From 0f02a5c6e63a95f910e6aba572729ca8085ac3ab Mon Sep 17 00:00:00 2001
From: Dongjoon Hyun <dongjoon@apache.org>
Date: Sun, 27 Mar 2016 20:07:31 +0100
Subject: [MINOR][MLLIB] Remove TODO comment DecisionTreeModel.scala

## What changes were proposed in this pull request?

This PR fixes the following line and the related code. Historically, this code was added in [SPARK-5597](https://issues.apache.org/jira/browse/SPARK-5597). After [SPARK-5597](https://issues.apache.org/jira/browse/SPARK-5597) was committed, [SPARK-3365](https://issues.apache.org/jira/browse/SPARK-3365) is fixed now. Now, we had better remove the comment without changing persistent code.

```scala
-        categories: Seq[Double]) { // TODO: Change to List once SPARK-3365 is fixed
+        categories: Seq[Double]) {
```

## How was this patch tested?

Pass the Jenkins tests.

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #11966 from dongjoon-hyun/change_categories_type.
---
 .../scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala     | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala
index ea68ff64a8..a87f8a6cde 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/DecisionTreeModel.scala
@@ -156,7 +156,7 @@ object DecisionTreeModel extends Loader[DecisionTreeModel] with Logging {
         feature: Int,
         threshold: Double,
         featureType: Int,
-        categories: Seq[Double]) { // TODO: Change to List once SPARK-3365 is fixed
+        categories: Seq[Double]) {
       def toSplit: Split = {
         new Split(feature, threshold, FeatureType(featureType), categories.toList)
       }
-- 
cgit v1.2.3


From 8ef493760f58687df766d03ccf64039635a2609f Mon Sep 17 00:00:00 2001
From: "Joseph K. Bradley" <joseph@databricks.com>
Date: Sun, 27 Mar 2016 19:04:18 -0700
Subject: [SPARK-10691][ML] Make LogisticRegressionModel, LinearRegressionModel
 evaluate() public

## What changes were proposed in this pull request?

Made evaluate method public.  Fixed LogisticRegressionModel evaluate to handle case when probabilityCol is not specified.

## How was this patch tested?

There were already unit tests for these methods.

Author: Joseph K. Bradley <joseph@databricks.com>

Closes #11928 from jkbradley/public-evaluate.
---
 .../apache/spark/ml/classification/LogisticRegression.scala  | 12 +++++++-----
 .../org/apache/spark/ml/regression/LinearRegression.scala    |  8 ++++----
 2 files changed, 11 insertions(+), 9 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
index 861b1d4b66..3d1d5b6892 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
@@ -539,13 +539,15 @@ class LogisticRegressionModel private[spark] (
   def hasSummary: Boolean = trainingSummary.isDefined
 
   /**
-   * Evaluates the model on a testset.
+   * Evaluates the model on a test dataset.
    * @param dataset Test dataset to evaluate model on.
    */
-  // TODO: decide on a good name before exposing to public API
-  private[classification] def evaluate(dataset: DataFrame): LogisticRegressionSummary = {
-    new BinaryLogisticRegressionSummary(
-      this.transform(dataset), $(probabilityCol), $(labelCol), $(featuresCol))
+  @Since("2.0.0")
+  def evaluate(dataset: DataFrame): LogisticRegressionSummary = {
+    // Handle possible missing or invalid prediction columns
+    val (summaryModel, probabilityColName) = findSummaryModelAndProbabilityCol()
+    new BinaryLogisticRegressionSummary(summaryModel.transform(dataset),
+      probabilityColName, $(labelCol), $(featuresCol))
   }
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
index b81c588e44..5ec02135cc 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
@@ -412,15 +412,15 @@ class LinearRegressionModel private[ml] (
   def hasSummary: Boolean = trainingSummary.isDefined
 
   /**
-   * Evaluates the model on a testset.
+   * Evaluates the model on a test dataset.
    * @param dataset Test dataset to evaluate model on.
    */
-  // TODO: decide on a good name before exposing to public API
-  private[regression] def evaluate(dataset: DataFrame): LinearRegressionSummary = {
+  @Since("2.0.0")
+  def evaluate(dataset: DataFrame): LinearRegressionSummary = {
     // Handle possible missing or invalid prediction columns
     val (summaryModel, predictionColName) = findSummaryModelAndPredictionCol()
     new LinearRegressionSummary(summaryModel.transform(dataset), predictionColName,
-      $(labelCol), this, Array(0D))
+      $(labelCol), summaryModel, Array(0D))
   }
 
   /**
-- 
cgit v1.2.3


From 7b841540180e8d1403d6c95b02e93f129267b34f Mon Sep 17 00:00:00 2001
From: Sean Owen <sowen@cloudera.com>
Date: Mon, 28 Mar 2016 12:01:33 +0100
Subject: [SPARK-12494][MLLIB] Array out of bound Exception in KMeans Yarn Mode

## What changes were proposed in this pull request?

Better error message with k-means init can't be enough samples from input (because it is perhaps empty)

## How was this patch tested?

Jenkins tests.

Author: Sean Owen <sowen@cloudera.com>

Closes #11979 from srowen/SPARK-12494.
---
 mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala | 2 ++
 1 file changed, 2 insertions(+)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
index a7beb81980..37a21cd879 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
@@ -390,6 +390,8 @@ class KMeans private (
     // Initialize each run's first center to a random point.
     val seed = new XORShiftRandom(this.seed).nextInt()
     val sample = data.takeSample(true, runs, seed).toSeq
+    // Could be empty if data is empty; fail with a better message early:
+    require(sample.size >= runs, s"Required $runs samples but got ${sample.size} from $data")
     val newCenters = Array.tabulate(runs)(r => ArrayBuffer(sample(r).toDense))
 
     /** Merges new centers to centers. */
-- 
cgit v1.2.3


From c8388297c436691a236520d2396deaf556aedb0e Mon Sep 17 00:00:00 2001
From: Chenliang Xu <chexu@groupon.com>
Date: Mon, 28 Mar 2016 08:33:37 -0700
Subject: [SPARK-14187][MLLIB] Fix incorrect use of binarySearch in
 SparseMatrix

## What changes were proposed in this pull request?

Fix incorrect use of binarySearch in SparseMatrix

## How was this patch tested?

Unit test added.

Author: Chenliang Xu <chexu@groupon.com>

Closes #11992 from luckyrandom/SPARK-14187.
---
 mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala     | 2 +-
 .../src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala  | 4 ++++
 2 files changed, 5 insertions(+), 1 deletion(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
index c6de7751f5..a09bc65cf3 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
@@ -613,7 +613,7 @@ class SparseMatrix @Since("1.3.0") (
 
   private[mllib] def update(i: Int, j: Int, v: Double): Unit = {
     val ind = index(i, j)
-    if (ind == -1) {
+    if (ind < 0) {
       throw new NoSuchElementException("The given row and column indices correspond to a zero " +
         "value. Only non-zero elements in Sparse Matrices can be updated.")
     } else {
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
index a02b8c9635..57907f415c 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
@@ -150,6 +150,10 @@ class MatricesSuite extends SparkFunSuite {
       sparseMat.update(0, 0, 10.0)
     }
 
+    intercept[NoSuchElementException] {
+      sparseMat.update(2, 1, 10.0)
+    }
+
     sparseMat.update(0, 1, 10.0)
     assert(sparseMat(0, 1) === 10.0)
     assert(sparseMat.values(2) === 10.0)
-- 
cgit v1.2.3


From 8c11d1aab8522c75d78bc6b30402c64e8d9ff065 Mon Sep 17 00:00:00 2001
From: Xusen Yin <yinxusen@gmail.com>
Date: Mon, 28 Mar 2016 15:40:06 -0700
Subject: [SPARK-11893] Model export/import for spark.ml: TrainValidationSplit

https://issues.apache.org/jira/browse/SPARK-11893

jkbradley In order to share read/write with `TrainValidationSplit`, I move the `SharedReadWrite` out of `CrossValidator` into a new trait `SharedReadWrite` in the tunning package.

To reduce the repeated tests, I move the complex tests from `CrossValidatorSuite` to `SharedReadWriteSuite`, and create a fake validator called `MyValidator` to test the shared code.

With `SharedReadWrite`, potential newly added `Validator` can share the read/write common part, and only need to implement their extra params save/load.

Author: Xusen Yin <yinxusen@gmail.com>
Author: Joseph K. Bradley <joseph@databricks.com>

Closes #9971 from yinxusen/SPARK-11893.
---
 .../apache/spark/ml/tuning/CrossValidator.scala    | 148 ++-------------------
 .../spark/ml/tuning/TrainValidationSplit.scala     | 100 +++++++++++++-
 .../apache/spark/ml/tuning/ValidatorParams.scala   | 117 +++++++++++++++-
 .../scala/org/apache/spark/ml/util/ReadWrite.scala |  42 +++++-
 .../ml/tuning/TrainValidationSplitSuite.scala      |  45 ++++++-
 5 files changed, 310 insertions(+), 142 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
index 963f81cb3e..040b0093b9 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
@@ -19,25 +19,19 @@ package org.apache.spark.ml.tuning
 
 import com.github.fommil.netlib.F2jBLAS
 import org.apache.hadoop.fs.Path
-import org.json4s.{DefaultFormats, JObject}
-import org.json4s.jackson.JsonMethods._
+import org.json4s.DefaultFormats
 
-import org.apache.spark.SparkContext
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml._
-import org.apache.spark.ml.classification.OneVsRestParams
 import org.apache.spark.ml.evaluation.Evaluator
-import org.apache.spark.ml.feature.RFormulaModel
 import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared.HasSeed
 import org.apache.spark.ml.util._
-import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.mllib.util.MLUtils
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.types.StructType
 
-
 /**
  * Params for [[CrossValidator]] and [[CrossValidatorModel]].
  */
@@ -45,6 +39,7 @@ private[ml] trait CrossValidatorParams extends ValidatorParams with HasSeed {
   /**
    * Param for number of folds for cross validation.  Must be >= 2.
    * Default: 3
+   *
    * @group param
    */
   val numFolds: IntParam = new IntParam(this, "numFolds",
@@ -163,10 +158,10 @@ object CrossValidator extends MLReadable[CrossValidator] {
 
   private[CrossValidator] class CrossValidatorWriter(instance: CrossValidator) extends MLWriter {
 
-    SharedReadWrite.validateParams(instance)
+    ValidatorParams.validateParams(instance)
 
     override protected def saveImpl(path: String): Unit =
-      SharedReadWrite.saveImpl(path, instance, sc)
+      ValidatorParams.saveImpl(path, instance, sc)
   }
 
   private class CrossValidatorReader extends MLReader[CrossValidator] {
@@ -175,8 +170,11 @@ object CrossValidator extends MLReadable[CrossValidator] {
     private val className = classOf[CrossValidator].getName
 
     override def load(path: String): CrossValidator = {
-      val (metadata, estimator, evaluator, estimatorParamMaps, numFolds) =
-        SharedReadWrite.load(path, sc, className)
+      implicit val format = DefaultFormats
+
+      val (metadata, estimator, evaluator, estimatorParamMaps) =
+        ValidatorParams.loadImpl(path, sc, className)
+      val numFolds = (metadata.params \ "numFolds").extract[Int]
       new CrossValidator(metadata.uid)
         .setEstimator(estimator)
         .setEvaluator(evaluator)
@@ -184,123 +182,6 @@ object CrossValidator extends MLReadable[CrossValidator] {
         .setNumFolds(numFolds)
     }
   }
-
-  private object CrossValidatorReader {
-    /**
-     * Examine the given estimator (which may be a compound estimator) and extract a mapping
-     * from UIDs to corresponding [[Params]] instances.
-     */
-    def getUidMap(instance: Params): Map[String, Params] = {
-      val uidList = getUidMapImpl(instance)
-      val uidMap = uidList.toMap
-      if (uidList.size != uidMap.size) {
-        throw new RuntimeException("CrossValidator.load found a compound estimator with stages" +
-          s" with duplicate UIDs.  List of UIDs: ${uidList.map(_._1).mkString(", ")}")
-      }
-      uidMap
-    }
-
-    def getUidMapImpl(instance: Params): List[(String, Params)] = {
-      val subStages: Array[Params] = instance match {
-        case p: Pipeline => p.getStages.asInstanceOf[Array[Params]]
-        case pm: PipelineModel => pm.stages.asInstanceOf[Array[Params]]
-        case v: ValidatorParams => Array(v.getEstimator, v.getEvaluator)
-        case ovr: OneVsRestParams =>
-          // TODO: SPARK-11892: This case may require special handling.
-          throw new UnsupportedOperationException("CrossValidator write will fail because it" +
-            " cannot yet handle an estimator containing type: ${ovr.getClass.getName}")
-        case rformModel: RFormulaModel => Array(rformModel.pipelineModel)
-        case _: Params => Array()
-      }
-      val subStageMaps = subStages.map(getUidMapImpl).foldLeft(List.empty[(String, Params)])(_ ++ _)
-      List((instance.uid, instance)) ++ subStageMaps
-    }
-  }
-
-  private[tuning] object SharedReadWrite {
-
-    /**
-     * Check that [[CrossValidator.evaluator]] and [[CrossValidator.estimator]] are Writable.
-     * This does not check [[CrossValidator.estimatorParamMaps]].
-     */
-    def validateParams(instance: ValidatorParams): Unit = {
-      def checkElement(elem: Params, name: String): Unit = elem match {
-        case stage: MLWritable => // good
-        case other =>
-          throw new UnsupportedOperationException("CrossValidator write will fail " +
-            s" because it contains $name which does not implement Writable." +
-            s" Non-Writable $name: ${other.uid} of type ${other.getClass}")
-      }
-      checkElement(instance.getEvaluator, "evaluator")
-      checkElement(instance.getEstimator, "estimator")
-      // Check to make sure all Params apply to this estimator.  Throw an error if any do not.
-      // Extraneous Params would cause problems when loading the estimatorParamMaps.
-      val uidToInstance: Map[String, Params] = CrossValidatorReader.getUidMap(instance)
-      instance.getEstimatorParamMaps.foreach { case pMap: ParamMap =>
-        pMap.toSeq.foreach { case ParamPair(p, v) =>
-          require(uidToInstance.contains(p.parent), s"CrossValidator save requires all Params in" +
-            s" estimatorParamMaps to apply to this CrossValidator, its Estimator, or its" +
-            s" Evaluator.  An extraneous Param was found: $p")
-        }
-      }
-    }
-
-    private[tuning] def saveImpl(
-        path: String,
-        instance: CrossValidatorParams,
-        sc: SparkContext,
-        extraMetadata: Option[JObject] = None): Unit = {
-      import org.json4s.JsonDSL._
-
-      val estimatorParamMapsJson = compact(render(
-        instance.getEstimatorParamMaps.map { case paramMap =>
-          paramMap.toSeq.map { case ParamPair(p, v) =>
-            Map("parent" -> p.parent, "name" -> p.name, "value" -> p.jsonEncode(v))
-          }
-        }.toSeq
-      ))
-      val jsonParams = List(
-        "numFolds" -> parse(instance.numFolds.jsonEncode(instance.getNumFolds)),
-        "estimatorParamMaps" -> parse(estimatorParamMapsJson)
-      )
-      DefaultParamsWriter.saveMetadata(instance, path, sc, extraMetadata, Some(jsonParams))
-
-      val evaluatorPath = new Path(path, "evaluator").toString
-      instance.getEvaluator.asInstanceOf[MLWritable].save(evaluatorPath)
-      val estimatorPath = new Path(path, "estimator").toString
-      instance.getEstimator.asInstanceOf[MLWritable].save(estimatorPath)
-    }
-
-    private[tuning] def load[M <: Model[M]](
-        path: String,
-        sc: SparkContext,
-        expectedClassName: String): (Metadata, Estimator[M], Evaluator, Array[ParamMap], Int) = {
-
-      val metadata = DefaultParamsReader.loadMetadata(path, sc, expectedClassName)
-
-      implicit val format = DefaultFormats
-      val evaluatorPath = new Path(path, "evaluator").toString
-      val evaluator = DefaultParamsReader.loadParamsInstance[Evaluator](evaluatorPath, sc)
-      val estimatorPath = new Path(path, "estimator").toString
-      val estimator = DefaultParamsReader.loadParamsInstance[Estimator[M]](estimatorPath, sc)
-
-      val uidToParams = Map(evaluator.uid -> evaluator) ++ CrossValidatorReader.getUidMap(estimator)
-
-      val numFolds = (metadata.params \ "numFolds").extract[Int]
-      val estimatorParamMaps: Array[ParamMap] =
-        (metadata.params \ "estimatorParamMaps").extract[Seq[Seq[Map[String, String]]]].map {
-          pMap =>
-            val paramPairs = pMap.map { case pInfo: Map[String, String] =>
-              val est = uidToParams(pInfo("parent"))
-              val param = est.getParam(pInfo("name"))
-              val value = param.jsonDecode(pInfo("value"))
-              param -> value
-            }
-            ParamMap(paramPairs: _*)
-        }.toArray
-      (metadata, estimator, evaluator, estimatorParamMaps, numFolds)
-    }
-  }
 }
 
 /**
@@ -346,8 +227,6 @@ class CrossValidatorModel private[ml] (
 @Since("1.6.0")
 object CrossValidatorModel extends MLReadable[CrossValidatorModel] {
 
-  import CrossValidator.SharedReadWrite
-
   @Since("1.6.0")
   override def read: MLReader[CrossValidatorModel] = new CrossValidatorModelReader
 
@@ -357,12 +236,12 @@ object CrossValidatorModel extends MLReadable[CrossValidatorModel] {
   private[CrossValidatorModel]
   class CrossValidatorModelWriter(instance: CrossValidatorModel) extends MLWriter {
 
-    SharedReadWrite.validateParams(instance)
+    ValidatorParams.validateParams(instance)
 
     override protected def saveImpl(path: String): Unit = {
       import org.json4s.JsonDSL._
       val extraMetadata = "avgMetrics" -> instance.avgMetrics.toSeq
-      SharedReadWrite.saveImpl(path, instance, sc, Some(extraMetadata))
+      ValidatorParams.saveImpl(path, instance, sc, Some(extraMetadata))
       val bestModelPath = new Path(path, "bestModel").toString
       instance.bestModel.asInstanceOf[MLWritable].save(bestModelPath)
     }
@@ -376,8 +255,9 @@ object CrossValidatorModel extends MLReadable[CrossValidatorModel] {
     override def load(path: String): CrossValidatorModel = {
       implicit val format = DefaultFormats
 
-      val (metadata, estimator, evaluator, estimatorParamMaps, numFolds) =
-        SharedReadWrite.load(path, sc, className)
+      val (metadata, estimator, evaluator, estimatorParamMaps) =
+        ValidatorParams.loadImpl(path, sc, className)
+      val numFolds = (metadata.params \ "numFolds").extract[Int]
       val bestModelPath = new Path(path, "bestModel").toString
       val bestModel = DefaultParamsReader.loadParamsInstance[Model[_]](bestModelPath, sc)
       val avgMetrics = (metadata.metadata \ "avgMetrics").extract[Seq[Double]].toArray
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
index 70fa5f0234..4d1d6364d7 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
@@ -17,12 +17,15 @@
 
 package org.apache.spark.ml.tuning
 
+import org.apache.hadoop.fs.Path
+import org.json4s.DefaultFormats
+
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.evaluation.Evaluator
 import org.apache.spark.ml.param.{DoubleParam, ParamMap, ParamValidators}
-import org.apache.spark.ml.util.Identifiable
+import org.apache.spark.ml.util._
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.types.StructType
 
@@ -33,6 +36,7 @@ private[ml] trait TrainValidationSplitParams extends ValidatorParams {
   /**
    * Param for ratio between train and validation data. Must be between 0 and 1.
    * Default: 0.75
+   *
    * @group param
    */
   val trainRatio: DoubleParam = new DoubleParam(this, "trainRatio",
@@ -55,7 +59,7 @@ private[ml] trait TrainValidationSplitParams extends ValidatorParams {
 @Experimental
 class TrainValidationSplit @Since("1.5.0") (@Since("1.5.0") override val uid: String)
   extends Estimator[TrainValidationSplitModel]
-  with TrainValidationSplitParams with Logging {
+  with TrainValidationSplitParams with MLWritable with Logging {
 
   @Since("1.5.0")
   def this() = this(Identifiable.randomUID("tvs"))
@@ -130,6 +134,47 @@ class TrainValidationSplit @Since("1.5.0") (@Since("1.5.0") override val uid: St
     }
     copied
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new TrainValidationSplit.TrainValidationSplitWriter(this)
+}
+
+@Since("2.0.0")
+object TrainValidationSplit extends MLReadable[TrainValidationSplit] {
+
+  @Since("2.0.0")
+  override def read: MLReader[TrainValidationSplit] = new TrainValidationSplitReader
+
+  @Since("2.0.0")
+  override def load(path: String): TrainValidationSplit = super.load(path)
+
+  private[TrainValidationSplit] class TrainValidationSplitWriter(instance: TrainValidationSplit)
+    extends MLWriter {
+
+    ValidatorParams.validateParams(instance)
+
+    override protected def saveImpl(path: String): Unit =
+      ValidatorParams.saveImpl(path, instance, sc)
+  }
+
+  private class TrainValidationSplitReader extends MLReader[TrainValidationSplit] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[TrainValidationSplit].getName
+
+    override def load(path: String): TrainValidationSplit = {
+      implicit val format = DefaultFormats
+
+      val (metadata, estimator, evaluator, estimatorParamMaps) =
+        ValidatorParams.loadImpl(path, sc, className)
+      val trainRatio = (metadata.params \ "trainRatio").extract[Double]
+      new TrainValidationSplit(metadata.uid)
+        .setEstimator(estimator)
+        .setEvaluator(evaluator)
+        .setEstimatorParamMaps(estimatorParamMaps)
+        .setTrainRatio(trainRatio)
+    }
+  }
 }
 
 /**
@@ -146,7 +191,7 @@ class TrainValidationSplitModel private[ml] (
     @Since("1.5.0") override val uid: String,
     @Since("1.5.0") val bestModel: Model[_],
     @Since("1.5.0") val validationMetrics: Array[Double])
-  extends Model[TrainValidationSplitModel] with TrainValidationSplitParams {
+  extends Model[TrainValidationSplitModel] with TrainValidationSplitParams with MLWritable {
 
   @Since("1.5.0")
   override def transform(dataset: DataFrame): DataFrame = {
@@ -167,4 +212,53 @@ class TrainValidationSplitModel private[ml] (
       validationMetrics.clone())
     copyValues(copied, extra)
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new TrainValidationSplitModel.TrainValidationSplitModelWriter(this)
+}
+
+@Since("2.0.0")
+object TrainValidationSplitModel extends MLReadable[TrainValidationSplitModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[TrainValidationSplitModel] = new TrainValidationSplitModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): TrainValidationSplitModel = super.load(path)
+
+  private[TrainValidationSplitModel]
+  class TrainValidationSplitModelWriter(instance: TrainValidationSplitModel) extends MLWriter {
+
+    ValidatorParams.validateParams(instance)
+
+    override protected def saveImpl(path: String): Unit = {
+      import org.json4s.JsonDSL._
+      val extraMetadata = "validationMetrics" -> instance.validationMetrics.toSeq
+      ValidatorParams.saveImpl(path, instance, sc, Some(extraMetadata))
+      val bestModelPath = new Path(path, "bestModel").toString
+      instance.bestModel.asInstanceOf[MLWritable].save(bestModelPath)
+    }
+  }
+
+  private class TrainValidationSplitModelReader extends MLReader[TrainValidationSplitModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[TrainValidationSplitModel].getName
+
+    override def load(path: String): TrainValidationSplitModel = {
+      implicit val format = DefaultFormats
+
+      val (metadata, estimator, evaluator, estimatorParamMaps) =
+        ValidatorParams.loadImpl(path, sc, className)
+      val trainRatio = (metadata.params \ "trainRatio").extract[Double]
+      val bestModelPath = new Path(path, "bestModel").toString
+      val bestModel = DefaultParamsReader.loadParamsInstance[Model[_]](bestModelPath, sc)
+      val validationMetrics = (metadata.metadata \ "validationMetrics").extract[Seq[Double]].toArray
+      val tvs = new TrainValidationSplitModel(metadata.uid, bestModel, validationMetrics)
+      tvs.set(tvs.estimator, estimator)
+        .set(tvs.evaluator, evaluator)
+        .set(tvs.estimatorParamMaps, estimatorParamMaps)
+        .set(tvs.trainRatio, trainRatio)
+    }
+  }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/ValidatorParams.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/ValidatorParams.scala
index 953456e8f0..7a4e106aeb 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/ValidatorParams.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/ValidatorParams.scala
@@ -17,9 +17,17 @@
 
 package org.apache.spark.ml.tuning
 
-import org.apache.spark.ml.Estimator
+import org.apache.hadoop.fs.Path
+import org.json4s.{DefaultFormats, _}
+import org.json4s.jackson.JsonMethods._
+
+import org.apache.spark.SparkContext
+import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.evaluation.Evaluator
-import org.apache.spark.ml.param.{Param, ParamMap, Params}
+import org.apache.spark.ml.param.{Param, ParamMap, ParamPair, Params}
+import org.apache.spark.ml.util.{DefaultParamsReader, DefaultParamsWriter, MetaAlgorithmReadWrite,
+  MLWritable}
+import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.sql.types.StructType
 
 /**
@@ -69,3 +77,108 @@ private[ml] trait ValidatorParams extends Params {
     est.copy(firstEstimatorParamMap).transformSchema(schema)
   }
 }
+
+private[ml] object ValidatorParams {
+  /**
+   * Check that [[ValidatorParams.evaluator]] and [[ValidatorParams.estimator]] are Writable.
+   * This does not check [[ValidatorParams.estimatorParamMaps]].
+   */
+  def validateParams(instance: ValidatorParams): Unit = {
+    def checkElement(elem: Params, name: String): Unit = elem match {
+      case stage: MLWritable => // good
+      case other =>
+        throw new UnsupportedOperationException(instance.getClass.getName + " write will fail " +
+          s" because it contains $name which does not implement Writable." +
+          s" Non-Writable $name: ${other.uid} of type ${other.getClass}")
+    }
+    checkElement(instance.getEvaluator, "evaluator")
+    checkElement(instance.getEstimator, "estimator")
+    // Check to make sure all Params apply to this estimator.  Throw an error if any do not.
+    // Extraneous Params would cause problems when loading the estimatorParamMaps.
+    val uidToInstance: Map[String, Params] = MetaAlgorithmReadWrite.getUidMap(instance)
+    instance.getEstimatorParamMaps.foreach { case pMap: ParamMap =>
+      pMap.toSeq.foreach { case ParamPair(p, v) =>
+        require(uidToInstance.contains(p.parent), s"ValidatorParams save requires all Params in" +
+          s" estimatorParamMaps to apply to this ValidatorParams, its Estimator, or its" +
+          s" Evaluator. An extraneous Param was found: $p")
+      }
+    }
+  }
+
+  /**
+   * Generic implementation of save for [[ValidatorParams]] types.
+   * This handles all [[ValidatorParams]] fields and saves [[Param]] values, but the implementing
+   * class needs to handle model data.
+   */
+  def saveImpl(
+      path: String,
+      instance: ValidatorParams,
+      sc: SparkContext,
+      extraMetadata: Option[JObject] = None): Unit = {
+    import org.json4s.JsonDSL._
+
+    val estimatorParamMapsJson = compact(render(
+      instance.getEstimatorParamMaps.map { case paramMap =>
+        paramMap.toSeq.map { case ParamPair(p, v) =>
+          Map("parent" -> p.parent, "name" -> p.name, "value" -> p.jsonEncode(v))
+        }
+      }.toSeq
+    ))
+
+    val validatorSpecificParams = instance match {
+      case cv: CrossValidatorParams =>
+        List("numFolds" -> parse(cv.numFolds.jsonEncode(cv.getNumFolds)))
+      case tvs: TrainValidationSplitParams =>
+        List("trainRatio" -> parse(tvs.trainRatio.jsonEncode(tvs.getTrainRatio)))
+      case _ =>
+        // This should not happen.
+        throw new NotImplementedError("ValidatorParams.saveImpl does not handle type: " +
+          instance.getClass.getCanonicalName)
+    }
+
+    val jsonParams = validatorSpecificParams ++ List(
+      "estimatorParamMaps" -> parse(estimatorParamMapsJson))
+
+    DefaultParamsWriter.saveMetadata(instance, path, sc, extraMetadata, Some(jsonParams))
+
+    val evaluatorPath = new Path(path, "evaluator").toString
+    instance.getEvaluator.asInstanceOf[MLWritable].save(evaluatorPath)
+    val estimatorPath = new Path(path, "estimator").toString
+    instance.getEstimator.asInstanceOf[MLWritable].save(estimatorPath)
+  }
+
+  /**
+   * Generic implementation of load for [[ValidatorParams]] types.
+   * This handles all [[ValidatorParams]] fields, but the implementing
+   * class needs to handle model data and special [[Param]] values.
+   */
+  def loadImpl[M <: Model[M]](
+      path: String,
+      sc: SparkContext,
+      expectedClassName: String): (Metadata, Estimator[M], Evaluator, Array[ParamMap]) = {
+
+    val metadata = DefaultParamsReader.loadMetadata(path, sc, expectedClassName)
+
+    implicit val format = DefaultFormats
+    val evaluatorPath = new Path(path, "evaluator").toString
+    val evaluator = DefaultParamsReader.loadParamsInstance[Evaluator](evaluatorPath, sc)
+    val estimatorPath = new Path(path, "estimator").toString
+    val estimator = DefaultParamsReader.loadParamsInstance[Estimator[M]](estimatorPath, sc)
+
+    val uidToParams = Map(evaluator.uid -> evaluator) ++ MetaAlgorithmReadWrite.getUidMap(estimator)
+
+    val estimatorParamMaps: Array[ParamMap] =
+      (metadata.params \ "estimatorParamMaps").extract[Seq[Seq[Map[String, String]]]].map {
+        pMap =>
+          val paramPairs = pMap.map { case pInfo: Map[String, String] =>
+            val est = uidToParams(pInfo("parent"))
+            val param = est.getParam(pInfo("name"))
+            val value = param.jsonDecode(pInfo("value"))
+            param -> value
+          }
+          ParamMap(paramPairs: _*)
+      }.toArray
+
+    (metadata, estimator, evaluator, estimatorParamMaps)
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala b/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
index c95e536abd..5a596cad06 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
@@ -21,13 +21,18 @@ import java.io.IOException
 
 import org.apache.hadoop.fs.Path
 import org.json4s._
-import org.json4s.jackson.JsonMethods._
+import org.json4s.{DefaultFormats, JObject}
 import org.json4s.JsonDSL._
+import org.json4s.jackson.JsonMethods._
 
 import org.apache.spark.SparkContext
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
+import org.apache.spark.ml._
+import org.apache.spark.ml.classification.OneVsRestParams
+import org.apache.spark.ml.feature.RFormulaModel
 import org.apache.spark.ml.param.{ParamPair, Params}
+import org.apache.spark.ml.tuning.ValidatorParams
 import org.apache.spark.sql.SQLContext
 import org.apache.spark.util.Utils
 
@@ -352,3 +357,38 @@ private[ml] object DefaultParamsReader {
     cls.getMethod("read").invoke(null).asInstanceOf[MLReader[T]].load(path)
   }
 }
+
+/**
+ * Default Meta-Algorithm read and write implementation.
+ */
+private[ml] object MetaAlgorithmReadWrite {
+  /**
+   * Examine the given estimator (which may be a compound estimator) and extract a mapping
+   * from UIDs to corresponding [[Params]] instances.
+   */
+  def getUidMap(instance: Params): Map[String, Params] = {
+    val uidList = getUidMapImpl(instance)
+    val uidMap = uidList.toMap
+    if (uidList.size != uidMap.size) {
+      throw new RuntimeException(s"${instance.getClass.getName}.load found a compound estimator" +
+        s" with stages with duplicate UIDs. List of UIDs: ${uidList.map(_._1).mkString(", ")}.")
+    }
+    uidMap
+  }
+
+  private def getUidMapImpl(instance: Params): List[(String, Params)] = {
+    val subStages: Array[Params] = instance match {
+      case p: Pipeline => p.getStages.asInstanceOf[Array[Params]]
+      case pm: PipelineModel => pm.stages.asInstanceOf[Array[Params]]
+      case v: ValidatorParams => Array(v.getEstimator, v.getEvaluator)
+      case ovr: OneVsRestParams =>
+        // TODO: SPARK-11892: This case may require special handling.
+        throw new UnsupportedOperationException(s"${instance.getClass.getName} write will fail" +
+          s" because it cannot yet handle an estimator containing type: ${ovr.getClass.getName}.")
+      case rformModel: RFormulaModel => Array(rformModel.pipelineModel)
+      case _: Params => Array()
+    }
+    val subStageMaps = subStages.map(getUidMapImpl).foldLeft(List.empty[(String, Params)])(_ ++ _)
+    List((instance.uid, instance)) ++ subStageMaps
+  }
+}
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
index cf8dcefebc..7cf7b3e087 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
@@ -19,17 +19,20 @@ package org.apache.spark.ml.tuning
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.{Estimator, Model}
-import org.apache.spark.ml.classification.LogisticRegression
+import org.apache.spark.ml.classification.{LogisticRegression, LogisticRegressionModel}
 import org.apache.spark.ml.evaluation.{BinaryClassificationEvaluator, Evaluator, RegressionEvaluator}
 import org.apache.spark.ml.param.ParamMap
 import org.apache.spark.ml.param.shared.HasInputCol
 import org.apache.spark.ml.regression.LinearRegression
+import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInput
+import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.util.{LinearDataGenerator, MLlibTestSparkContext}
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.types.StructType
 
-class TrainValidationSplitSuite extends SparkFunSuite with MLlibTestSparkContext {
+class TrainValidationSplitSuite
+  extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
   test("train validation with logistic regression") {
     val dataset = sqlContext.createDataFrame(
       sc.parallelize(generateLogisticInput(1.0, 1.0, 100, 42), 2))
@@ -105,6 +108,44 @@ class TrainValidationSplitSuite extends SparkFunSuite with MLlibTestSparkContext
       cv.transformSchema(new StructType())
     }
   }
+
+  test("read/write: TrainValidationSplit") {
+    val lr = new LogisticRegression().setMaxIter(3)
+    val evaluator = new BinaryClassificationEvaluator()
+    val paramMaps = new ParamGridBuilder()
+        .addGrid(lr.regParam, Array(0.1, 0.2))
+        .build()
+    val tvs = new TrainValidationSplit()
+      .setEstimator(lr)
+      .setEvaluator(evaluator)
+      .setTrainRatio(0.5)
+      .setEstimatorParamMaps(paramMaps)
+
+    val tvs2 = testDefaultReadWrite(tvs, testParams = false)
+
+    assert(tvs.getTrainRatio === tvs2.getTrainRatio)
+  }
+
+  test("read/write: TrainValidationSplitModel") {
+    val lr = new LogisticRegression()
+      .setThreshold(0.6)
+    val lrModel = new LogisticRegressionModel(lr.uid, Vectors.dense(1.0, 2.0), 1.2)
+      .setThreshold(0.6)
+    val evaluator = new BinaryClassificationEvaluator()
+    val paramMaps = new ParamGridBuilder()
+        .addGrid(lr.regParam, Array(0.1, 0.2))
+        .build()
+    val tvs = new TrainValidationSplitModel("cvUid", lrModel, Array(0.3, 0.6))
+    tvs.set(tvs.estimator, lr)
+      .set(tvs.evaluator, evaluator)
+      .set(tvs.trainRatio, 0.5)
+      .set(tvs.estimatorParamMaps, paramMaps)
+
+    val tvs2 = testDefaultReadWrite(tvs, testParams = false)
+
+    assert(tvs.getTrainRatio === tvs2.getTrainRatio)
+    assert(tvs.validationMetrics === tvs2.validationMetrics)
+  }
 }
 
 object TrainValidationSplitSuite {
-- 
cgit v1.2.3


From f6066b0c3c35ceea1706378145e15776c9b4415a Mon Sep 17 00:00:00 2001
From: sethah <seth.hendrickson16@gmail.com>
Date: Mon, 28 Mar 2016 22:27:53 -0700
Subject: [SPARK-11730][ML] Add feature importances for GBTs.

## What changes were proposed in this pull request?

Now that GBTs have been moved to ML, they can use the implementation of feature importance for random forests. This patch simply adds a `featureImportances` attribute to `GBTClassifier` and `GBTRegressor` and adds tests for each.

GBT feature importances here simply average the feature importances for each tree in its ensemble. This follows the implementation from scikit-learn. This method is also suggested by J Friedman in [this paper](https://statweb.stanford.edu/~jhf/ftp/trebst.pdf).

## How was this patch tested?

Unit tests were added to `GBTClassifierSuite` and `GBTRegressorSuite` to validate feature importances.

Author: sethah <seth.hendrickson16@gmail.com>

Closes #11961 from sethah/SPARK-11730.
---
 .../ml/classification/DecisionTreeClassifier.scala |   2 +-
 .../spark/ml/classification/GBTClassifier.scala    |  13 +++
 .../ml/classification/RandomForestClassifier.scala |  16 ++-
 .../ml/regression/DecisionTreeRegressor.scala      |   2 +-
 .../apache/spark/ml/regression/GBTRegressor.scala  |  13 +++
 .../ml/regression/RandomForestRegressor.scala      |  16 ++-
 .../spark/ml/tree/impl/GradientBoostedTrees.scala  |   2 +
 .../apache/spark/ml/tree/impl/RandomForest.scala   | 110 -------------------
 .../org/apache/spark/ml/tree/treeModels.scala      | 120 +++++++++++++++++++++
 .../ml/classification/GBTClassifierSuite.scala     |  25 +++++
 .../spark/ml/regression/GBTRegressorSuite.scala    |  23 ++++
 .../spark/ml/tree/impl/RandomForestSuite.scala     |   6 +-
 12 files changed, 213 insertions(+), 135 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
index 3e4b21bff6..23c4af17f9 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
@@ -203,7 +203,7 @@ final class DecisionTreeClassificationModel private[ml] (
    *       to determine feature importance instead.
    */
   @Since("2.0.0")
-  lazy val featureImportances: Vector = RandomForest.featureImportances(this, numFeatures)
+  lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(this, numFeatures)
 
   /** Convert to spark.mllib DecisionTreeModel (losing some infomation) */
   override private[spark] def toOld: OldDecisionTreeModel = {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
index c31df3aa18..48ce051d0a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -238,6 +238,19 @@ final class GBTClassificationModel private[ml](
     s"GBTClassificationModel (uid=$uid) with $numTrees trees"
   }
 
+  /**
+   * Estimate of the importance of each feature.
+   *
+   * Each feature's importance is the average of its importance across all trees in the ensemble
+   * The importance vector is normalized to sum to 1. This method is suggested by Hastie et al.
+   * (Hastie, Tibshirani, Friedman. "The Elements of Statistical Learning, 2nd Edition." 2001.)
+   * and follows the implementation from scikit-learn.
+   *
+   * @see [[DecisionTreeClassificationModel.featureImportances]]
+   */
+  @Since("2.0.0")
+  lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(trees, numFeatures)
+
   /** (private[ml]) Convert to a model in the old API */
   private[ml] def toOld: OldGBTModel = {
     new OldGBTModel(OldAlgo.Classification, _trees.map(_.toOld), _treeWeights)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
index 5da04d341d..82fa05a604 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
@@ -222,19 +222,15 @@ final class RandomForestClassificationModel private[ml] (
   /**
    * Estimate of the importance of each feature.
    *
-   * This generalizes the idea of "Gini" importance to other losses,
-   * following the explanation of Gini importance from "Random Forests" documentation
-   * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
+   * Each feature's importance is the average of its importance across all trees in the ensemble
+   * The importance vector is normalized to sum to 1. This method is suggested by Hastie et al.
+   * (Hastie, Tibshirani, Friedman. "The Elements of Statistical Learning, 2nd Edition." 2001.)
+   * and follows the implementation from scikit-learn.
    *
-   * This feature importance is calculated as follows:
-   *  - Average over trees:
-   *     - importance(feature j) = sum (over nodes which split on feature j) of the gain,
-   *       where gain is scaled by the number of instances passing through node
-   *     - Normalize importances for tree to sum to 1.
-   *  - Normalize feature importance vector to sum to 1.
+   * @see [[DecisionTreeClassificationModel.featureImportances]]
    */
   @Since("1.5.0")
-  lazy val featureImportances: Vector = RandomForest.featureImportances(trees, numFeatures)
+  lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(trees, numFeatures)
 
   /** (private[ml]) Convert to a model in the old API */
   private[ml] def toOld: OldRandomForestModel = {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
index 50ac96eb5e..0a3d00e470 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
@@ -203,7 +203,7 @@ final class DecisionTreeRegressionModel private[ml] (
    *       to determine feature importance instead.
    */
   @Since("2.0.0")
-  lazy val featureImportances: Vector = RandomForest.featureImportances(this, numFeatures)
+  lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(this, numFeatures)
 
   /** Convert to spark.mllib DecisionTreeModel (losing some infomation) */
   override private[spark] def toOld: OldDecisionTreeModel = {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
index da5b77e8fa..8fca35da51 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -224,6 +224,19 @@ final class GBTRegressionModel private[ml](
     s"GBTRegressionModel (uid=$uid) with $numTrees trees"
   }
 
+  /**
+   * Estimate of the importance of each feature.
+   *
+   * Each feature's importance is the average of its importance across all trees in the ensemble
+   * The importance vector is normalized to sum to 1. This method is suggested by Hastie et al.
+   * (Hastie, Tibshirani, Friedman. "The Elements of Statistical Learning, 2nd Edition." 2001.)
+   * and follows the implementation from scikit-learn.
+   *
+   * @see [[DecisionTreeRegressionModel.featureImportances]]
+   */
+  @Since("2.0.0")
+  lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(trees, numFeatures)
+
   /** (private[ml]) Convert to a model in the old API */
   private[ml] def toOld: OldGBTModel = {
     new OldGBTModel(OldAlgo.Regression, _trees.map(_.toOld), _treeWeights)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
index 798947b94a..5b3f3a1f5d 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
@@ -181,19 +181,15 @@ final class RandomForestRegressionModel private[ml] (
   /**
    * Estimate of the importance of each feature.
    *
-   * This generalizes the idea of "Gini" importance to other losses,
-   * following the explanation of Gini importance from "Random Forests" documentation
-   * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
+   * Each feature's importance is the average of its importance across all trees in the ensemble
+   * The importance vector is normalized to sum to 1. This method is suggested by Hastie et al.
+   * (Hastie, Tibshirani, Friedman. "The Elements of Statistical Learning, 2nd Edition." 2001.)
+   * and follows the implementation from scikit-learn.
    *
-   * This feature importance is calculated as follows:
-   *  - Average over trees:
-   *     - importance(feature j) = sum (over nodes which split on feature j) of the gain,
-   *       where gain is scaled by the number of instances passing through node
-   *     - Normalize importances for tree to sum to 1.
-   *  - Normalize feature importance vector to sum to 1.
+   * @see [[DecisionTreeRegressionModel.featureImportances]]
    */
   @Since("1.5.0")
-  lazy val featureImportances: Vector = RandomForest.featureImportances(trees, numFeatures)
+  lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(trees, numFeatures)
 
   /** (private[ml]) Convert to a model in the old API */
   private[ml] def toOld: OldRandomForestModel = {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
index 1c8a9b4dfe..b37f4e891e 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
@@ -19,7 +19,9 @@ package org.apache.spark.ml.tree.impl
 
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.regression.{DecisionTreeRegressionModel, DecisionTreeRegressor}
+import org.apache.spark.ml.tree.DecisionTreeModel
 import org.apache.spark.mllib.impl.PeriodicRDDCheckpointer
+import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
 import org.apache.spark.mllib.tree.configuration.{BoostingStrategy => OldBoostingStrategy}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala
index 7774ae64e5..cccf052b3e 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala
@@ -26,7 +26,6 @@ import org.apache.spark.internal.Logging
 import org.apache.spark.ml.classification.DecisionTreeClassificationModel
 import org.apache.spark.ml.regression.DecisionTreeRegressionModel
 import org.apache.spark.ml.tree._
-import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, Strategy => OldStrategy}
 import org.apache.spark.mllib.tree.impl.{BaggedPoint, DecisionTreeMetadata, DTStatsAggregator,
@@ -35,7 +34,6 @@ import org.apache.spark.mllib.tree.impurity.ImpurityCalculator
 import org.apache.spark.mllib.tree.model.ImpurityStats
 import org.apache.spark.rdd.RDD
 import org.apache.spark.storage.StorageLevel
-import org.apache.spark.util.collection.OpenHashMap
 import org.apache.spark.util.random.{SamplingUtils, XORShiftRandom}
 
 
@@ -1105,112 +1103,4 @@ private[spark] object RandomForest extends Logging {
     }
   }
 
-  /**
-   * Given a Random Forest model, compute the importance of each feature.
-   * This generalizes the idea of "Gini" importance to other losses,
-   * following the explanation of Gini importance from "Random Forests" documentation
-   * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
-   *
-   * This feature importance is calculated as follows:
-   *  - Average over trees:
-   *     - importance(feature j) = sum (over nodes which split on feature j) of the gain,
-   *       where gain is scaled by the number of instances passing through node
-   *     - Normalize importances for tree to sum to 1.
-   *  - Normalize feature importance vector to sum to 1.
-   *
-   * @param trees  Unweighted forest of trees
-   * @param numFeatures  Number of features in model (even if not all are explicitly used by
-   *                     the model).
-   *                     If -1, then numFeatures is set based on the max feature index in all trees.
-   * @return  Feature importance values, of length numFeatures.
-   */
-  private[ml] def featureImportances(trees: Array[DecisionTreeModel], numFeatures: Int): Vector = {
-    val totalImportances = new OpenHashMap[Int, Double]()
-    trees.foreach { tree =>
-      // Aggregate feature importance vector for this tree
-      val importances = new OpenHashMap[Int, Double]()
-      computeFeatureImportance(tree.rootNode, importances)
-      // Normalize importance vector for this tree, and add it to total.
-      // TODO: In the future, also support normalizing by tree.rootNode.impurityStats.count?
-      val treeNorm = importances.map(_._2).sum
-      if (treeNorm != 0) {
-        importances.foreach { case (idx, impt) =>
-          val normImpt = impt / treeNorm
-          totalImportances.changeValue(idx, normImpt, _ + normImpt)
-        }
-      }
-    }
-    // Normalize importances
-    normalizeMapValues(totalImportances)
-    // Construct vector
-    val d = if (numFeatures != -1) {
-      numFeatures
-    } else {
-      // Find max feature index used in trees
-      val maxFeatureIndex = trees.map(_.maxSplitFeatureIndex()).max
-      maxFeatureIndex + 1
-    }
-    if (d == 0) {
-      assert(totalImportances.size == 0, s"Unknown error in computing feature" +
-        s" importance: No splits found, but some non-zero importances.")
-    }
-    val (indices, values) = totalImportances.iterator.toSeq.sortBy(_._1).unzip
-    Vectors.sparse(d, indices.toArray, values.toArray)
-  }
-
-  /**
-   * Given a Decision Tree model, compute the importance of each feature.
-   * This generalizes the idea of "Gini" importance to other losses,
-   * following the explanation of Gini importance from "Random Forests" documentation
-   * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
-   *
-   * This feature importance is calculated as follows:
-   *  - importance(feature j) = sum (over nodes which split on feature j) of the gain,
-   *    where gain is scaled by the number of instances passing through node
-   *  - Normalize importances for tree to sum to 1.
-   *
-   * @param tree  Decision tree to compute importances for.
-   * @param numFeatures  Number of features in model (even if not all are explicitly used by
-   *                     the model).
-   *                     If -1, then numFeatures is set based on the max feature index in all trees.
-   * @return  Feature importance values, of length numFeatures.
-   */
-  private[ml] def featureImportances(tree: DecisionTreeModel, numFeatures: Int): Vector = {
-    featureImportances(Array(tree), numFeatures)
-  }
-
-  /**
-   * Recursive method for computing feature importances for one tree.
-   * This walks down the tree, adding to the importance of 1 feature at each node.
-   * @param node  Current node in recursion
-   * @param importances  Aggregate feature importances, modified by this method
-   */
-  private[impl] def computeFeatureImportance(
-      node: Node,
-      importances: OpenHashMap[Int, Double]): Unit = {
-    node match {
-      case n: InternalNode =>
-        val feature = n.split.featureIndex
-        val scaledGain = n.gain * n.impurityStats.count
-        importances.changeValue(feature, scaledGain, _ + scaledGain)
-        computeFeatureImportance(n.leftChild, importances)
-        computeFeatureImportance(n.rightChild, importances)
-      case n: LeafNode =>
-        // do nothing
-    }
-  }
-
-  /**
-   * Normalize the values of this map to sum to 1, in place.
-   * If all values are 0, this method does nothing.
-   * @param map  Map with non-negative values.
-   */
-  private[impl] def normalizeMapValues(map: OpenHashMap[Int, Double]): Unit = {
-    val total = map.map(_._2).sum
-    if (total != 0) {
-      val keys = map.iterator.map(_._1).toArray
-      keys.foreach { key => map.changeValue(key, 0.0, _ / total) }
-    }
-  }
-
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
index ef40c9068f..1fad9d6d8c 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
@@ -27,6 +27,7 @@ import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.tree.impurity.ImpurityCalculator
 import org.apache.spark.mllib.tree.model.{DecisionTreeModel => OldDecisionTreeModel}
 import org.apache.spark.sql.SQLContext
+import org.apache.spark.util.collection.OpenHashMap
 
 /**
  * Abstraction for Decision Tree models.
@@ -115,6 +116,125 @@ private[ml] trait TreeEnsembleModel {
   lazy val totalNumNodes: Int = trees.map(_.numNodes).sum
 }
 
+private[ml] object TreeEnsembleModel {
+
+  /**
+   * Given a tree ensemble model, compute the importance of each feature.
+   * This generalizes the idea of "Gini" importance to other losses,
+   * following the explanation of Gini importance from "Random Forests" documentation
+   * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
+   *
+   *  For collections of trees, including boosting and bagging, Hastie et al.
+   *  propose to use the average of single tree importances across all trees in the ensemble.
+   *
+   * This feature importance is calculated as follows:
+   *  - Average over trees:
+   *     - importance(feature j) = sum (over nodes which split on feature j) of the gain,
+   *       where gain is scaled by the number of instances passing through node
+   *     - Normalize importances for tree to sum to 1.
+   *  - Normalize feature importance vector to sum to 1.
+   *
+   *  References:
+   *  - Hastie, Tibshirani, Friedman. "The Elements of Statistical Learning, 2nd Edition." 2001.
+   *
+   * @param trees  Unweighted collection of trees
+   * @param numFeatures  Number of features in model (even if not all are explicitly used by
+   *                     the model).
+   *                     If -1, then numFeatures is set based on the max feature index in all trees.
+   * @return  Feature importance values, of length numFeatures.
+   */
+  def featureImportances(trees: Array[DecisionTreeModel], numFeatures: Int): Vector = {
+    val totalImportances = new OpenHashMap[Int, Double]()
+    trees.foreach { tree =>
+      // Aggregate feature importance vector for this tree
+      val importances = new OpenHashMap[Int, Double]()
+      computeFeatureImportance(tree.rootNode, importances)
+      // Normalize importance vector for this tree, and add it to total.
+      // TODO: In the future, also support normalizing by tree.rootNode.impurityStats.count?
+      val treeNorm = importances.map(_._2).sum
+      if (treeNorm != 0) {
+        importances.foreach { case (idx, impt) =>
+          val normImpt = impt / treeNorm
+          totalImportances.changeValue(idx, normImpt, _ + normImpt)
+        }
+      }
+    }
+    // Normalize importances
+    normalizeMapValues(totalImportances)
+    // Construct vector
+    val d = if (numFeatures != -1) {
+      numFeatures
+    } else {
+      // Find max feature index used in trees
+      val maxFeatureIndex = trees.map(_.maxSplitFeatureIndex()).max
+      maxFeatureIndex + 1
+    }
+    if (d == 0) {
+      assert(totalImportances.size == 0, s"Unknown error in computing feature" +
+        s" importance: No splits found, but some non-zero importances.")
+    }
+    val (indices, values) = totalImportances.iterator.toSeq.sortBy(_._1).unzip
+    Vectors.sparse(d, indices.toArray, values.toArray)
+  }
+
+  /**
+   * Given a Decision Tree model, compute the importance of each feature.
+   * This generalizes the idea of "Gini" importance to other losses,
+   * following the explanation of Gini importance from "Random Forests" documentation
+   * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
+   *
+   * This feature importance is calculated as follows:
+   *  - importance(feature j) = sum (over nodes which split on feature j) of the gain,
+   *    where gain is scaled by the number of instances passing through node
+   *  - Normalize importances for tree to sum to 1.
+   *
+   * @param tree  Decision tree to compute importances for.
+   * @param numFeatures  Number of features in model (even if not all are explicitly used by
+   *                     the model).
+   *                     If -1, then numFeatures is set based on the max feature index in all trees.
+   * @return  Feature importance values, of length numFeatures.
+   */
+  def featureImportances(tree: DecisionTreeModel, numFeatures: Int): Vector = {
+    featureImportances(Array(tree), numFeatures)
+  }
+
+  /**
+   * Recursive method for computing feature importances for one tree.
+   * This walks down the tree, adding to the importance of 1 feature at each node.
+   *
+   * @param node  Current node in recursion
+   * @param importances  Aggregate feature importances, modified by this method
+   */
+  def computeFeatureImportance(
+      node: Node,
+      importances: OpenHashMap[Int, Double]): Unit = {
+    node match {
+      case n: InternalNode =>
+        val feature = n.split.featureIndex
+        val scaledGain = n.gain * n.impurityStats.count
+        importances.changeValue(feature, scaledGain, _ + scaledGain)
+        computeFeatureImportance(n.leftChild, importances)
+        computeFeatureImportance(n.rightChild, importances)
+      case n: LeafNode =>
+      // do nothing
+    }
+  }
+
+  /**
+   * Normalize the values of this map to sum to 1, in place.
+   * If all values are 0, this method does nothing.
+   *
+   * @param map  Map with non-negative values.
+   */
+  def normalizeMapValues(map: OpenHashMap[Int, Double]): Unit = {
+    val total = map.map(_._2).sum
+    if (total != 0) {
+      val keys = map.iterator.map(_._1).toArray
+      keys.foreach { key => map.changeValue(key, 0.0, _ / total) }
+    }
+  }
+}
+
 /** Helper classes for tree model persistence */
 private[ml] object DecisionTreeModelReadWrite {
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
index f3680ed044..bf7481e8a3 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
@@ -120,6 +120,31 @@ class GBTClassifierSuite extends SparkFunSuite with MLlibTestSparkContext {
   }
   */
 
+  /////////////////////////////////////////////////////////////////////////////
+  // Tests of feature importance
+  /////////////////////////////////////////////////////////////////////////////
+  test("Feature importance with toy data") {
+    val numClasses = 2
+    val gbt = new GBTClassifier()
+      .setImpurity("Gini")
+      .setMaxDepth(3)
+      .setMaxIter(5)
+      .setSubsamplingRate(1.0)
+      .setStepSize(0.5)
+      .setSeed(123)
+
+    // In this data, feature 1 is very important.
+    val data: RDD[LabeledPoint] = TreeTests.featureImportanceData(sc)
+    val categoricalFeatures = Map.empty[Int, Int]
+    val df: DataFrame = TreeTests.setMetadata(data, categoricalFeatures, numClasses)
+
+    val importances = gbt.fit(df).featureImportances
+    val mostImportantFeature = importances.argmax
+    assert(mostImportantFeature === 1)
+    assert(importances.toArray.sum === 1.0)
+    assert(importances.toArray.forall(_ >= 0.0))
+  }
+
   /////////////////////////////////////////////////////////////////////////////
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
index 84148a8a4a..dfb8418086 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
@@ -131,6 +131,29 @@ class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
   }
   */
 
+  /////////////////////////////////////////////////////////////////////////////
+  // Tests of feature importance
+  /////////////////////////////////////////////////////////////////////////////
+  test("Feature importance with toy data") {
+    val gbt = new GBTRegressor()
+      .setMaxDepth(3)
+      .setMaxIter(5)
+      .setSubsamplingRate(1.0)
+      .setStepSize(0.5)
+      .setSeed(123)
+
+    // In this data, feature 1 is very important.
+    val data: RDD[LabeledPoint] = TreeTests.featureImportanceData(sc)
+    val categoricalFeatures = Map.empty[Int, Int]
+    val df: DataFrame = TreeTests.setMetadata(data, categoricalFeatures, 0)
+
+    val importances = gbt.fit(df).featureImportances
+    val mostImportantFeature = importances.argmax
+    assert(mostImportantFeature === 1)
+    assert(importances.toArray.sum === 1.0)
+    assert(importances.toArray.forall(_ >= 0.0))
+  }
+
   /////////////////////////////////////////////////////////////////////////////
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
index 361366fde7..441338e74e 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
@@ -471,7 +471,7 @@ class RandomForestSuite extends SparkFunSuite with MLlibTestSparkContext {
     // Test feature importance computed at different subtrees.
     def testNode(node: Node, expected: Map[Int, Double]): Unit = {
       val map = new OpenHashMap[Int, Double]()
-      RandomForest.computeFeatureImportance(node, map)
+      TreeEnsembleModel.computeFeatureImportance(node, map)
       assert(mapToVec(map.toMap) ~== mapToVec(expected) relTol 0.01)
     }
 
@@ -493,7 +493,7 @@ class RandomForestSuite extends SparkFunSuite with MLlibTestSparkContext {
       new DecisionTreeClassificationModel(root, numFeatures = 2, numClasses = 3)
         .asInstanceOf[DecisionTreeModel]
     }
-    val importances: Vector = RandomForest.featureImportances(trees, 2)
+    val importances: Vector = TreeEnsembleModel.featureImportances(trees, 2)
     val tree2norm = feature0importance + feature1importance
     val expected = Vectors.dense((1.0 + feature0importance / tree2norm) / 2.0,
       (feature1importance / tree2norm) / 2.0)
@@ -504,7 +504,7 @@ class RandomForestSuite extends SparkFunSuite with MLlibTestSparkContext {
     val map = new OpenHashMap[Int, Double]()
     map(0) = 1.0
     map(2) = 2.0
-    RandomForest.normalizeMapValues(map)
+    TreeEnsembleModel.normalizeMapValues(map)
     val expected = Map(0 -> 1.0 / 3.0, 2 -> 2.0 / 3.0)
     assert(mapToVec(map.toMap) ~== mapToVec(expected) relTol 0.01)
   }
-- 
cgit v1.2.3


From 425bcf6d6844732fe402af05472ad87b4e032cb6 Mon Sep 17 00:00:00 2001
From: Bryan Cutler <cutlerb@gmail.com>
Date: Tue, 29 Mar 2016 12:30:30 +0200
Subject: [SPARK-13963][ML] Adding binary toggle param to HashingTF

## What changes were proposed in this pull request?
Adding binary toggle parameter to ml.feature.HashingTF, as well as mllib.feature.HashingTF since the former wraps this functionality.  This parameter, if true, will set non-zero valued term counts to 1 to transform term count features to binary values that are well suited for discrete probability models.

## How was this patch tested?
Added unit tests for ML and MLlib

Author: Bryan Cutler <cutlerb@gmail.com>

Closes #11832 from BryanCutler/binary-param-HashingTF-SPARK-13963.
---
 .../org/apache/spark/ml/feature/HashingTF.scala    | 23 ++++++++++++++++++---
 .../org/apache/spark/mllib/feature/HashingTF.scala | 15 +++++++++++++-
 .../apache/spark/ml/feature/HashingTFSuite.scala   | 24 +++++++++++++++++++++-
 .../spark/mllib/feature/HashingTFSuite.scala       | 12 +++++++++++
 4 files changed, 69 insertions(+), 5 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
index 61a78d73c4..0f7ae5a100 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
@@ -20,7 +20,7 @@ package org.apache.spark.ml.feature
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml.Transformer
 import org.apache.spark.ml.attribute.AttributeGroup
-import org.apache.spark.ml.param.{IntParam, ParamMap, ParamValidators}
+import org.apache.spark.ml.param.{BooleanParam, IntParam, ParamMap, ParamValidators}
 import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.feature
@@ -52,7 +52,18 @@ class HashingTF(override val uid: String)
   val numFeatures = new IntParam(this, "numFeatures", "number of features (> 0)",
     ParamValidators.gt(0))
 
-  setDefault(numFeatures -> (1 << 18))
+  /**
+   * Binary toggle to control term frequency counts.
+   * If true, all non-zero counts are set to 1.  This is useful for discrete probabilistic
+   * models that model binary events rather than integer counts.
+   * (default = false)
+   * @group param
+   */
+  val binary = new BooleanParam(this, "binary", "If true, all non zero counts are set to 1. " +
+    "This is useful for discrete probabilistic models that model binary events rather " +
+    "than integer counts")
+
+  setDefault(numFeatures -> (1 << 18), binary -> false)
 
   /** @group getParam */
   def getNumFeatures: Int = $(numFeatures)
@@ -60,9 +71,15 @@ class HashingTF(override val uid: String)
   /** @group setParam */
   def setNumFeatures(value: Int): this.type = set(numFeatures, value)
 
+  /** @group getParam */
+  def getBinary: Boolean = $(binary)
+
+  /** @group setParam */
+  def setBinary(value: Boolean): this.type = set(binary, value)
+
   override def transform(dataset: DataFrame): DataFrame = {
     val outputSchema = transformSchema(dataset.schema)
-    val hashingTF = new feature.HashingTF($(numFeatures))
+    val hashingTF = new feature.HashingTF($(numFeatures)).setBinary($(binary))
     val t = udf { terms: Seq[_] => hashingTF.transform(terms) }
     val metadata = outputSchema($(outputCol)).metadata
     dataset.select(col("*"), t(col($(inputCol))).as($(outputCol), metadata))
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/feature/HashingTF.scala b/mllib/src/main/scala/org/apache/spark/mllib/feature/HashingTF.scala
index c93ed64183..47c9e850a0 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/feature/HashingTF.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/feature/HashingTF.scala
@@ -36,11 +36,23 @@ import org.apache.spark.util.Utils
 @Since("1.1.0")
 class HashingTF(val numFeatures: Int) extends Serializable {
 
+  private var binary = false
+
   /**
    */
   @Since("1.1.0")
   def this() = this(1 << 20)
 
+  /**
+   * If true, term frequency vector will be binary such that non-zero term counts will be set to 1
+   * (default: false)
+   */
+  @Since("2.0.0")
+  def setBinary(value: Boolean): this.type = {
+    binary = value
+    this
+  }
+
   /**
    * Returns the index of the input term.
    */
@@ -53,9 +65,10 @@ class HashingTF(val numFeatures: Int) extends Serializable {
   @Since("1.1.0")
   def transform(document: Iterable[_]): Vector = {
     val termFrequencies = mutable.HashMap.empty[Int, Double]
+    val setTF = if (binary) (i: Int) => 1.0 else (i: Int) => termFrequencies.getOrElse(i, 0.0) + 1.0
     document.foreach { term =>
       val i = indexOf(term)
-      termFrequencies.put(i, termFrequencies.getOrElse(i, 0.0) + 1.0)
+      termFrequencies.put(i, setTF(i))
     }
     Vectors.sparse(numFeatures, termFrequencies.toSeq)
   }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/HashingTFSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/HashingTFSuite.scala
index 0dcd0f4946..addd733c20 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/HashingTFSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/HashingTFSuite.scala
@@ -46,12 +46,30 @@ class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
     require(attrGroup.numAttributes === Some(n))
     val features = output.select("features").first().getAs[Vector](0)
     // Assume perfect hash on "a", "b", "c", and "d".
-    def idx(any: Any): Int = Utils.nonNegativeMod(any.##, n)
+    def idx: Any => Int = featureIdx(n)
     val expected = Vectors.sparse(n,
       Seq((idx("a"), 2.0), (idx("b"), 2.0), (idx("c"), 1.0), (idx("d"), 1.0)))
     assert(features ~== expected absTol 1e-14)
   }
 
+  test("applying binary term freqs") {
+    val df = sqlContext.createDataFrame(Seq(
+      (0, "a a b c c c".split(" ").toSeq)
+    )).toDF("id", "words")
+    val n = 100
+    val hashingTF = new HashingTF()
+        .setInputCol("words")
+        .setOutputCol("features")
+        .setNumFeatures(n)
+        .setBinary(true)
+    val output = hashingTF.transform(df)
+    val features = output.select("features").first().getAs[Vector](0)
+    def idx: Any => Int = featureIdx(n)  // Assume perfect hash on input features
+    val expected = Vectors.sparse(n,
+      Seq((idx("a"), 1.0), (idx("b"), 1.0), (idx("c"), 1.0)))
+    assert(features ~== expected absTol 1e-14)
+  }
+
   test("read/write") {
     val t = new HashingTF()
       .setInputCol("myInputCol")
@@ -59,4 +77,8 @@ class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
       .setNumFeatures(10)
     testDefaultReadWrite(t)
   }
+
+  private def featureIdx(numFeatures: Int)(term: Any): Int = {
+    Utils.nonNegativeMod(term.##, numFeatures)
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/feature/HashingTFSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/feature/HashingTFSuite.scala
index cf279c0233..6c07e3a5ce 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/feature/HashingTFSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/feature/HashingTFSuite.scala
@@ -20,6 +20,7 @@ package org.apache.spark.mllib.feature
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 
 class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext {
 
@@ -48,4 +49,15 @@ class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext {
     val docs = sc.parallelize(localDocs, 2)
     assert(hashingTF.transform(docs).collect().toSet === localDocs.map(hashingTF.transform).toSet)
   }
+
+  test("applying binary term freqs") {
+    val hashingTF = new HashingTF(100).setBinary(true)
+    val doc = "a a b c c c".split(" ")
+    val n = hashingTF.numFeatures
+    val expected = Vectors.sparse(n, Seq(
+      (hashingTF.indexOf("a"), 1.0),
+      (hashingTF.indexOf("b"), 1.0),
+      (hashingTF.indexOf("c"), 1.0)))
+    assert(hashingTF.transform(doc) ~== expected absTol 1e-14)
+  }
 }
-- 
cgit v1.2.3


From d2a819a6363190b946986ebf6f8001d520098c3b Mon Sep 17 00:00:00 2001
From: Yuhao Yang <hhbyyh@gmail.com>
Date: Tue, 29 Mar 2016 09:16:50 -0700
Subject: [SPARK-14154][MLLIB] Simplify the implementation for
 Kolmogorov–Smirnov test
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## What changes were proposed in this pull request?
jira: https://issues.apache.org/jira/browse/SPARK-14154

I just read the code for KolmogorovSmirnovTest and find it could be much simplified following the original definition.

Send a PR for discussion

## How was this patch tested?
unit test

Author: Yuhao Yang <hhbyyh@gmail.com>

Closes #11954 from hhbyyh/ksoptimize.
---
 .../mllib/stat/test/KolmogorovSmirnovTest.scala    | 77 ++--------------------
 1 file changed, 4 insertions(+), 73 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala b/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
index baf9e5e7d1..0ec8975fed 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
@@ -64,11 +64,10 @@ private[stat] object KolmogorovSmirnovTest extends Logging {
    */
   def testOneSample(data: RDD[Double], cdf: Double => Double): KolmogorovSmirnovTestResult = {
     val n = data.count().toDouble
-    val localData = data.sortBy(x => x).mapPartitions { part =>
-      val partDiffs = oneSampleDifferences(part, n, cdf) // local distances
-      searchOneSampleCandidates(partDiffs) // candidates: local extrema
-    }.collect()
-    val ksStat = searchOneSampleStatistic(localData, n) // result: global extreme
+    val ksStat = data.sortBy(x => x).zipWithIndex().map { case (v, i) =>
+      val f = cdf(v)
+      math.max(f - i / n, (i + 1) / n - f)
+    }.max()
     evalOneSampleP(ksStat, n.toLong)
   }
 
@@ -84,74 +83,6 @@ private[stat] object KolmogorovSmirnovTest extends Logging {
     testOneSample(data, cdf)
   }
 
-  /**
-   * Calculate unadjusted distances between the empirical CDF and the theoretical CDF in a
-   * partition
-   * @param partData `Iterator[Double]` 1 partition of a sorted RDD
-   * @param n `Double` the total size of the RDD
-   * @param cdf `Double => Double` a function the calculates the theoretical CDF of a value
-   * @return `Iterator[(Double, Double)] `Unadjusted (ie. off by a constant) potential extrema
-   *        in a partition. The first element corresponds to the (empirical CDF - 1/N) - CDF,
-   *        the second element corresponds to empirical CDF - CDF.  We can then search the resulting
-   *        iterator for the minimum of the first and the maximum of the second element, and provide
-   *        this as a partition's candidate extrema
-   */
-  private def oneSampleDifferences(partData: Iterator[Double], n: Double, cdf: Double => Double)
-    : Iterator[(Double, Double)] = {
-    // zip data with index (within that partition)
-    // calculate local (unadjusted) empirical CDF and subtract CDF
-    partData.zipWithIndex.map { case (v, ix) =>
-      // dp and dl are later adjusted by constant, when global info is available
-      val dp = (ix + 1) / n
-      val dl = ix / n
-      val cdfVal = cdf(v)
-      (dl - cdfVal, dp - cdfVal)
-    }
-  }
-
-  /**
-   * Search the unadjusted differences in a partition and return the
-   * two extrema (furthest below and furthest above CDF), along with a count of elements in that
-   * partition
-   * @param partDiffs `Iterator[(Double, Double)]` the unadjusted differences between empirical CDF
-   *                 and CDFin a partition, which come as a tuple of
-   *                 (empirical CDF - 1/N - CDF, empirical CDF - CDF)
-   * @return `Iterator[(Double, Double, Double)]` the local extrema and a count of elements
-   */
-  private def searchOneSampleCandidates(partDiffs: Iterator[(Double, Double)])
-    : Iterator[(Double, Double, Double)] = {
-    val initAcc = (Double.MaxValue, Double.MinValue, 0.0)
-    val pResults = partDiffs.foldLeft(initAcc) { case ((pMin, pMax, pCt), (dl, dp)) =>
-      (math.min(pMin, dl), math.max(pMax, dp), pCt + 1)
-    }
-    val results = if (pResults == initAcc) Array[(Double, Double, Double)]() else Array(pResults)
-    results.iterator
-  }
-
-  /**
-   * Find the global maximum distance between empirical CDF and CDF (i.e. the KS statistic) after
-   * adjusting local extrema estimates from individual partitions with the amount of elements in
-   * preceding partitions
-   * @param localData `Array[(Double, Double, Double)]` A local array containing the collected
-   *                 results of `searchOneSampleCandidates` across all partitions
-   * @param n `Double`The size of the RDD
-   * @return The one-sample Kolmogorov Smirnov Statistic
-   */
-  private def searchOneSampleStatistic(localData: Array[(Double, Double, Double)], n: Double)
-    : Double = {
-    val initAcc = (Double.MinValue, 0.0)
-    // adjust differences based on the number of elements preceding it, which should provide
-    // the correct distance between empirical CDF and CDF
-    val results = localData.foldLeft(initAcc) { case ((prevMax, prevCt), (minCand, maxCand, ct)) =>
-      val adjConst = prevCt / n
-      val dist1 = math.abs(minCand + adjConst)
-      val dist2 = math.abs(maxCand + adjConst)
-      val maxVal = Array(prevMax, dist1, dist2).max
-      (maxVal, prevCt + ct)
-    }
-    results._1
-  }
-
   /**
    * A convenience function that allows running the KS test for 1 set of sample data against
    * a named distribution
-- 
cgit v1.2.3


From 529d6ce8f96ef2b4a57c2d9066c7d80466e36209 Mon Sep 17 00:00:00 2001
From: Xusen Yin <yinxusen@gmail.com>
Date: Wed, 30 Mar 2016 14:32:29 -0700
Subject: [SPARK-14181] TrainValidationSplit should have HasSeed

https://issues.apache.org/jira/browse/SPARK-14181

TrainValidationSplit should have HasSeed for the random split of RDD. I also changed the random split from the RDD function to the DataFrame function.

Author: Xusen Yin <yinxusen@gmail.com>

Closes #11985 from yinxusen/SPARK-14181.
---
 .../org/apache/spark/ml/tuning/TrainValidationSplit.scala | 15 ++++++++++-----
 .../spark/ml/tuning/TrainValidationSplitSuite.scala       |  4 ++++
 2 files changed, 14 insertions(+), 5 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
index 4d1d6364d7..07330bb6b0 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
@@ -25,6 +25,7 @@ import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.evaluation.Evaluator
 import org.apache.spark.ml.param.{DoubleParam, ParamMap, ParamValidators}
+import org.apache.spark.ml.param.shared.HasSeed
 import org.apache.spark.ml.util._
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.types.StructType
@@ -32,7 +33,7 @@ import org.apache.spark.sql.types.StructType
 /**
  * Params for [[TrainValidationSplit]] and [[TrainValidationSplitModel]].
  */
-private[ml] trait TrainValidationSplitParams extends ValidatorParams {
+private[ml] trait TrainValidationSplitParams extends ValidatorParams with HasSeed {
   /**
    * Param for ratio between train and validation data. Must be between 0 and 1.
    * Default: 0.75
@@ -80,6 +81,10 @@ class TrainValidationSplit @Since("1.5.0") (@Since("1.5.0") override val uid: St
   @Since("1.5.0")
   def setTrainRatio(value: Double): this.type = set(trainRatio, value)
 
+  /** @group setParam */
+  @Since("2.0.0")
+  def setSeed(value: Long): this.type = set(seed, value)
+
   @Since("1.5.0")
   override def fit(dataset: DataFrame): TrainValidationSplitModel = {
     val schema = dataset.schema
@@ -91,10 +96,10 @@ class TrainValidationSplit @Since("1.5.0") (@Since("1.5.0") override val uid: St
     val numModels = epm.length
     val metrics = new Array[Double](epm.length)
 
-    val Array(training, validation) =
-      dataset.rdd.randomSplit(Array($(trainRatio), 1 - $(trainRatio)))
-    val trainingDataset = sqlCtx.createDataFrame(training, schema).cache()
-    val validationDataset = sqlCtx.createDataFrame(validation, schema).cache()
+    val Array(trainingDataset, validationDataset) =
+      dataset.randomSplit(Array($(trainRatio), 1 - $(trainRatio)), $(seed))
+    trainingDataset.cache()
+    validationDataset.cache()
 
     // multi-model training
     logDebug(s"Train split with multiple sets of parameters.")
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
index 7cf7b3e087..4030956fab 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
@@ -48,6 +48,7 @@ class TrainValidationSplitSuite
       .setEstimatorParamMaps(lrParamMaps)
       .setEvaluator(eval)
       .setTrainRatio(0.5)
+      .setSeed(42L)
     val cvModel = cv.fit(dataset)
     val parent = cvModel.bestModel.parent.asInstanceOf[LogisticRegression]
     assert(cv.getTrainRatio === 0.5)
@@ -72,6 +73,7 @@ class TrainValidationSplitSuite
       .setEstimatorParamMaps(lrParamMaps)
       .setEvaluator(eval)
       .setTrainRatio(0.5)
+      .setSeed(42L)
     val cvModel = cv.fit(dataset)
     val parent = cvModel.bestModel.parent.asInstanceOf[LinearRegression]
     assert(parent.getRegParam === 0.001)
@@ -120,6 +122,7 @@ class TrainValidationSplitSuite
       .setEvaluator(evaluator)
       .setTrainRatio(0.5)
       .setEstimatorParamMaps(paramMaps)
+      .setSeed(42L)
 
     val tvs2 = testDefaultReadWrite(tvs, testParams = false)
 
@@ -140,6 +143,7 @@ class TrainValidationSplitSuite
       .set(tvs.evaluator, evaluator)
       .set(tvs.trainRatio, 0.5)
       .set(tvs.estimatorParamMaps, paramMaps)
+      .set(tvs.seed, 42L)
 
     val tvs2 = testDefaultReadWrite(tvs, testParams = false)
 
-- 
cgit v1.2.3


From 5dc948e8125fd27646a7f1e8991948a45b3f9c50 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Wed, 30 Mar 2016 14:57:38 -0700
Subject: [MINOR][ML] Fix the wrong param name of LDA topicDistributionCol

## What changes were proposed in this pull request?
Fix the wrong param name of LDA ```topicDistributionCol```.
## How was this patch tested?
No tests.

cc jkbradley

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12065 from yanboliang/lda-topicDistributionCol.
---
 mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
index fe6a37fd6d..60cc345565 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
@@ -176,7 +176,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * @group param
    */
   @Since("1.6.0")
-  final val topicDistributionCol = new Param[String](this, "topicDistribution", "Output column" +
+  final val topicDistributionCol = new Param[String](this, "topicDistributionCol", "Output column" +
     " with estimates of the topic mixture distribution for each document (often called \"theta\"" +
     " in the literature).  Returns a vector of zeros for an empty document.")
 
-- 
cgit v1.2.3


From ca458618d8ee659ffa9a081083cd475a440fa8ff Mon Sep 17 00:00:00 2001
From: Yuhao Yang <hhbyyh@gmail.com>
Date: Wed, 30 Mar 2016 15:58:19 -0700
Subject: [SPARK-11507][MLLIB] add compact in Matrices fromBreeze

jira: https://issues.apache.org/jira/browse/SPARK-11507
"In certain situations when adding two block matrices, I get an error regarding colPtr and the operation fails. External issue URL includes full error and code for reproducing the problem."

root cause: colPtr.last does NOT always equal to values.length in breeze SCSMatrix, which fails the require in SparseMatrix.

easy step to repro:
```
val m1: BM[Double] = new CSCMatrix[Double] (Array (1.0, 1, 1), 3, 3, Array (0, 1, 2, 3), Array (0, 1, 2) )
val m2: BM[Double] = new CSCMatrix[Double] (Array (1.0, 2, 2, 4), 3, 3, Array (0, 0, 2, 4), Array (1, 2, 1, 2) )
val sum = m1 + m2
Matrices.fromBreeze(sum)
```

Solution: By checking the code in [CSCMatrix](https://github.com/scalanlp/breeze/blob/28000a7b901bc3cfbbbf5c0bce1d0a5dda8281b0/math/src/main/scala/breeze/linalg/CSCMatrix.scala), CSCMatrix in breeze can have extra zeros in the end of data array. Invoking compact will make sure it aligns with the require of SparseMatrix. This should add limited overhead as the actual compact operation is only performed when necessary.

Author: Yuhao Yang <hhbyyh@gmail.com>

Closes #9520 from hhbyyh/matricesFromBreeze.
---
 .../main/scala/org/apache/spark/mllib/linalg/Matrices.scala  | 10 +++++++++-
 .../scala/org/apache/spark/mllib/linalg/MatricesSuite.scala  | 12 ++++++++++++
 2 files changed, 21 insertions(+), 1 deletion(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
index a09bc65cf3..6e571fe35a 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
@@ -940,8 +940,16 @@ object Matrices {
       case dm: BDM[Double] =>
         new DenseMatrix(dm.rows, dm.cols, dm.data, dm.isTranspose)
       case sm: BSM[Double] =>
+        // Spark-11507. work around breeze issue 479.
+        val mat = if (sm.colPtrs.last != sm.data.length) {
+          val matCopy = sm.copy
+          matCopy.compact()
+          matCopy
+        } else {
+          sm
+        }
         // There is no isTranspose flag for sparse matrices in Breeze
-        new SparseMatrix(sm.rows, sm.cols, sm.colPtrs, sm.rowIndices, sm.data)
+        new SparseMatrix(mat.rows, mat.cols, mat.colPtrs, mat.rowIndices, mat.data)
       case _ =>
         throw new UnsupportedOperationException(
           s"Do not support conversion from type ${breeze.getClass.getName}.")
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
index 57907f415c..e289724cda 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
@@ -19,6 +19,7 @@ package org.apache.spark.mllib.linalg
 
 import java.util.Random
 
+import breeze.linalg.{CSCMatrix, Matrix => BM}
 import org.mockito.Mockito.when
 import org.scalatest.mock.MockitoSugar._
 import scala.collection.mutable.{Map => MutableMap}
@@ -499,6 +500,17 @@ class MatricesSuite extends SparkFunSuite {
     assert(sm1.numActives === 3)
   }
 
+  test("fromBreeze with sparse matrix") {
+    // colPtr.last does NOT always equal to values.length in breeze SCSMatrix and
+    // invocation of compact() may be necessary. Refer to SPARK-11507
+    val bm1: BM[Double] = new CSCMatrix[Double](
+      Array(1.0, 1, 1), 3, 3, Array(0, 1, 2, 3), Array(0, 1, 2))
+    val bm2: BM[Double] = new CSCMatrix[Double](
+      Array(1.0, 2, 2, 4), 3, 3, Array(0, 0, 2, 4), Array(1, 2, 1, 2))
+    val sum = bm1 + bm2
+    Matrices.fromBreeze(sum)
+  }
+
   test("row/col iterator") {
     val dm = new DenseMatrix(3, 2, Array(0, 1, 2, 3, 4, 0))
     val sm = dm.toSparse
-- 
cgit v1.2.3


From 208fff3ac87f200fd4e6f0407d70bf81cf8c556f Mon Sep 17 00:00:00 2001
From: Dongjoon Hyun <dongjoon@apache.org>
Date: Thu, 31 Mar 2016 09:39:15 -0700
Subject: [SPARK-14164][MLLIB] Improve input layer validation of
 MultilayerPerceptronClassifier

## What changes were proposed in this pull request?

This issue improves an input layer validation and adds related testcases to MultilayerPerceptronClassifier.

```scala
-    // TODO: how to check ALSO that all elements are greater than 0?
-    ParamValidators.arrayLengthGt(1)
+    (t: Array[Int]) => t.forall(ParamValidators.gt(0)) && t.length > 1
```

## How was this patch tested?

Pass the Jenkins tests including the new testcases.

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #11964 from dongjoon-hyun/SPARK-14164.
---
 .../classification/MultilayerPerceptronClassifier.scala |  3 +--
 .../MultilayerPerceptronClassifierSuite.scala           | 17 +++++++++++++++++
 2 files changed, 18 insertions(+), 2 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
index f6de5f2df4..7ce3ec68da 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
@@ -43,8 +43,7 @@ private[ml] trait MultilayerPerceptronParams extends PredictorParams
     "Sizes of layers from input layer to output layer" +
       " E.g., Array(780, 100, 10) means 780 inputs, " +
       "one hidden layer with 100 neurons and output layer of 10 neurons.",
-    // TODO: how to check ALSO that all elements are greater than 0?
-    ParamValidators.arrayLengthGt(1)
+    (t: Array[Int]) => t.forall(ParamValidators.gt(0)) && t.length > 1
   )
 
   /** @group getParam */
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
index 5df8e6a847..53c7a559e3 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
@@ -43,6 +43,23 @@ class MultilayerPerceptronClassifierSuite
     ).toDF("features", "label")
   }
 
+  test("Input Validation") {
+    val mlpc = new MultilayerPerceptronClassifier()
+    intercept[IllegalArgumentException] {
+      mlpc.setLayers(Array[Int]())
+    }
+    intercept[IllegalArgumentException] {
+      mlpc.setLayers(Array[Int](1))
+    }
+    intercept[IllegalArgumentException] {
+      mlpc.setLayers(Array[Int](0, 1))
+    }
+    intercept[IllegalArgumentException] {
+      mlpc.setLayers(Array[Int](1, 0))
+    }
+    mlpc.setLayers(Array[Int](1, 1))
+  }
+
   test("XOR function learning as binary classification problem with two outputs.") {
     val layers = Array[Int](2, 5, 2)
     val trainer = new MultilayerPerceptronClassifier()
-- 
cgit v1.2.3


From a0a1991580ed24230f88cae9f5a4dfbe58f03b28 Mon Sep 17 00:00:00 2001
From: Yuhao Yang <hhbyyh@gmail.com>
Date: Thu, 31 Mar 2016 11:12:40 -0700
Subject: [SPARK-13782][ML] Model export/import for spark.ml: BisectingKMeans

## What changes were proposed in this pull request?
jira: https://issues.apache.org/jira/browse/SPARK-13782
Model export/import for BisectingKMeans in spark.ml and mllib

## How was this patch tested?

unit tests

Author: Yuhao Yang <hhbyyh@gmail.com>

Closes #11933 from hhbyyh/bisectingsave.
---
 .../spark/ml/clustering/BisectingKMeans.scala      | 59 +++++++++++--
 .../spark/mllib/clustering/BisectingKMeans.scala   |  2 +-
 .../mllib/clustering/BisectingKMeansModel.scala    | 98 +++++++++++++++++++++-
 .../spark/ml/clustering/BisectingKMeansSuite.scala | 22 ++++-
 .../mllib/clustering/BisectingKMeansSuite.scala    | 18 ++++
 5 files changed, 190 insertions(+), 9 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala
index f014a1d572..55f751c57f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala
@@ -17,11 +17,13 @@
 
 package org.apache.spark.ml.clustering
 
+import org.apache.hadoop.fs.Path
+
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml.{Estimator, Model}
-import org.apache.spark.ml.param.{IntParam, Param, ParamMap, Params}
+import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared._
-import org.apache.spark.ml.util.{Identifiable, SchemaUtils}
+import org.apache.spark.ml.util._
 import org.apache.spark.mllib.clustering.
   {BisectingKMeans => MLlibBisectingKMeans, BisectingKMeansModel => MLlibBisectingKMeansModel}
 import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
@@ -49,7 +51,7 @@ private[clustering] trait BisectingKMeansParams extends Params
 
   /** @group expertParam */
   @Since("2.0.0")
-  final val minDivisibleClusterSize = new Param[Double](
+  final val minDivisibleClusterSize = new DoubleParam(
     this,
     "minDivisibleClusterSize",
     "the minimum number of points (if >= 1.0) or the minimum proportion",
@@ -81,7 +83,7 @@ private[clustering] trait BisectingKMeansParams extends Params
 class BisectingKMeansModel private[ml] (
     @Since("2.0.0") override val uid: String,
     private val parentModel: MLlibBisectingKMeansModel
-  ) extends Model[BisectingKMeansModel] with BisectingKMeansParams {
+  ) extends Model[BisectingKMeansModel] with BisectingKMeansParams with MLWritable {
 
   @Since("2.0.0")
   override def copy(extra: ParamMap): BisectingKMeansModel = {
@@ -115,6 +117,44 @@ class BisectingKMeansModel private[ml] (
     val data = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
     parentModel.computeCost(data)
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new BisectingKMeansModel.BisectingKMeansModelWriter(this)
+}
+
+object BisectingKMeansModel extends MLReadable[BisectingKMeansModel] {
+  @Since("2.0.0")
+  override def read: MLReader[BisectingKMeansModel] = new BisectingKMeansModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): BisectingKMeansModel = super.load(path)
+
+  /** [[MLWriter]] instance for [[BisectingKMeansModel]] */
+  private[BisectingKMeansModel]
+  class BisectingKMeansModelWriter(instance: BisectingKMeansModel) extends MLWriter {
+
+    override protected def saveImpl(path: String): Unit = {
+      // Save metadata and Params
+      DefaultParamsWriter.saveMetadata(instance, path, sc)
+      val dataPath = new Path(path, "data").toString
+      instance.parentModel.save(sc, dataPath)
+    }
+  }
+
+  private class BisectingKMeansModelReader extends MLReader[BisectingKMeansModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[BisectingKMeansModel].getName
+
+    override def load(path: String): BisectingKMeansModel = {
+      val metadata = DefaultParamsReader.loadMetadata(path, sc, className)
+      val dataPath = new Path(path, "data").toString
+      val mllibModel = MLlibBisectingKMeansModel.load(sc, dataPath)
+      val model = new BisectingKMeansModel(metadata.uid, mllibModel)
+      DefaultParamsReader.getAndSetParams(model, metadata)
+      model
+    }
+  }
 }
 
 /**
@@ -137,7 +177,7 @@ class BisectingKMeansModel private[ml] (
 @Experimental
 class BisectingKMeans @Since("2.0.0") (
     @Since("2.0.0") override val uid: String)
-  extends Estimator[BisectingKMeansModel] with BisectingKMeansParams {
+  extends Estimator[BisectingKMeansModel] with BisectingKMeansParams with DefaultParamsWritable {
 
   setDefault(
     k -> 4,
@@ -148,7 +188,7 @@ class BisectingKMeans @Since("2.0.0") (
   override def copy(extra: ParamMap): BisectingKMeans = defaultCopy(extra)
 
   @Since("2.0.0")
-  def this() = this(Identifiable.randomUID("bisecting k-means"))
+  def this() = this(Identifiable.randomUID("bisecting-kmeans"))
 
   /** @group setParam */
   @Since("2.0.0")
@@ -194,3 +234,10 @@ class BisectingKMeans @Since("2.0.0") (
   }
 }
 
+
+@Since("2.0.0")
+object BisectingKMeans extends DefaultParamsReadable[BisectingKMeans] {
+
+  @Since("2.0.0")
+  override def load(path: String): BisectingKMeans = super.load(path)
+}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeans.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeans.scala
index 64b838a1db..e4bd0dc25e 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeans.scala
@@ -411,7 +411,7 @@ private object BisectingKMeans extends Serializable {
 private[clustering] class ClusteringTreeNode private[clustering] (
     val index: Int,
     val size: Long,
-    private val centerWithNorm: VectorWithNorm,
+    private[clustering] val centerWithNorm: VectorWithNorm,
     val cost: Double,
     val height: Double,
     val children: Array[ClusteringTreeNode]) extends Serializable {
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeansModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeansModel.scala
index 01a0d31f14..c3b5b8b790 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeansModel.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/BisectingKMeansModel.scala
@@ -17,11 +17,19 @@
 
 package org.apache.spark.mllib.clustering
 
+import org.json4s._
+import org.json4s.DefaultFormats
+import org.json4s.jackson.JsonMethods._
+import org.json4s.JsonDSL._
+
+import org.apache.spark.SparkContext
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.internal.Logging
 import org.apache.spark.mllib.linalg.Vector
+import org.apache.spark.mllib.util.{Loader, Saveable}
 import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.{Row, SQLContext}
 
 /**
  * Clustering model produced by [[BisectingKMeans]].
@@ -34,7 +42,7 @@ import org.apache.spark.rdd.RDD
 @Experimental
 class BisectingKMeansModel private[clustering] (
     private[clustering] val root: ClusteringTreeNode
-  ) extends Serializable with Logging {
+  ) extends Serializable with Saveable with Logging {
 
   /**
    * Leaf cluster centers.
@@ -92,4 +100,92 @@ class BisectingKMeansModel private[clustering] (
    */
   @Since("1.6.0")
   def computeCost(data: JavaRDD[Vector]): Double = this.computeCost(data.rdd)
+
+  @Since("2.0.0")
+  override def save(sc: SparkContext, path: String): Unit = {
+    BisectingKMeansModel.SaveLoadV1_0.save(sc, this, path)
+  }
+
+  override protected def formatVersion: String = "1.0"
+}
+
+@Since("2.0.0")
+object BisectingKMeansModel extends Loader[BisectingKMeansModel] {
+
+  @Since("2.0.0")
+  override def load(sc: SparkContext, path: String): BisectingKMeansModel = {
+    val (loadedClassName, formatVersion, metadata) = Loader.loadMetadata(sc, path)
+    implicit val formats = DefaultFormats
+    val rootId = (metadata \ "rootId").extract[Int]
+    val classNameV1_0 = SaveLoadV1_0.thisClassName
+    (loadedClassName, formatVersion) match {
+      case (classNameV1_0, "1.0") =>
+        val model = SaveLoadV1_0.load(sc, path, rootId)
+        model
+      case _ => throw new Exception(
+        s"BisectingKMeansModel.load did not recognize model with (className, format version):" +
+          s"($loadedClassName, $formatVersion).  Supported:\n" +
+          s"  ($classNameV1_0, 1.0)")
+    }
+  }
+
+  private case class Data(index: Int, size: Long, center: Vector, norm: Double, cost: Double,
+     height: Double, children: Seq[Int])
+
+  private object Data {
+    def apply(r: Row): Data = Data(r.getInt(0), r.getLong(1), r.getAs[Vector](2), r.getDouble(3),
+      r.getDouble(4), r.getDouble(5), r.getSeq[Int](6))
+  }
+
+  private[clustering] object SaveLoadV1_0 {
+    private val thisFormatVersion = "1.0"
+
+    private[clustering]
+    val thisClassName = "org.apache.spark.mllib.clustering.BisectingKMeansModel"
+
+    def save(sc: SparkContext, model: BisectingKMeansModel, path: String): Unit = {
+      val sqlContext = SQLContext.getOrCreate(sc)
+      import sqlContext.implicits._
+      val metadata = compact(render(
+        ("class" -> thisClassName) ~ ("version" -> thisFormatVersion)
+          ~ ("rootId" -> model.root.index)))
+      sc.parallelize(Seq(metadata), 1).saveAsTextFile(Loader.metadataPath(path))
+
+      val data = getNodes(model.root).map(node => Data(node.index, node.size,
+        node.centerWithNorm.vector, node.centerWithNorm.norm, node.cost, node.height,
+        node.children.map(_.index)))
+      val dataRDD = sc.parallelize(data).toDF()
+      dataRDD.write.parquet(Loader.dataPath(path))
+    }
+
+    private def getNodes(node: ClusteringTreeNode): Array[ClusteringTreeNode] = {
+      if (node.children.isEmpty) {
+        Array(node)
+      } else {
+        node.children.flatMap(getNodes(_)) ++ Array(node)
+      }
+    }
+
+    def load(sc: SparkContext, path: String, rootId: Int): BisectingKMeansModel = {
+      val sqlContext = SQLContext.getOrCreate(sc)
+      val rows = sqlContext.read.parquet(Loader.dataPath(path))
+      Loader.checkSchema[Data](rows.schema)
+      val data = rows.select("index", "size", "center", "norm", "cost", "height", "children")
+      val nodes = data.rdd.map(Data.apply).collect().map(d => (d.index, d)).toMap
+      val rootNode = buildTree(rootId, nodes)
+      new BisectingKMeansModel(rootNode)
+    }
+
+    private def buildTree(rootId: Int, nodes: Map[Int, Data]): ClusteringTreeNode = {
+      val root = nodes.get(rootId).get
+      if (root.children.isEmpty) {
+        new ClusteringTreeNode(root.index, root.size, new VectorWithNorm(root.center, root.norm),
+          root.cost, root.height, new Array[ClusteringTreeNode](0))
+      } else {
+        val children = root.children.map(c => buildTree(c, nodes))
+        new ClusteringTreeNode(root.index, root.size, new VectorWithNorm(root.center, root.norm),
+          root.cost, root.height, children.toArray)
+      }
+    }
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala
index b719a8c7e7..18f2c994b4 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala
@@ -18,10 +18,12 @@
 package org.apache.spark.ml.clustering
 
 import org.apache.spark.SparkFunSuite
+import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.sql.DataFrame
 
-class BisectingKMeansSuite extends SparkFunSuite with MLlibTestSparkContext {
+class BisectingKMeansSuite
+  extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
   final val k = 5
   @transient var dataset: DataFrame = _
@@ -84,4 +86,22 @@ class BisectingKMeansSuite extends SparkFunSuite with MLlibTestSparkContext {
     assert(model.computeCost(dataset) < 0.1)
     assert(model.hasParent)
   }
+
+  test("read/write") {
+    def checkModelData(model: BisectingKMeansModel, model2: BisectingKMeansModel): Unit = {
+      assert(model.clusterCenters === model2.clusterCenters)
+    }
+    val bisectingKMeans = new BisectingKMeans()
+    testEstimatorAndModelReadWrite(
+      bisectingKMeans, dataset, BisectingKMeansSuite.allParamSettings, checkModelData)
+  }
+}
+
+object BisectingKMeansSuite {
+  val allParamSettings: Map[String, Any] = Map(
+    "k" -> 3,
+    "maxIter" -> 2,
+    "seed" -> -1L,
+    "minDivisibleClusterSize" -> 2.0
+  )
 }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/clustering/BisectingKMeansSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/clustering/BisectingKMeansSuite.scala
index 41b9d5c0d9..35f7932ae8 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/clustering/BisectingKMeansSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/clustering/BisectingKMeansSuite.scala
@@ -21,6 +21,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
+import org.apache.spark.util.Utils
 
 class BisectingKMeansSuite extends SparkFunSuite with MLlibTestSparkContext {
 
@@ -179,4 +180,21 @@ class BisectingKMeansSuite extends SparkFunSuite with MLlibTestSparkContext {
       }
     }
   }
+
+  test("BisectingKMeans model save/load") {
+    val tempDir = Utils.createTempDir()
+    val path = tempDir.toURI.toString
+
+    val points = (1 until 8).map(i => Vectors.dense(i))
+    val data = sc.parallelize(points, 2)
+    val model = new BisectingKMeans().run(data)
+    try {
+      model.save(sc, path)
+      val sameModel = BisectingKMeansModel.load(sc, path)
+      assert(model.k === sameModel.k)
+      model.clusterCenters.zip(sameModel.clusterCenters).foreach(c => c._1 === c._2)
+    } finally {
+      Utils.deleteRecursively(tempDir)
+    }
+  }
 }
-- 
cgit v1.2.3


From 8b207f3b6a0eb617d38091f3b9001830ac3651fe Mon Sep 17 00:00:00 2001
From: Xusen Yin <yinxusen@gmail.com>
Date: Thu, 31 Mar 2016 11:17:32 -0700
Subject: [SPARK-11892][ML] Model export/import for spark.ml: OneVsRest

# What changes were proposed in this pull request?

https://issues.apache.org/jira/browse/SPARK-11892

Add save/load for spark ml.OneVsRest and its model. Also add OneVsRest and OneVsRestModel in MetaAlgorithmReadWrite.

# How was this patch tested?

Test with Scala unit test.

Author: Xusen Yin <yinxusen@gmail.com>

Closes #9934 from yinxusen/SPARK-11892.
---
 .../apache/spark/ml/classification/OneVsRest.scala | 165 +++++++++++++++++++--
 .../scala/org/apache/spark/ml/util/ReadWrite.scala |   8 +-
 .../spark/ml/classification/OneVsRestSuite.scala   |  68 ++++++++-
 3 files changed, 223 insertions(+), 18 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
index c41a611f1c..98b99a3485 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
@@ -21,22 +21,24 @@ import java.util.UUID
 
 import scala.language.existentials
 
+import org.apache.hadoop.fs.Path
+import org.json4s.{DefaultFormats, JObject, _}
+import org.json4s.JsonDSL._
+import org.json4s.jackson.JsonMethods._
+
+import org.apache.spark.SparkContext
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml._
 import org.apache.spark.ml.attribute._
-import org.apache.spark.ml.param.{Param, ParamMap}
-import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
+import org.apache.spark.ml.param.{Param, ParamMap, ParamPair, Params}
+import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.sql.{DataFrame, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 import org.apache.spark.storage.StorageLevel
 
-/**
- * Params for [[OneVsRest]].
- */
-private[ml] trait OneVsRestParams extends PredictorParams {
-
+private[ml] trait ClassifierTypeTrait {
   // scalastyle:off structural.type
   type ClassifierType = Classifier[F, E, M] forSome {
     type F
@@ -44,6 +46,12 @@ private[ml] trait OneVsRestParams extends PredictorParams {
     type E <: Classifier[F, E, M]
   }
   // scalastyle:on structural.type
+}
+
+/**
+ * Params for [[OneVsRest]].
+ */
+private[ml] trait OneVsRestParams extends PredictorParams with ClassifierTypeTrait {
 
   /**
    * param for the base binary classifier that we reduce multiclass classification into.
@@ -57,6 +65,55 @@ private[ml] trait OneVsRestParams extends PredictorParams {
   def getClassifier: ClassifierType = $(classifier)
 }
 
+private[ml] object OneVsRestParams extends ClassifierTypeTrait {
+
+  def validateParams(instance: OneVsRestParams): Unit = {
+    def checkElement(elem: Params, name: String): Unit = elem match {
+      case stage: MLWritable => // good
+      case other =>
+        throw new UnsupportedOperationException("OneVsRest write will fail " +
+          s" because it contains $name which does not implement MLWritable." +
+          s" Non-Writable $name: ${other.uid} of type ${other.getClass}")
+    }
+
+    instance match {
+      case ovrModel: OneVsRestModel => ovrModel.models.foreach(checkElement(_, "model"))
+      case _ => // no need to check OneVsRest here
+    }
+
+    checkElement(instance.getClassifier, "classifier")
+  }
+
+  def saveImpl(
+      path: String,
+      instance: OneVsRestParams,
+      sc: SparkContext,
+      extraMetadata: Option[JObject] = None): Unit = {
+
+    val params = instance.extractParamMap().toSeq
+    val jsonParams = render(params
+      .filter { case ParamPair(p, v) => p.name != "classifier" }
+      .map { case ParamPair(p, v) => p.name -> parse(p.jsonEncode(v)) }
+      .toList)
+
+    DefaultParamsWriter.saveMetadata(instance, path, sc, extraMetadata, Some(jsonParams))
+
+    val classifierPath = new Path(path, "classifier").toString
+    instance.getClassifier.asInstanceOf[MLWritable].save(classifierPath)
+  }
+
+  def loadImpl(
+      path: String,
+      sc: SparkContext,
+      expectedClassName: String): (DefaultParamsReader.Metadata, ClassifierType) = {
+
+    val metadata = DefaultParamsReader.loadMetadata(path, sc, expectedClassName)
+    val classifierPath = new Path(path, "classifier").toString
+    val estimator = DefaultParamsReader.loadParamsInstance[ClassifierType](classifierPath, sc)
+    (metadata, estimator)
+  }
+}
+
 /**
  * :: Experimental ::
  * Model produced by [[OneVsRest]].
@@ -73,10 +130,10 @@ private[ml] trait OneVsRestParams extends PredictorParams {
 @Since("1.4.0")
 @Experimental
 final class OneVsRestModel private[ml] (
-    @Since("1.4.0")  override val uid: String,
-    @Since("1.4.0") labelMetadata: Metadata,
+    @Since("1.4.0") override val uid: String,
+    private[ml] val labelMetadata: Metadata,
     @Since("1.4.0") val models: Array[_ <: ClassificationModel[_, _]])
-  extends Model[OneVsRestModel] with OneVsRestParams {
+  extends Model[OneVsRestModel] with OneVsRestParams with MLWritable {
 
   @Since("1.4.0")
   override def transformSchema(schema: StructType): StructType = {
@@ -143,6 +200,56 @@ final class OneVsRestModel private[ml] (
       uid, labelMetadata, models.map(_.copy(extra).asInstanceOf[ClassificationModel[_, _]]))
     copyValues(copied, extra).setParent(parent)
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new OneVsRestModel.OneVsRestModelWriter(this)
+}
+
+@Since("2.0.0")
+object OneVsRestModel extends MLReadable[OneVsRestModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[OneVsRestModel] = new OneVsRestModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): OneVsRestModel = super.load(path)
+
+  /** [[MLWriter]] instance for [[OneVsRestModel]] */
+  private[OneVsRestModel] class OneVsRestModelWriter(instance: OneVsRestModel) extends MLWriter {
+
+    OneVsRestParams.validateParams(instance)
+
+    override protected def saveImpl(path: String): Unit = {
+      val extraJson = ("labelMetadata" -> instance.labelMetadata.json) ~
+        ("numClasses" -> instance.models.length)
+      OneVsRestParams.saveImpl(path, instance, sc, Some(extraJson))
+      instance.models.zipWithIndex.foreach { case (model: MLWritable, idx) =>
+        val modelPath = new Path(path, s"model_$idx").toString
+        model.save(modelPath)
+      }
+    }
+  }
+
+  private class OneVsRestModelReader extends MLReader[OneVsRestModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[OneVsRestModel].getName
+
+    override def load(path: String): OneVsRestModel = {
+      implicit val format = DefaultFormats
+      val (metadata, classifier) = OneVsRestParams.loadImpl(path, sc, className)
+      val labelMetadata = Metadata.fromJson((metadata.metadata \ "labelMetadata").extract[String])
+      val numClasses = (metadata.metadata \ "numClasses").extract[Int]
+      val models = Range(0, numClasses).toArray.map { idx =>
+        val modelPath = new Path(path, s"model_$idx").toString
+        DefaultParamsReader.loadParamsInstance[ClassificationModel[_, _]](modelPath, sc)
+      }
+      val ovrModel = new OneVsRestModel(metadata.uid, labelMetadata, models)
+      DefaultParamsReader.getAndSetParams(ovrModel, metadata)
+      ovrModel.set("classifier", classifier)
+      ovrModel
+    }
+  }
 }
 
 /**
@@ -158,7 +265,7 @@ final class OneVsRestModel private[ml] (
 @Experimental
 final class OneVsRest @Since("1.4.0") (
     @Since("1.4.0") override val uid: String)
-  extends Estimator[OneVsRestModel] with OneVsRestParams {
+  extends Estimator[OneVsRestModel] with OneVsRestParams with MLWritable {
 
   @Since("1.4.0")
   def this() = this(Identifiable.randomUID("oneVsRest"))
@@ -243,4 +350,40 @@ final class OneVsRest @Since("1.4.0") (
     }
     copied
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new OneVsRest.OneVsRestWriter(this)
+}
+
+@Since("2.0.0")
+object OneVsRest extends MLReadable[OneVsRest] {
+
+  @Since("2.0.0")
+  override def read: MLReader[OneVsRest] = new OneVsRestReader
+
+  @Since("2.0.0")
+  override def load(path: String): OneVsRest = super.load(path)
+
+  /** [[MLWriter]] instance for [[OneVsRest]] */
+  private[OneVsRest] class OneVsRestWriter(instance: OneVsRest) extends MLWriter {
+
+    OneVsRestParams.validateParams(instance)
+
+    override protected def saveImpl(path: String): Unit = {
+      OneVsRestParams.saveImpl(path, instance, sc)
+    }
+  }
+
+  private class OneVsRestReader extends MLReader[OneVsRest] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[OneVsRest].getName
+
+    override def load(path: String): OneVsRest = {
+      val (metadata, classifier) = OneVsRestParams.loadImpl(path, sc, className)
+      val ovr = new OneVsRest(metadata.uid)
+      DefaultParamsReader.getAndSetParams(ovr, metadata)
+      ovr.setClassifier(classifier)
+    }
+  }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala b/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
index 5a596cad06..39999ede30 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
@@ -29,7 +29,7 @@ import org.apache.spark.SparkContext
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml._
-import org.apache.spark.ml.classification.OneVsRestParams
+import org.apache.spark.ml.classification.{OneVsRest, OneVsRestModel}
 import org.apache.spark.ml.feature.RFormulaModel
 import org.apache.spark.ml.param.{ParamPair, Params}
 import org.apache.spark.ml.tuning.ValidatorParams
@@ -381,10 +381,8 @@ private[ml] object MetaAlgorithmReadWrite {
       case p: Pipeline => p.getStages.asInstanceOf[Array[Params]]
       case pm: PipelineModel => pm.stages.asInstanceOf[Array[Params]]
       case v: ValidatorParams => Array(v.getEstimator, v.getEvaluator)
-      case ovr: OneVsRestParams =>
-        // TODO: SPARK-11892: This case may require special handling.
-        throw new UnsupportedOperationException(s"${instance.getClass.getName} write will fail" +
-          s" because it cannot yet handle an estimator containing type: ${ovr.getClass.getName}.")
+      case ovr: OneVsRest => Array(ovr.getClassifier)
+      case ovrModel: OneVsRestModel => Array(ovrModel.getClassifier) ++ ovrModel.models
       case rformModel: RFormulaModel => Array(rformModel.pipelineModel)
       case _: Params => Array()
     }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
index 2ae74a2090..51c1baf682 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
@@ -21,7 +21,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.attribute.NominalAttribute
 import org.apache.spark.ml.feature.StringIndexer
 import org.apache.spark.ml.param.{ParamMap, ParamsSuite}
-import org.apache.spark.ml.util.{MetadataUtils, MLTestingUtils}
+import org.apache.spark.ml.util.{DefaultReadWriteTest, MetadataUtils, MLTestingUtils}
 import org.apache.spark.mllib.classification.LogisticRegressionSuite._
 import org.apache.spark.mllib.classification.LogisticRegressionWithLBFGS
 import org.apache.spark.mllib.evaluation.MulticlassMetrics
@@ -33,7 +33,7 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.types.Metadata
 
-class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext {
+class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
   @transient var dataset: DataFrame = _
   @transient var rdd: RDD[LabeledPoint] = _
@@ -160,6 +160,70 @@ class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext {
       require(m.getThreshold === 0.1, "copy should handle extra model params")
     }
   }
+
+  test("read/write: OneVsRest") {
+    val lr = new LogisticRegression().setMaxIter(10).setRegParam(0.01)
+
+    val ova = new OneVsRest()
+      .setClassifier(lr)
+      .setLabelCol("myLabel")
+      .setFeaturesCol("myFeature")
+      .setPredictionCol("myPrediction")
+
+    val ova2 = testDefaultReadWrite(ova, testParams = false)
+    assert(ova.uid === ova2.uid)
+    assert(ova.getFeaturesCol === ova2.getFeaturesCol)
+    assert(ova.getLabelCol === ova2.getLabelCol)
+    assert(ova.getPredictionCol === ova2.getPredictionCol)
+
+    ova2.getClassifier match {
+      case lr2: LogisticRegression =>
+        assert(lr.uid === lr2.uid)
+        assert(lr.getMaxIter === lr2.getMaxIter)
+        assert(lr.getRegParam === lr2.getRegParam)
+      case other =>
+        throw new AssertionError(s"Loaded OneVsRest expected classifier of type" +
+          s" LogisticRegression but found ${other.getClass.getName}")
+    }
+  }
+
+  test("read/write: OneVsRestModel") {
+    def checkModelData(model: OneVsRestModel, model2: OneVsRestModel): Unit = {
+      assert(model.uid === model2.uid)
+      assert(model.getFeaturesCol === model2.getFeaturesCol)
+      assert(model.getLabelCol === model2.getLabelCol)
+      assert(model.getPredictionCol === model2.getPredictionCol)
+
+      val classifier = model.getClassifier.asInstanceOf[LogisticRegression]
+
+      model2.getClassifier match {
+        case lr2: LogisticRegression =>
+          assert(classifier.uid === lr2.uid)
+          assert(classifier.getMaxIter === lr2.getMaxIter)
+          assert(classifier.getRegParam === lr2.getRegParam)
+        case other =>
+          throw new AssertionError(s"Loaded OneVsRestModel expected classifier of type" +
+            s" LogisticRegression but found ${other.getClass.getName}")
+      }
+
+      assert(model.labelMetadata === model2.labelMetadata)
+      model.models.zip(model2.models).foreach {
+        case (lrModel1: LogisticRegressionModel, lrModel2: LogisticRegressionModel) =>
+          assert(lrModel1.uid === lrModel2.uid)
+          assert(lrModel1.coefficients === lrModel2.coefficients)
+          assert(lrModel1.intercept === lrModel2.intercept)
+        case other =>
+          throw new AssertionError(s"Loaded OneVsRestModel expected model of type" +
+            s" LogisticRegressionModel but found ${other.getClass.getName}")
+      }
+    }
+
+    val lr = new LogisticRegression().setMaxIter(10).setRegParam(0.01)
+    val ova = new OneVsRest().setClassifier(lr)
+    val ovaModel = ova.fit(dataset)
+    val newOvaModel = testDefaultReadWrite(ovaModel, testParams = false)
+    checkModelData(ovaModel, newOvaModel)
+  }
 }
 
 private class MockLogisticRegression(uid: String) extends LogisticRegression(uid) {
-- 
cgit v1.2.3


From 1b070637fa03ab4966f76427b15e433050eaa956 Mon Sep 17 00:00:00 2001
From: Cheng Lian <lian@databricks.com>
Date: Thu, 31 Mar 2016 23:46:08 -0700
Subject: [SPARK-14295][SPARK-14274][SQL] Implements buildReader() for LibSVM

## What changes were proposed in this pull request?

This PR implements `FileFormat.buildReader()` for the LibSVM data source. Besides that, a new interface method `prepareRead()` is added to `FileFormat`:

```scala
  def prepareRead(
      sqlContext: SQLContext,
      options: Map[String, String],
      files: Seq[FileStatus]): Map[String, String] = options
```

After migrating from `buildInternalScan()` to `buildReader()`, we lost the opportunity to collect necessary global information, since `buildReader()` works in a per-partition manner. For example, LibSVM needs to infer the total number of features if the `numFeatures` data source option is not set. Any necessary collected global information should be returned using the data source options map. By default, this method just returns the original options untouched.

An alternative approach is to absorb `inferSchema()` into `prepareRead()`, since schema inference is also some kind of global information gathering. However, this approach wasn't chosen because schema inference is optional, while `prepareRead()` must be called whenever a `HadoopFsRelation` based data source relation is instantiated.

One unaddressed problem is that, when `numFeatures` is absent, now the input data will be scanned twice. The `buildInternalScan()` code path doesn't need to do this because it caches the raw parsed RDD in memory before computing the total number of features. However, with `FileScanRDD`, the raw parsed RDD is created in a different way (e.g. partitioning) from the final RDD.

## How was this patch tested?

Tested using existing test suites.

Author: Cheng Lian <lian@databricks.com>

Closes #12088 from liancheng/spark-14295-libsvm-build-reader.
---
 .../spark/ml/source/libsvm/LibSVMRelation.scala    | 87 +++++++++++++++++++++-
 .../org/apache/spark/mllib/util/MLUtils.scala      | 73 ++++++++++--------
 .../sql/execution/datasources/DataSource.scala     |  5 +-
 .../execution/datasources/FileSourceStrategy.scala |  1 +
 .../org/apache/spark/sql/sources/interfaces.scala  |  9 +++
 5 files changed, 141 insertions(+), 34 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala b/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
index 13a13f0a7e..2e9b6be9a2 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
@@ -19,6 +19,7 @@ package org.apache.spark.ml.source.libsvm
 
 import java.io.IOException
 
+import org.apache.hadoop.conf.Configuration
 import org.apache.hadoop.fs.{FileStatus, Path}
 import org.apache.hadoop.io.{NullWritable, Text}
 import org.apache.hadoop.mapreduce.{Job, RecordWriter, TaskAttemptContext}
@@ -26,12 +27,16 @@ import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat
 
 import org.apache.spark.annotation.Since
 import org.apache.spark.broadcast.Broadcast
-import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
+import org.apache.spark.mllib.linalg.{Vector, Vectors, VectorUDT}
+import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.util.MLUtils
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, DataFrameReader, Row, SQLContext}
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.encoders.RowEncoder
+import org.apache.spark.sql.catalyst.expressions.{AttributeReference, JoinedRow}
+import org.apache.spark.sql.catalyst.expressions.codegen.GenerateUnsafeProjection
+import org.apache.spark.sql.execution.datasources.{CaseInsensitiveMap, HadoopFileLinesReader, PartitionedFile}
 import org.apache.spark.sql.sources._
 import org.apache.spark.sql.types._
 import org.apache.spark.util.SerializableConfiguration
@@ -110,13 +115,16 @@ class DefaultSource extends FileFormat with DataSourceRegister {
   @Since("1.6.0")
   override def shortName(): String = "libsvm"
 
+  override def toString: String = "LibSVM"
+
   private def verifySchema(dataSchema: StructType): Unit = {
     if (dataSchema.size != 2 ||
       (!dataSchema(0).dataType.sameType(DataTypes.DoubleType)
         || !dataSchema(1).dataType.sameType(new VectorUDT()))) {
-      throw new IOException(s"Illegal schema for libsvm data, schema=${dataSchema}")
+      throw new IOException(s"Illegal schema for libsvm data, schema=$dataSchema")
     }
   }
+
   override def inferSchema(
       sqlContext: SQLContext,
       options: Map[String, String],
@@ -127,6 +135,32 @@ class DefaultSource extends FileFormat with DataSourceRegister {
         StructField("features", new VectorUDT(), nullable = false) :: Nil))
   }
 
+  override def prepareRead(
+      sqlContext: SQLContext,
+      options: Map[String, String],
+      files: Seq[FileStatus]): Map[String, String] = {
+    def computeNumFeatures(): Int = {
+      val dataFiles = files.filterNot(_.getPath.getName startsWith "_")
+      val path = if (dataFiles.length == 1) {
+        dataFiles.head.getPath.toUri.toString
+      } else if (dataFiles.isEmpty) {
+        throw new IOException("No input path specified for libsvm data")
+      } else {
+        throw new IOException("Multiple input paths are not supported for libsvm data.")
+      }
+
+      val sc = sqlContext.sparkContext
+      val parsed = MLUtils.parseLibSVMFile(sc, path, sc.defaultParallelism)
+      MLUtils.computeNumFeatures(parsed)
+    }
+
+    val numFeatures = options.get("numFeatures").filter(_.toInt > 0).getOrElse {
+      computeNumFeatures()
+    }
+
+    new CaseInsensitiveMap(options + ("numFeatures" -> numFeatures.toString))
+  }
+
   override def prepareWrite(
       sqlContext: SQLContext,
       job: Job,
@@ -158,7 +192,7 @@ class DefaultSource extends FileFormat with DataSourceRegister {
     verifySchema(dataSchema)
     val dataFiles = inputFiles.filterNot(_.getPath.getName startsWith "_")
 
-    val path = if (dataFiles.length == 1) dataFiles(0).getPath.toUri.toString
+    val path = if (dataFiles.length == 1) dataFiles.head.getPath.toUri.toString
     else if (dataFiles.isEmpty) throw new IOException("No input path specified for libsvm data")
     else throw new IOException("Multiple input paths are not supported for libsvm data.")
 
@@ -176,4 +210,51 @@ class DefaultSource extends FileFormat with DataSourceRegister {
       externalRows.map(converter.toRow)
     }
   }
+
+  override def buildReader(
+      sqlContext: SQLContext,
+      dataSchema: StructType,
+      partitionSchema: StructType,
+      requiredSchema: StructType,
+      filters: Seq[Filter],
+      options: Map[String, String]): (PartitionedFile) => Iterator[InternalRow] = {
+    val numFeatures = options("numFeatures").toInt
+    assert(numFeatures > 0)
+
+    val sparse = options.getOrElse("vectorType", "sparse") == "sparse"
+
+    val broadcastedConf = sqlContext.sparkContext.broadcast(
+      new SerializableConfiguration(new Configuration(sqlContext.sparkContext.hadoopConfiguration))
+    )
+
+    (file: PartitionedFile) => {
+      val points =
+        new HadoopFileLinesReader(file, broadcastedConf.value.value)
+          .map(_.toString.trim)
+          .filterNot(line => line.isEmpty || line.startsWith("#"))
+          .map { line =>
+            val (label, indices, values) = MLUtils.parseLibSVMRecord(line)
+            LabeledPoint(label, Vectors.sparse(numFeatures, indices, values))
+          }
+
+      val converter = RowEncoder(requiredSchema)
+
+      val unsafeRowIterator = points.map { pt =>
+        val features = if (sparse) pt.features.toSparse else pt.features.toDense
+        converter.toRow(Row(pt.label, features))
+      }
+
+      def toAttribute(f: StructField): AttributeReference =
+        AttributeReference(f.name, f.dataType, f.nullable, f.metadata)()
+
+      // Appends partition values
+      val fullOutput = (requiredSchema ++ partitionSchema).map(toAttribute)
+      val joinedRow = new JoinedRow()
+      val appendPartitionColumns = GenerateUnsafeProjection.generate(fullOutput, fullOutput)
+
+      unsafeRowIterator.map { dataRow =>
+        appendPartitionColumns(joinedRow(dataRow, file.partitionValues))
+      }
+    }
+  }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala b/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
index c3b1d5cdd7..4b9d77949f 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
@@ -67,42 +67,14 @@ object MLUtils {
       path: String,
       numFeatures: Int,
       minPartitions: Int): RDD[LabeledPoint] = {
-    val parsed = sc.textFile(path, minPartitions)
-      .map(_.trim)
-      .filter(line => !(line.isEmpty || line.startsWith("#")))
-      .map { line =>
-        val items = line.split(' ')
-        val label = items.head.toDouble
-        val (indices, values) = items.tail.filter(_.nonEmpty).map { item =>
-          val indexAndValue = item.split(':')
-          val index = indexAndValue(0).toInt - 1 // Convert 1-based indices to 0-based.
-          val value = indexAndValue(1).toDouble
-          (index, value)
-        }.unzip
-
-        // check if indices are one-based and in ascending order
-        var previous = -1
-        var i = 0
-        val indicesLength = indices.length
-        while (i < indicesLength) {
-          val current = indices(i)
-          require(current > previous, s"indices should be one-based and in ascending order;"
-            + " found current=$current, previous=$previous; line=\"$line\"")
-          previous = current
-          i += 1
-        }
-
-        (label, indices.toArray, values.toArray)
-      }
+    val parsed = parseLibSVMFile(sc, path, minPartitions)
 
     // Determine number of features.
     val d = if (numFeatures > 0) {
       numFeatures
     } else {
       parsed.persist(StorageLevel.MEMORY_ONLY)
-      parsed.map { case (label, indices, values) =>
-        indices.lastOption.getOrElse(0)
-      }.reduce(math.max) + 1
+      computeNumFeatures(parsed)
     }
 
     parsed.map { case (label, indices, values) =>
@@ -110,6 +82,47 @@ object MLUtils {
     }
   }
 
+  private[spark] def computeNumFeatures(rdd: RDD[(Double, Array[Int], Array[Double])]): Int = {
+    rdd.map { case (label, indices, values) =>
+      indices.lastOption.getOrElse(0)
+    }.reduce(math.max) + 1
+  }
+
+  private[spark] def parseLibSVMFile(
+      sc: SparkContext,
+      path: String,
+      minPartitions: Int): RDD[(Double, Array[Int], Array[Double])] = {
+    sc.textFile(path, minPartitions)
+      .map(_.trim)
+      .filter(line => !(line.isEmpty || line.startsWith("#")))
+      .map(parseLibSVMRecord)
+  }
+
+  private[spark] def parseLibSVMRecord(line: String): (Double, Array[Int], Array[Double]) = {
+    val items = line.split(' ')
+    val label = items.head.toDouble
+    val (indices, values) = items.tail.filter(_.nonEmpty).map { item =>
+      val indexAndValue = item.split(':')
+      val index = indexAndValue(0).toInt - 1 // Convert 1-based indices to 0-based.
+    val value = indexAndValue(1).toDouble
+      (index, value)
+    }.unzip
+
+    // check if indices are one-based and in ascending order
+    var previous = -1
+    var i = 0
+    val indicesLength = indices.length
+    while (i < indicesLength) {
+      val current = indices(i)
+      require(current > previous, s"indices should be one-based and in ascending order;"
+        + " found current=$current, previous=$previous; line=\"$line\"")
+      previous = current
+      i += 1
+    }
+
+    (label, indices, values)
+  }
+
   /**
    * Loads labeled data in the LIBSVM format into an RDD[LabeledPoint], with the default number of
    * partitions.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSource.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSource.scala
index c66921f485..1850810270 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSource.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSource.scala
@@ -299,6 +299,9 @@ case class DataSource(
             "It must be specified manually")
         }
 
+        val enrichedOptions =
+          format.prepareRead(sqlContext, caseInsensitiveOptions, fileCatalog.allFiles())
+
         HadoopFsRelation(
           sqlContext,
           fileCatalog,
@@ -306,7 +309,7 @@ case class DataSource(
           dataSchema = dataSchema.asNullable,
           bucketSpec = bucketSpec,
           format,
-          options)
+          enrichedOptions)
 
       case _ =>
         throw new AnalysisException(
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala
index 554298772a..a143ac6aec 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala
@@ -59,6 +59,7 @@ private[sql] object FileSourceStrategy extends Strategy with Logging {
       if (files.fileFormat.toString == "TestFileFormat" ||
          files.fileFormat.isInstanceOf[parquet.DefaultSource] ||
          files.fileFormat.toString == "ORC" ||
+         files.fileFormat.toString == "LibSVM" ||
          files.fileFormat.isInstanceOf[csv.DefaultSource] ||
          files.fileFormat.isInstanceOf[text.DefaultSource] ||
          files.fileFormat.isInstanceOf[json.DefaultSource]) &&
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
index 6b95a3d25b..e8834d052c 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
@@ -438,6 +438,15 @@ trait FileFormat {
       options: Map[String, String],
       files: Seq[FileStatus]): Option[StructType]
 
+  /**
+   * Prepares a read job and returns a potentially updated data source option [[Map]]. This method
+   * can be useful for collecting necessary global information for scanning input data.
+   */
+  def prepareRead(
+      sqlContext: SQLContext,
+      options: Map[String, String],
+      files: Seq[FileStatus]): Map[String, String] = options
+
   /**
    * Prepares a write job and returns an [[OutputWriterFactory]].  Client side job preparation can
    * be put here.  For example, user defined output committer can be configured here
-- 
cgit v1.2.3


From 26867ebc67edab97376c5d8fee76df294359e461 Mon Sep 17 00:00:00 2001
From: Alexander Ulanov <nashb@yandex.ru>
Date: Thu, 31 Mar 2016 23:48:36 -0700
Subject: [SPARK-11262][ML] Unit test for gradient, loss layers, memory
 management for multilayer perceptron

1.Implement LossFunction trait and implement squared error and cross entropy
loss with it
2.Implement unit test for gradient and loss
3.Implement InPlace trait and in-place layer evaluation
4.Refactor interface for ActivationFunction
5.Update of Layer and LayerModel interfaces
6.Fix random weights assignment
7.Implement memory allocation by MLP model instead of individual layers

These features decreased the memory usage and increased flexibility of
internal API.

Author: Alexander Ulanov <nashb@yandex.ru>
Author: avulanov <avulanov@gmail.com>

Closes #9229 from avulanov/mlp-refactoring.
---
 .../main/scala/org/apache/spark/ml/ann/Layer.scala | 662 ++++++++++-----------
 .../org/apache/spark/ml/ann/LossFunction.scala     | 124 ++++
 .../MultilayerPerceptronClassifier.scala           |  82 ++-
 .../JavaMultilayerPerceptronClassifierSuite.java   |   2 +-
 .../scala/org/apache/spark/ml/ann/ANNSuite.scala   |   9 +-
 .../org/apache/spark/ml/ann/GradientSuite.scala    |  76 +++
 .../MultilayerPerceptronClassifierSuite.scala      |  26 +-
 project/MimaExcludes.scala                         |   5 +
 python/pyspark/ml/classification.py                |   2 +-
 9 files changed, 601 insertions(+), 387 deletions(-)
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/ann/LossFunction.scala
 create mode 100644 mllib/src/test/scala/org/apache/spark/ml/ann/GradientSuite.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala b/mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala
index 2cd94fa8f5..a5b84116e6 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala
@@ -17,9 +17,9 @@
 
 package org.apache.spark.ml.ann
 
-import breeze.linalg.{*, axpy => Baxpy, sum => Bsum, DenseMatrix => BDM, DenseVector => BDV,
-  Vector => BV}
-import breeze.numerics.{log => Blog, sigmoid => Bsigmoid}
+import java.util.Random
+
+import breeze.linalg.{*, axpy => Baxpy, DenseMatrix => BDM, DenseVector => BDV, Vector => BV}
 
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.optimization._
@@ -32,20 +32,46 @@ import org.apache.spark.util.random.XORShiftRandom
  *
  */
 private[ann] trait Layer extends Serializable {
+
   /**
-   * Returns the instance of the layer based on weights provided
-   * @param weights vector with layer weights
-   * @param position position of weights in the vector
-   * @return the layer model
+   * Number of weights that is used to allocate memory for the weights vector
+   */
+  val weightSize: Int
+
+  /**
+   * Returns the output size given the input size (not counting the stack size).
+   * Output size is used to allocate memory for the output.
+   *
+   * @param inputSize input size
+   * @return output size
    */
-  def getInstance(weights: Vector, position: Int): LayerModel
+  def getOutputSize(inputSize: Int): Int
 
+  /**
+   * If true, the memory is not allocated for the output of this layer.
+   * The memory allocated to the previous layer is used to write the output of this layer.
+   * Developer can set this to true if computing delta of a previous layer
+   * does not involve its output, so the current layer can write there.
+   * This also mean that both layers have the same number of outputs.
+   */
+  val inPlace: Boolean
+
+  /**
+   * Returns the instance of the layer based on weights provided.
+   * Size of weights must be equal to weightSize
+   *
+   * @param initialWeights vector with layer weights
+   * @return the layer model
+   */
+  def createModel(initialWeights: BDV[Double]): LayerModel
   /**
    * Returns the instance of the layer with random generated weights
-   * @param seed seed
+   *
+   * @param weights vector for weights initialization, must be equal to weightSize
+   * @param random random number generator
    * @return the layer model
    */
-  def getInstance(seed: Long): LayerModel
+  def initModel(weights: BDV[Double], random: Random): LayerModel
 }
 
 /**
@@ -54,92 +80,102 @@ private[ann] trait Layer extends Serializable {
  * Can return weights in Vector format.
  */
 private[ann] trait LayerModel extends Serializable {
-  /**
-   * number of weights
-   */
-  val size: Int
 
+  val weights: BDV[Double]
   /**
    * Evaluates the data (process the data through the layer)
+   * Output is allocated based on the size provided by the
+   * LayerModel implementation and the stack (batch) size
+   * Developer is responsible for checking the size of output
+   * when writing to it
+ *
    * @param data data
-   * @return processed data
+   * @param output output (modified in place)
    */
-  def eval(data: BDM[Double]): BDM[Double]
+  def eval(data: BDM[Double], output: BDM[Double]): Unit
 
   /**
    * Computes the delta for back propagation
-   * @param nextDelta delta of the next layer
-   * @param input input data
-   * @return delta
+   * Delta is allocated based on the size provided by the
+   * LayerModel implementation and the stack (batch) size
+   * Developer is responsible for checking the size of
+   * prevDelta when writing to it
+    *
+    * @param delta delta of this layer
+   * @param output output of this layer
+   * @param prevDelta the previous delta (modified in place)
    */
-  def prevDelta(nextDelta: BDM[Double], input: BDM[Double]): BDM[Double]
+  def computePrevDelta(delta: BDM[Double], output: BDM[Double], prevDelta: BDM[Double]): Unit
 
   /**
    * Computes the gradient
+   * cumGrad is a wrapper on the part of the weight vector
+   * size of cumGrad is based on weightSize provided by
+   * implementation of LayerModel
+   *
    * @param delta delta for this layer
    * @param input input data
-   * @return gradient
+   * @param cumGrad cumulative gradient (modified in place)
    */
-  def grad(delta: BDM[Double], input: BDM[Double]): Array[Double]
-
-  /**
-   * Returns weights for the layer in a single vector
-   * @return layer weights
-   */
-  def weights(): Vector
+  def grad(delta: BDM[Double], input: BDM[Double], cumGrad: BDV[Double]): Unit
 }
 
 /**
  * Layer properties of affine transformations, that is y=A*x+b
+ *
  * @param numIn number of inputs
  * @param numOut number of outputs
  */
 private[ann] class AffineLayer(val numIn: Int, val numOut: Int) extends Layer {
 
-  override def getInstance(weights: Vector, position: Int): LayerModel = {
-    AffineLayerModel(this, weights, position)
-  }
+  override val weightSize = numIn * numOut + numOut
 
-  override def getInstance(seed: Long = 11L): LayerModel = {
-    AffineLayerModel(this, seed)
-  }
+  override def getOutputSize(inputSize: Int): Int = numOut
+
+  override val inPlace = false
+
+  override def createModel(weights: BDV[Double]): LayerModel = new AffineLayerModel(weights, this)
+
+  override def initModel(weights: BDV[Double], random: Random): LayerModel =
+    AffineLayerModel(this, weights, random)
 }
 
 /**
- * Model of Affine layer y=A*x+b
- * @param w weights (matrix A)
- * @param b bias (vector b)
+ * Model of Affine layer
+ *
+ * @param weights weights
+ * @param layer layer properties
  */
-private[ann] class AffineLayerModel private(w: BDM[Double], b: BDV[Double]) extends LayerModel {
-  val size = w.size + b.length
-  val gwb = new Array[Double](size)
-  private lazy val gw: BDM[Double] = new BDM[Double](w.rows, w.cols, gwb)
-  private lazy val gb: BDV[Double] = new BDV[Double](gwb, w.size)
-  private var z: BDM[Double] = null
-  private var d: BDM[Double] = null
+private[ann] class AffineLayerModel private[ann] (
+    val weights: BDV[Double],
+    val layer: AffineLayer) extends LayerModel {
+  val w = new BDM[Double](layer.numOut, layer.numIn, weights.data, weights.offset)
+  val b =
+    new BDV[Double](weights.data, weights.offset + (layer.numOut * layer.numIn), 1, layer.numOut)
+
   private var ones: BDV[Double] = null
 
-  override def eval(data: BDM[Double]): BDM[Double] = {
-    if (z == null || z.cols != data.cols) z = new BDM[Double](w.rows, data.cols)
-    z(::, *) := b
-    BreezeUtil.dgemm(1.0, w, data, 1.0, z)
-    z
+  override def eval(data: BDM[Double], output: BDM[Double]): Unit = {
+    output(::, *) := b
+    BreezeUtil.dgemm(1.0, w, data, 1.0, output)
   }
 
-  override def prevDelta(nextDelta: BDM[Double], input: BDM[Double]): BDM[Double] = {
-    if (d == null || d.cols != nextDelta.cols) d = new BDM[Double](w.cols, nextDelta.cols)
-    BreezeUtil.dgemm(1.0, w.t, nextDelta, 0.0, d)
-    d
+  override def computePrevDelta(
+    delta: BDM[Double],
+    output: BDM[Double],
+    prevDelta: BDM[Double]): Unit = {
+    BreezeUtil.dgemm(1.0, w.t, delta, 0.0, prevDelta)
   }
 
-  override def grad(delta: BDM[Double], input: BDM[Double]): Array[Double] = {
-    BreezeUtil.dgemm(1.0 / input.cols, delta, input.t, 0.0, gw)
+  override def grad(delta: BDM[Double], input: BDM[Double], cumGrad: BDV[Double]): Unit = {
+    // compute gradient of weights
+    val cumGradientOfWeights = new BDM[Double](w.rows, w.cols, cumGrad.data, cumGrad.offset)
+    BreezeUtil.dgemm(1.0 / input.cols, delta, input.t, 1.0, cumGradientOfWeights)
     if (ones == null || ones.length != delta.cols) ones = BDV.ones[Double](delta.cols)
-    BreezeUtil.dgemv(1.0 / input.cols, delta, ones, 0.0, gb)
-    gwb
+    // compute gradient of bias
+    val cumGradientOfBias = new BDV[Double](cumGrad.data, cumGrad.offset + w.size, 1, b.length)
+    BreezeUtil.dgemv(1.0 / input.cols, delta, ones, 1.0, cumGradientOfBias)
   }
-
-  override def weights(): Vector = AffineLayerModel.roll(w, b)
 }
 
 /**
@@ -149,73 +185,40 @@ private[ann] object AffineLayerModel {
 
   /**
    * Creates a model of Affine layer
+ *
    * @param layer layer properties
-   * @param weights vector with weights
-   * @param position position of weights in the vector
-   * @return model of Affine layer
-   */
-  def apply(layer: AffineLayer, weights: Vector, position: Int): AffineLayerModel = {
-    val (w, b) = unroll(weights, position, layer.numIn, layer.numOut)
-    new AffineLayerModel(w, b)
-  }
-
-  /**
-   * Creates a model of Affine layer
-   * @param layer layer properties
-   * @param seed seed
+   * @param weights vector for weights initialization
+   * @param random random number generator
    * @return model of Affine layer
    */
-  def apply(layer: AffineLayer, seed: Long): AffineLayerModel = {
-    val (w, b) = randomWeights(layer.numIn, layer.numOut, seed)
-    new AffineLayerModel(w, b)
-  }
-
-  /**
-   * Unrolls the weights from the vector
-   * @param weights vector with weights
-   * @param position position of weights for this layer
-   * @param numIn number of layer inputs
-   * @param numOut number of layer outputs
-   * @return matrix A and vector b
-   */
-  def unroll(
-    weights: Vector,
-    position: Int,
-    numIn: Int,
-    numOut: Int): (BDM[Double], BDV[Double]) = {
-    val weightsCopy = weights.toArray
-    // TODO: the array is not copied to BDMs, make sure this is OK!
-    val a = new BDM[Double](numOut, numIn, weightsCopy, position)
-    val b = new BDV[Double](weightsCopy, position + (numOut * numIn), 1, numOut)
-    (a, b)
-  }
-
-  /**
-   * Roll the layer weights into a vector
-   * @param a matrix A
-   * @param b vector b
-   * @return vector of weights
-   */
-  def roll(a: BDM[Double], b: BDV[Double]): Vector = {
-    val result = new Array[Double](a.size + b.length)
-    // TODO: make sure that we need to copy!
-    System.arraycopy(a.toArray, 0, result, 0, a.size)
-    System.arraycopy(b.toArray, 0, result, a.size, b.length)
-    Vectors.dense(result)
+  def apply(layer: AffineLayer, weights: BDV[Double], random: Random): AffineLayerModel = {
+    randomWeights(layer.numIn, layer.numOut, weights, random)
+    new AffineLayerModel(weights, layer)
   }
 
   /**
-   * Generate random weights for the layer
-   * @param numIn number of inputs
+   * Initialize weights randomly in the interval
+   * Uses [Bottou-88] heuristic [-a/sqrt(in); a/sqrt(in)]
+   * where a is chosen in a such way that the weight variance corresponds
+   * to the points to the maximal curvature of the activation function
+   * (which is approximately 2.38 for a standard sigmoid)
+    *
+    * @param numIn number of inputs
    * @param numOut number of outputs
-   * @param seed seed
-   * @return (matrix A, vector b)
+   * @param weights vector for weights initialization
+   * @param random random number generator
    */
-  def randomWeights(numIn: Int, numOut: Int, seed: Long = 11L): (BDM[Double], BDV[Double]) = {
-    val rand: XORShiftRandom = new XORShiftRandom(seed)
-    val weights = BDM.fill[Double](numOut, numIn) { (rand.nextDouble * 4.8 - 2.4) / numIn }
-    val bias = BDV.fill[Double](numOut) { (rand.nextDouble * 4.8 - 2.4) / numIn }
-    (weights, bias)
+  def randomWeights(
+    numIn: Int,
+    numOut: Int,
+    weights: BDV[Double],
+    random: Random): Unit = {
+    var i = 0
+    val sqrtIn = math.sqrt(numIn)
+    while (i < weights.length) {
+      weights(i) = (random.nextDouble * 4.8 - 2.4) / sqrtIn
+      i += 1
+    }
   }
 }
 
@@ -226,44 +229,21 @@ private[ann] trait ActivationFunction extends Serializable {
 
   /**
    * Implements a function
-   * @param x input data
-   * @param y output data
    */
-  def eval(x: BDM[Double], y: BDM[Double]): Unit
+  def eval: Double => Double
 
   /**
    * Implements a derivative of a function (needed for the back propagation)
-   * @param x input data
-   * @param y output data
    */
-  def derivative(x: BDM[Double], y: BDM[Double]): Unit
-
-  /**
-   * Implements a cross entropy error of a function.
-   * Needed if the functional layer that contains this function is the output layer
-   * of the network.
-   * @param target target output
-   * @param output computed output
-   * @param result intermediate result
-   * @return cross-entropy
-   */
-  def crossEntropy(target: BDM[Double], output: BDM[Double], result: BDM[Double]): Double
-
-  /**
-   * Implements a mean squared error of a function
-   * @param target target output
-   * @param output computed output
-   * @param result intermediate result
-   * @return mean squared error
-   */
-  def squared(target: BDM[Double], output: BDM[Double], result: BDM[Double]): Double
+  def derivative: Double => Double
 }
 
 /**
- * Implements in-place application of functions
+ * Implements in-place application of functions in the arrays
  */
-private[ann] object ActivationFunction {
+private[ann] object ApplyInPlace {
 
+  // TODO: use Breeze UFunc
   def apply(x: BDM[Double], y: BDM[Double], func: Double => Double): Unit = {
     var i = 0
     while (i < x.rows) {
@@ -276,6 +256,7 @@ private[ann] object ActivationFunction {
     }
   }
 
+  // TODO: use Breeze UFunc
   def apply(
     x1: BDM[Double],
     x2: BDM[Double],
@@ -293,180 +274,87 @@ private[ann] object ActivationFunction {
   }
 }
 
-/**
- * Implements SoftMax activation function
- */
-private[ann] class SoftmaxFunction extends ActivationFunction {
-  override def eval(x: BDM[Double], y: BDM[Double]): Unit = {
-    var j = 0
-    // find max value to make sure later that exponent is computable
-    while (j < x.cols) {
-      var i = 0
-      var max = Double.MinValue
-      while (i < x.rows) {
-        if (x(i, j) > max) {
-          max = x(i, j)
-        }
-        i += 1
-      }
-      var sum = 0.0
-      i = 0
-      while (i < x.rows) {
-        val res = Math.exp(x(i, j) - max)
-        y(i, j) = res
-        sum += res
-        i += 1
-      }
-      i = 0
-      while (i < x.rows) {
-        y(i, j) /= sum
-        i += 1
-      }
-      j += 1
-    }
-  }
-
-  override def crossEntropy(
-    output: BDM[Double],
-    target: BDM[Double],
-    result: BDM[Double]): Double = {
-    def m(o: Double, t: Double): Double = o - t
-    ActivationFunction(output, target, result, m)
-    -Bsum( target :* Blog(output)) / output.cols
-  }
-
-  override def derivative(x: BDM[Double], y: BDM[Double]): Unit = {
-    def sd(z: Double): Double = (1 - z) * z
-    ActivationFunction(x, y, sd)
-  }
-
-  override def squared(output: BDM[Double], target: BDM[Double], result: BDM[Double]): Double = {
-    throw new UnsupportedOperationException("Sorry, squared error is not defined for SoftMax.")
-  }
-}
-
 /**
  * Implements Sigmoid activation function
  */
 private[ann] class SigmoidFunction extends ActivationFunction {
-  override def eval(x: BDM[Double], y: BDM[Double]): Unit = {
-    def s(z: Double): Double = Bsigmoid(z)
-    ActivationFunction(x, y, s)
-  }
-
-  override def crossEntropy(
-    output: BDM[Double],
-    target: BDM[Double],
-    result: BDM[Double]): Double = {
-    def m(o: Double, t: Double): Double = o - t
-    ActivationFunction(output, target, result, m)
-    -Bsum(target :* Blog(output)) / output.cols
-  }
 
-  override def derivative(x: BDM[Double], y: BDM[Double]): Unit = {
-    def sd(z: Double): Double = (1 - z) * z
-    ActivationFunction(x, y, sd)
-  }
+  override def eval: (Double) => Double = x => 1.0 / (1 + math.exp(-x))
 
-  override def squared(output: BDM[Double], target: BDM[Double], result: BDM[Double]): Double = {
-    // TODO: make it readable
-    def m(o: Double, t: Double): Double = (o - t)
-    ActivationFunction(output, target, result, m)
-    val e = Bsum(result :* result) / 2 / output.cols
-    def m2(x: Double, o: Double) = x * (o - o * o)
-    ActivationFunction(result, output, result, m2)
-    e
-  }
+  override def derivative: (Double) => Double = z => (1 - z) * z
 }
 
 /**
  * Functional layer properties, y = f(x)
+ *
  * @param activationFunction activation function
  */
 private[ann] class FunctionalLayer (val activationFunction: ActivationFunction) extends Layer {
-  override def getInstance(weights: Vector, position: Int): LayerModel = getInstance(0L)
 
-  override def getInstance(seed: Long): LayerModel =
-    FunctionalLayerModel(this)
+  override val weightSize = 0
+
+  override def getOutputSize(inputSize: Int): Int = inputSize
+
+  override val inPlace = true
+
+  override def createModel(weights: BDV[Double]): LayerModel = new FunctionalLayerModel(this)
+
+  override def initModel(weights: BDV[Double], random: Random): LayerModel =
+    createModel(weights)
 }
 
 /**
  * Functional layer model. Holds no weights.
- * @param activationFunction activation function
+ *
+ * @param layer functiona layer
  */
-private[ann] class FunctionalLayerModel private (val activationFunction: ActivationFunction)
+private[ann] class FunctionalLayerModel private[ann] (val layer: FunctionalLayer)
   extends LayerModel {
-  val size = 0
-  // matrices for in-place computations
-  // outputs
-  private var f: BDM[Double] = null
-  // delta
-  private var d: BDM[Double] = null
-  // matrix for error computation
-  private var e: BDM[Double] = null
-  // delta gradient
-  private lazy val dg = new Array[Double](0)
 
-  override def eval(data: BDM[Double]): BDM[Double] = {
-    if (f == null || f.cols != data.cols) f = new BDM[Double](data.rows, data.cols)
-    activationFunction.eval(data, f)
-    f
-  }
+  // empty weights
+  val weights = new BDV[Double](0)
 
-  override def prevDelta(nextDelta: BDM[Double], input: BDM[Double]): BDM[Double] = {
-    if (d == null || d.cols != nextDelta.cols) d = new BDM[Double](nextDelta.rows, nextDelta.cols)
-    activationFunction.derivative(input, d)
-    d :*= nextDelta
-    d
+  override def eval(data: BDM[Double], output: BDM[Double]): Unit = {
+    ApplyInPlace(data, output, layer.activationFunction.eval)
   }
 
-  override def grad(delta: BDM[Double], input: BDM[Double]): Array[Double] = dg
-
-  override def weights(): Vector = Vectors.dense(new Array[Double](0))
-
-  def crossEntropy(output: BDM[Double], target: BDM[Double]): (BDM[Double], Double) = {
-    if (e == null || e.cols != output.cols) e = new BDM[Double](output.rows, output.cols)
-    val error = activationFunction.crossEntropy(output, target, e)
-    (e, error)
+  override def computePrevDelta(
+    nextDelta: BDM[Double],
+    input: BDM[Double],
+    delta: BDM[Double]): Unit = {
+    ApplyInPlace(input, delta, layer.activationFunction.derivative)
+    delta :*= nextDelta
   }
 
-  def squared(output: BDM[Double], target: BDM[Double]): (BDM[Double], Double) = {
-    if (e == null || e.cols != output.cols) e = new BDM[Double](output.rows, output.cols)
-    val error = activationFunction.squared(output, target, e)
-    (e, error)
-  }
-
-  def error(output: BDM[Double], target: BDM[Double]): (BDM[Double], Double) = {
-    // TODO: allow user pick error
-    activationFunction match {
-      case sigmoid: SigmoidFunction => squared(output, target)
-      case softmax: SoftmaxFunction => crossEntropy(output, target)
-    }
-  }
-}
-
-/**
- * Fabric of functional layer models
- */
-private[ann] object FunctionalLayerModel {
-  def apply(layer: FunctionalLayer): FunctionalLayerModel =
-    new FunctionalLayerModel(layer.activationFunction)
+  override def grad(delta: BDM[Double], input: BDM[Double], cumGrad: BDV[Double]): Unit = {}
 }
 
 /**
  * Trait for the artificial neural network (ANN) topology properties
  */
-private[ann] trait Topology extends Serializable{
-  def getInstance(weights: Vector): TopologyModel
-  def getInstance(seed: Long): TopologyModel
+private[ann] trait Topology extends Serializable {
+  def model(weights: Vector): TopologyModel
+  def model(seed: Long): TopologyModel
 }
 
 /**
  * Trait for ANN topology model
  */
-private[ann] trait TopologyModel extends Serializable{
+private[ann] trait TopologyModel extends Serializable {
+
+  val weights: Vector
+  /**
+   * Array of layers
+   */
+  val layers: Array[Layer]
+
+  /**
+   * Array of layer models
+   */
+  val layerModels: Array[LayerModel]
   /**
    * Forward propagation
+ *
    * @param data input data
    * @return array of outputs for each of the layers
    */
@@ -474,6 +362,7 @@ private[ann] trait TopologyModel extends Serializable{
 
   /**
    * Prediction of the model
+ *
    * @param data input data
    * @return prediction
    */
@@ -481,6 +370,7 @@ private[ann] trait TopologyModel extends Serializable{
 
   /**
    * Computes gradient for the network
+ *
    * @param data input data
    * @param target target output
    * @param cumGradient cumulative gradient
@@ -489,22 +379,17 @@ private[ann] trait TopologyModel extends Serializable{
    */
   def computeGradient(data: BDM[Double], target: BDM[Double], cumGradient: Vector,
                       blockSize: Int): Double
-
-  /**
-   * Returns the weights of the ANN
-   * @return weights
-   */
-  def weights(): Vector
 }
 
 /**
  * Feed forward ANN
+ *
  * @param layers
  */
 private[ann] class FeedForwardTopology private(val layers: Array[Layer]) extends Topology {
-  override def getInstance(weights: Vector): TopologyModel = FeedForwardModel(this, weights)
+  override def model(weights: Vector): TopologyModel = FeedForwardModel(this, weights)
 
-  override def getInstance(seed: Long): TopologyModel = FeedForwardModel(this, seed)
+  override def model(seed: Long): TopologyModel = FeedForwardModel(this, seed)
 }
 
 /**
@@ -513,6 +398,7 @@ private[ann] class FeedForwardTopology private(val layers: Array[Layer]) extends
 private[ml] object FeedForwardTopology {
   /**
    * Creates a feed forward topology from the array of layers
+ *
    * @param layers array of layers
    * @return feed forward topology
    */
@@ -522,18 +408,26 @@ private[ml] object FeedForwardTopology {
 
   /**
    * Creates a multi-layer perceptron
+ *
    * @param layerSizes sizes of layers including input and output size
-   * @param softmax whether to use SoftMax or Sigmoid function for an output layer.
+   * @param softmaxOnTop wether to use SoftMax or Sigmoid function for an output layer.
    *                Softmax is default
    * @return multilayer perceptron topology
    */
-  def multiLayerPerceptron(layerSizes: Array[Int], softmax: Boolean = true): FeedForwardTopology = {
+  def multiLayerPerceptron(
+    layerSizes: Array[Int],
+    softmaxOnTop: Boolean = true): FeedForwardTopology = {
     val layers = new Array[Layer]((layerSizes.length - 1) * 2)
-    for(i <- 0 until layerSizes.length - 1) {
+    for (i <- 0 until layerSizes.length - 1) {
       layers(i * 2) = new AffineLayer(layerSizes(i), layerSizes(i + 1))
       layers(i * 2 + 1) =
-        if (softmax && i == layerSizes.length - 2) {
-          new FunctionalLayer(new SoftmaxFunction())
+        if (i == layerSizes.length - 2) {
+          if (softmaxOnTop) {
+            new SoftmaxLayerWithCrossEntropyLoss()
+          } else {
+            // TODO: squared error is more natural but converges slower
+            new SigmoidLayerWithSquaredError()
+          }
         } else {
           new FunctionalLayer(new SigmoidFunction())
         }
@@ -545,17 +439,45 @@ private[ml] object FeedForwardTopology {
 /**
  * Model of Feed Forward Neural Network.
  * Implements forward, gradient computation and can return weights in vector format.
- * @param layerModels models of layers
- * @param topology topology of the network
+ *
+ * @param weights network weights
+ * @param topology network topology
  */
 private[ml] class FeedForwardModel private(
-    val layerModels: Array[LayerModel],
+    val weights: Vector,
     val topology: FeedForwardTopology) extends TopologyModel {
+
+  val layers = topology.layers
+  val layerModels = new Array[LayerModel](layers.length)
+  private var offset = 0
+  for (i <- 0 until layers.length) {
+    layerModels(i) = layers(i).createModel(
+      new BDV[Double](weights.toArray, offset, 1, layers(i).weightSize))
+    offset += layers(i).weightSize
+  }
+  private var outputs: Array[BDM[Double]] = null
+  private var deltas: Array[BDM[Double]] = null
+
   override def forward(data: BDM[Double]): Array[BDM[Double]] = {
-    val outputs = new Array[BDM[Double]](layerModels.length)
-    outputs(0) = layerModels(0).eval(data)
+    // Initialize output arrays for all layers. Special treatment for InPlace
+    val currentBatchSize = data.cols
+    // TODO: allocate outputs as one big array and then create BDMs from it
+    if (outputs == null || outputs(0).cols != currentBatchSize) {
+      outputs = new Array[BDM[Double]](layers.length)
+      var inputSize = data.rows
+      for (i <- 0 until layers.length) {
+        if (layers(i).inPlace) {
+          outputs(i) = outputs(i - 1)
+        } else {
+          val outputSize = layers(i).getOutputSize(inputSize)
+          outputs(i) = new BDM[Double](outputSize, currentBatchSize)
+          inputSize = outputSize
+        }
+      }
+    }
+    layerModels(0).eval(data, outputs(0))
     for (i <- 1 until layerModels.length) {
-      outputs(i) = layerModels(i).eval(outputs(i-1))
+      layerModels(i).eval(outputs(i - 1), outputs(i))
     }
     outputs
   }
@@ -566,54 +488,36 @@ private[ml] class FeedForwardModel private(
     cumGradient: Vector,
     realBatchSize: Int): Double = {
     val outputs = forward(data)
-    val deltas = new Array[BDM[Double]](layerModels.length)
+    val currentBatchSize = data.cols
+    // TODO: allocate deltas as one big array and then create BDMs from it
+    if (deltas == null || deltas(0).cols != currentBatchSize) {
+      deltas = new Array[BDM[Double]](layerModels.length)
+      var inputSize = data.rows
+      for (i <- 0 until layerModels.length - 1) {
+        val outputSize = layers(i).getOutputSize(inputSize)
+        deltas(i) = new BDM[Double](outputSize, currentBatchSize)
+        inputSize = outputSize
+      }
+    }
     val L = layerModels.length - 1
-    val (newE, newError) = layerModels.last match {
-      case flm: FunctionalLayerModel => flm.error(outputs.last, target)
+    // TODO: explain why delta of top layer is null (because it might contain loss+layer)
+    val loss = layerModels.last match {
+      case levelWithError: LossFunction => levelWithError.loss(outputs.last, target, deltas(L - 1))
       case _ =>
-        throw new UnsupportedOperationException("Non-functional layer not supported at the top")
+        throw new UnsupportedOperationException("Top layer is required to have objective.")
     }
-    deltas(L) = new BDM[Double](0, 0)
-    deltas(L - 1) = newE
     for (i <- (L - 2) to (0, -1)) {
-      deltas(i) = layerModels(i + 1).prevDelta(deltas(i + 1), outputs(i + 1))
-    }
-    val grads = new Array[Array[Double]](layerModels.length)
-    for (i <- 0 until layerModels.length) {
-      val input = if (i==0) data else outputs(i - 1)
-      grads(i) = layerModels(i).grad(deltas(i), input)
+      layerModels(i + 1).computePrevDelta(deltas(i + 1), outputs(i + 1), deltas(i))
     }
-    // update cumGradient
     val cumGradientArray = cumGradient.toArray
     var offset = 0
-    // TODO: extract roll
-    for (i <- 0 until grads.length) {
-      val gradArray = grads(i)
-      var k = 0
-      while (k < gradArray.length) {
-        cumGradientArray(offset + k) += gradArray(k)
-        k += 1
-      }
-      offset += gradArray.length
-    }
-    newError
-  }
-
-  // TODO: do we really need to copy the weights? they should be read-only
-  override def weights(): Vector = {
-    // TODO: extract roll
-    var size = 0
-    for (i <- 0 until layerModels.length) {
-      size += layerModels(i).size
-    }
-    val array = new Array[Double](size)
-    var offset = 0
     for (i <- 0 until layerModels.length) {
-      val layerWeights = layerModels(i).weights().toArray
-      System.arraycopy(layerWeights, 0, array, offset, layerWeights.length)
-      offset += layerWeights.length
+      val input = if (i == 0) data else outputs(i - 1)
+      layerModels(i).grad(deltas(i), input,
+        new BDV[Double](cumGradientArray, offset, 1, layers(i).weightSize))
+      offset += layers(i).weightSize
     }
-    Vectors.dense(array)
+    loss
   }
 
   override def predict(data: Vector): Vector = {
@@ -630,23 +534,19 @@ private[ann] object FeedForwardModel {
 
   /**
    * Creates a model from a topology and weights
+ *
    * @param topology topology
    * @param weights weights
    * @return model
    */
   def apply(topology: FeedForwardTopology, weights: Vector): FeedForwardModel = {
-    val layers = topology.layers
-    val layerModels = new Array[LayerModel](layers.length)
-    var offset = 0
-    for (i <- 0 until layers.length) {
-      layerModels(i) = layers(i).getInstance(weights, offset)
-      offset += layerModels(i).size
-    }
-    new FeedForwardModel(layerModels, topology)
+    // TODO: check that weights size is equal to sum of layers sizes
+    new FeedForwardModel(weights, topology)
   }
 
   /**
    * Creates a model given a topology and seed
+ *
    * @param topology topology
    * @param seed seed for generating the weights
    * @return model
@@ -654,17 +554,25 @@ private[ann] object FeedForwardModel {
   def apply(topology: FeedForwardTopology, seed: Long = 11L): FeedForwardModel = {
     val layers = topology.layers
     val layerModels = new Array[LayerModel](layers.length)
+    var totalSize = 0
+    for (i <- 0 until topology.layers.length) {
+      totalSize += topology.layers(i).weightSize
+    }
+    val weights = BDV.zeros[Double](totalSize)
     var offset = 0
-    for(i <- 0 until layers.length) {
-      layerModels(i) = layers(i).getInstance(seed)
-      offset += layerModels(i).size
+    val random = new XORShiftRandom(seed)
+    for (i <- 0 until layers.length) {
+      layerModels(i) = layers(i).
+        initModel(new BDV[Double](weights.data, offset, 1, layers(i).weightSize), random)
+      offset += layers(i).weightSize
     }
-    new FeedForwardModel(layerModels, topology)
+    new FeedForwardModel(Vectors.fromBreeze(weights), topology)
   }
 }
 
 /**
  * Neural network gradient. Does nothing but calling Model's gradient
+ *
  * @param topology topology
  * @param dataStacker data stacker
  */
@@ -682,7 +590,7 @@ private[ann] class ANNGradient(topology: Topology, dataStacker: DataStacker) ext
     weights: Vector,
     cumGradient: Vector): Double = {
     val (input, target, realBatchSize) = dataStacker.unstack(data)
-    val model = topology.getInstance(weights)
+    val model = topology.model(weights)
     model.computeGradient(input, target, cumGradient, realBatchSize)
   }
 }
@@ -692,6 +600,7 @@ private[ann] class ANNGradient(topology: Topology, dataStacker: DataStacker) ext
  * through Optimizer/Gradient interfaces. If stackSize is more than one, makes blocks
  * or matrices of inputs and outputs and then stack them in one vector.
  * This can be used for further batch computations after unstacking.
+ *
  * @param stackSize stack size
  * @param inputSize size of the input vectors
  * @param outputSize size of the output vectors
@@ -701,6 +610,7 @@ private[ann] class DataStacker(stackSize: Int, inputSize: Int, outputSize: Int)
 
   /**
    * Stacks the data
+ *
    * @param data RDD of vector pairs
    * @return RDD of double (always zero) and vector that contains the stacked vectors
    */
@@ -733,6 +643,7 @@ private[ann] class DataStacker(stackSize: Int, inputSize: Int, outputSize: Int)
 
   /**
    * Unstack the stacked vectors into matrices for batch operations
+ *
    * @param data stacked vector
    * @return pair of matrices holding input and output data and the real stack size
    */
@@ -765,6 +676,7 @@ private[ann] class ANNUpdater extends Updater {
 
 /**
  * MLlib-style trainer class that trains a network given the data and topology
+ *
  * @param topology topology of ANN
  * @param inputSize input size
  * @param outputSize output size
@@ -774,36 +686,50 @@ private[ml] class FeedForwardTrainer(
     val inputSize: Int,
     val outputSize: Int) extends Serializable {
 
-  // TODO: what if we need to pass random seed?
-  private var _weights = topology.getInstance(11L).weights()
+  private var _seed = this.getClass.getName.hashCode.toLong
+  private var _weights: Vector = null
   private var _stackSize = 128
   private var dataStacker = new DataStacker(_stackSize, inputSize, outputSize)
   private var _gradient: Gradient = new ANNGradient(topology, dataStacker)
   private var _updater: Updater = new ANNUpdater()
   private var optimizer: Optimizer = LBFGSOptimizer.setConvergenceTol(1e-4).setNumIterations(100)
 
+  /**
+   * Returns seed
+   */
+  def getSeed: Long = _seed
+
+  /**
+   * Sets seed
+   */
+  def setSeed(value: Long): this.type = {
+    _seed = value
+    this
+  }
+
   /**
    * Returns weights
-   * @return weights
    */
   def getWeights: Vector = _weights
 
   /**
    * Sets weights
+ *
    * @param value weights
    * @return trainer
    */
-  def setWeights(value: Vector): FeedForwardTrainer = {
+  def setWeights(value: Vector): this.type = {
     _weights = value
     this
   }
 
   /**
    * Sets the stack size
+ *
    * @param value stack size
    * @return trainer
    */
-  def setStackSize(value: Int): FeedForwardTrainer = {
+  def setStackSize(value: Int): this.type = {
     _stackSize = value
     dataStacker = new DataStacker(value, inputSize, outputSize)
     this
@@ -811,6 +737,7 @@ private[ml] class FeedForwardTrainer(
 
   /**
    * Sets the SGD optimizer
+ *
    * @return SGD optimizer
    */
   def SGDOptimizer: GradientDescent = {
@@ -821,6 +748,7 @@ private[ml] class FeedForwardTrainer(
 
   /**
    * Sets the LBFGS optimizer
+ *
    * @return LBGS optimizer
    */
   def LBFGSOptimizer: LBFGS = {
@@ -831,10 +759,11 @@ private[ml] class FeedForwardTrainer(
 
   /**
    * Sets the updater
+ *
    * @param value updater
    * @return trainer
    */
-  def setUpdater(value: Updater): FeedForwardTrainer = {
+  def setUpdater(value: Updater): this.type = {
     _updater = value
     updateUpdater(value)
     this
@@ -842,10 +771,11 @@ private[ml] class FeedForwardTrainer(
 
   /**
    * Sets the gradient
+ *
    * @param value gradient
    * @return trainer
    */
-  def setGradient(value: Gradient): FeedForwardTrainer = {
+  def setGradient(value: Gradient): this.type = {
     _gradient = value
     updateGradient(value)
     this
@@ -871,12 +801,20 @@ private[ml] class FeedForwardTrainer(
 
   /**
    * Trains the ANN
+ *
    * @param data RDD of input and output vector pairs
    * @return model
    */
   def train(data: RDD[(Vector, Vector)]): TopologyModel = {
-    val newWeights = optimizer.optimize(dataStacker.stack(data), getWeights)
-    topology.getInstance(newWeights)
+    val w = if (getWeights == null) {
+      // TODO: will make a copy if vector is a subvector of BDV (see Vectors code)
+      topology.model(_seed).weights
+    } else {
+      getWeights
+    }
+    // TODO: deprecate standard optimizer because it needs Vector
+    val newWeights = optimizer.optimize(dataStacker.stack(data), w)
+    topology.model(newWeights)
   }
 
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/ann/LossFunction.scala b/mllib/src/main/scala/org/apache/spark/ml/ann/LossFunction.scala
new file mode 100644
index 0000000000..32d78e9b22
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/ann/LossFunction.scala
@@ -0,0 +1,124 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.ann
+
+import java.util.Random
+
+import breeze.linalg.{sum => Bsum, DenseMatrix => BDM, DenseVector => BDV}
+import breeze.numerics.{log => brzlog}
+
+/**
+ * Trait for loss function
+ */
+private[ann] trait LossFunction {
+  /**
+   * Returns the value of loss function.
+   * Computes loss based on target and output.
+   * Writes delta (error) to delta in place.
+   * Delta is allocated based on the outputSize
+   * of model implementation.
+   *
+   * @param output actual output
+   * @param target target output
+   * @param delta delta (updated in place)
+   * @return loss
+   */
+  def loss(output: BDM[Double], target: BDM[Double], delta: BDM[Double]): Double
+}
+
+private[ann] class SigmoidLayerWithSquaredError extends Layer {
+  override val weightSize = 0
+  override val inPlace = true
+
+  override def getOutputSize(inputSize: Int): Int = inputSize
+  override def createModel(weights: BDV[Double]): LayerModel =
+    new SigmoidLayerModelWithSquaredError()
+  override def initModel(weights: BDV[Double], random: Random): LayerModel =
+    new SigmoidLayerModelWithSquaredError()
+}
+
+private[ann] class SigmoidLayerModelWithSquaredError
+  extends FunctionalLayerModel(new FunctionalLayer(new SigmoidFunction)) with LossFunction {
+  override def loss(output: BDM[Double], target: BDM[Double], delta: BDM[Double]): Double = {
+    ApplyInPlace(output, target, delta, (o: Double, t: Double) => o - t)
+    val error = Bsum(delta :* delta) / 2 / output.cols
+    ApplyInPlace(delta, output, delta, (x: Double, o: Double) => x * (o - o * o))
+    error
+  }
+}
+
+private[ann] class SoftmaxLayerWithCrossEntropyLoss extends Layer {
+  override val weightSize = 0
+  override val inPlace = true
+
+  override def getOutputSize(inputSize: Int): Int = inputSize
+  override def createModel(weights: BDV[Double]): LayerModel =
+    new SoftmaxLayerModelWithCrossEntropyLoss()
+  override def initModel(weights: BDV[Double], random: Random): LayerModel =
+    new SoftmaxLayerModelWithCrossEntropyLoss()
+}
+
+private[ann] class SoftmaxLayerModelWithCrossEntropyLoss extends LayerModel with LossFunction {
+
+  // loss layer models do not have weights
+  val weights = new BDV[Double](0)
+
+  override def eval(data: BDM[Double], output: BDM[Double]): Unit = {
+    var j = 0
+    // find max value to make sure later that exponent is computable
+    while (j < data.cols) {
+      var i = 0
+      var max = Double.MinValue
+      while (i < data.rows) {
+        if (data(i, j) > max) {
+          max = data(i, j)
+        }
+        i += 1
+      }
+      var sum = 0.0
+      i = 0
+      while (i < data.rows) {
+        val res = math.exp(data(i, j) - max)
+        output(i, j) = res
+        sum += res
+        i += 1
+      }
+      i = 0
+      while (i < data.rows) {
+        output(i, j) /= sum
+        i += 1
+      }
+      j += 1
+    }
+  }
+  override def computePrevDelta(
+    nextDelta: BDM[Double],
+    input: BDM[Double],
+    delta: BDM[Double]): Unit = {
+    /* loss layer model computes delta in loss function */
+  }
+
+  override def grad(delta: BDM[Double], input: BDM[Double], cumGrad: BDV[Double]): Unit = {
+    /* loss layer model does not have weights */
+  }
+
+  override def loss(output: BDM[Double], target: BDM[Double], delta: BDM[Double]): Double = {
+    ApplyInPlace(output, target, delta, (o: Double, t: Double) => o - t)
+    -Bsum( target :* brzlog(output)) / output.cols
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
index 7ce3ec68da..79bb2a8855 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
@@ -24,8 +24,8 @@ import org.apache.hadoop.fs.Path
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml.{PredictionModel, Predictor, PredictorParams}
 import org.apache.spark.ml.ann.{FeedForwardTopology, FeedForwardTrainer}
-import org.apache.spark.ml.param.{IntArrayParam, IntParam, ParamMap, ParamValidators}
-import org.apache.spark.ml.param.shared.{HasMaxIter, HasSeed, HasTol}
+import org.apache.spark.ml.param._
+import org.apache.spark.ml.param.shared.{HasMaxIter, HasSeed, HasStepSize, HasTol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
@@ -33,11 +33,12 @@ import org.apache.spark.sql.DataFrame
 
 /** Params for Multilayer Perceptron. */
 private[ml] trait MultilayerPerceptronParams extends PredictorParams
-  with HasSeed with HasMaxIter with HasTol {
+  with HasSeed with HasMaxIter with HasTol with HasStepSize {
   /**
    * Layer sizes including input size and output size.
    * Default: Array(1, 1)
-   * @group param
+    *
+    * @group param
    */
   final val layers: IntArrayParam = new IntArrayParam(this, "layers",
     "Sizes of layers from input layer to output layer" +
@@ -55,7 +56,8 @@ private[ml] trait MultilayerPerceptronParams extends PredictorParams
    * a partition then it is adjusted to the size of this data.
    * Recommended size is between 10 and 1000.
    * Default: 128
-   * @group expertParam
+    *
+    * @group expertParam
    */
   final val blockSize: IntParam = new IntParam(this, "blockSize",
     "Block size for stacking input data in matrices. Data is stacked within partitions." +
@@ -66,7 +68,33 @@ private[ml] trait MultilayerPerceptronParams extends PredictorParams
   /** @group getParam */
   final def getBlockSize: Int = $(blockSize)
 
-  setDefault(maxIter -> 100, tol -> 1e-4, layers -> Array(1, 1), blockSize -> 128)
+  /**
+   * Allows setting the solver: minibatch gradient descent (gd) or l-bfgs.
+   * l-bfgs is the default one.
+ *
+   * @group expertParam
+   */
+  final val solver: Param[String] = new Param[String](this, "solver",
+    " Allows setting the solver: minibatch gradient descent (gd) or l-bfgs. " +
+      " l-bfgs is the default one.",
+    ParamValidators.inArray[String](Array("gd", "l-bfgs")))
+
+  /** @group getParam */
+  final def getOptimizer: String = $(solver)
+
+  /**
+   * Model weights. Can be returned either after training or after explicit setting
+    *
+    * @group expertParam
+   */
+  final val weights: Param[Vector] = new Param[Vector](this, "weights",
+    " Sets the weights of the model ")
+
+  /** @group getParam */
+  final def getWeights: Vector = $(weights)
+
+
+  setDefault(maxIter -> 100, tol -> 1e-4, blockSize -> 128, solver -> "l-bfgs", stepSize -> 0.03)
 }
 
 /** Label to vector converter. */
@@ -105,6 +133,7 @@ private object LabelConverter {
  * Each layer has sigmoid activation function, output layer has softmax.
  * Number of inputs has to be equal to the size of feature vectors.
  * Number of outputs has to be equal to the total number of labels.
+ *
  */
 @Since("1.5.0")
 @Experimental
@@ -127,7 +156,8 @@ class MultilayerPerceptronClassifier @Since("1.5.0") (
   /**
    * Set the maximum number of iterations.
    * Default is 100.
-   * @group setParam
+    *
+    * @group setParam
    */
   @Since("1.5.0")
   def setMaxIter(value: Int): this.type = set(maxIter, value)
@@ -136,18 +166,28 @@ class MultilayerPerceptronClassifier @Since("1.5.0") (
    * Set the convergence tolerance of iterations.
    * Smaller value will lead to higher accuracy with the cost of more iterations.
    * Default is 1E-4.
-   * @group setParam
+    *
+    * @group setParam
    */
   @Since("1.5.0")
   def setTol(value: Double): this.type = set(tol, value)
 
   /**
-   * Set the seed for weights initialization.
-   * @group setParam
+   * Set the seed for weights initialization if weights are not set
+    *
+    * @group setParam
    */
   @Since("1.5.0")
   def setSeed(value: Long): this.type = set(seed, value)
 
+  /**
+   * Sets the model weights.
+    *
+    * @group expertParam
+   */
+  @Since("2.0.0")
+  def setWeights(value: Vector): this.type = set(weights, value)
+
   @Since("1.5.0")
   override def copy(extra: ParamMap): MultilayerPerceptronClassifier = defaultCopy(extra)
 
@@ -165,11 +205,18 @@ class MultilayerPerceptronClassifier @Since("1.5.0") (
     val lpData = extractLabeledPoints(dataset)
     val data = lpData.map(lp => LabelConverter.encodeLabeledPoint(lp, labels))
     val topology = FeedForwardTopology.multiLayerPerceptron(myLayers, true)
-    val FeedForwardTrainer = new FeedForwardTrainer(topology, myLayers(0), myLayers.last)
-    FeedForwardTrainer.LBFGSOptimizer.setConvergenceTol($(tol)).setNumIterations($(maxIter))
-    FeedForwardTrainer.setStackSize($(blockSize))
-    val mlpModel = FeedForwardTrainer.train(data)
-    new MultilayerPerceptronClassificationModel(uid, myLayers, mlpModel.weights())
+    val trainer = new FeedForwardTrainer(topology, myLayers(0), myLayers.last)
+    if (isDefined(weights)) {
+      trainer.setWeights($(weights))
+    } else {
+      trainer.setSeed($(seed))
+    }
+    trainer.LBFGSOptimizer
+      .setConvergenceTol($(tol))
+      .setNumIterations($(maxIter))
+    trainer.setStackSize($(blockSize))
+    val mlpModel = trainer.train(data)
+    new MultilayerPerceptronClassificationModel(uid, myLayers, mlpModel.weights)
   }
 }
 
@@ -185,7 +232,8 @@ object MultilayerPerceptronClassifier
  * :: Experimental ::
  * Classification model based on the Multilayer Perceptron.
  * Each layer has sigmoid activation function, output layer has softmax.
- * @param uid uid
+  *
+  * @param uid uid
  * @param layers array of layer sizes including input and output layers
  * @param weights vector of initial weights for the model that consists of the weights of layers
  * @return prediction model
@@ -202,7 +250,7 @@ class MultilayerPerceptronClassificationModel private[ml] (
   @Since("1.6.0")
   override val numFeatures: Int = layers.head
 
-  private val mlpModel = FeedForwardTopology.multiLayerPerceptron(layers, true).getInstance(weights)
+  private val mlpModel = FeedForwardTopology.multiLayerPerceptron(layers, true).model(weights)
 
   /**
    * Returns layers in a Java List.
diff --git a/mllib/src/test/java/org/apache/spark/ml/classification/JavaMultilayerPerceptronClassifierSuite.java b/mllib/src/test/java/org/apache/spark/ml/classification/JavaMultilayerPerceptronClassifierSuite.java
index d499d363f1..bc955f3cf6 100644
--- a/mllib/src/test/java/org/apache/spark/ml/classification/JavaMultilayerPerceptronClassifierSuite.java
+++ b/mllib/src/test/java/org/apache/spark/ml/classification/JavaMultilayerPerceptronClassifierSuite.java
@@ -63,7 +63,7 @@ public class JavaMultilayerPerceptronClassifierSuite implements Serializable {
     MultilayerPerceptronClassifier mlpc = new MultilayerPerceptronClassifier()
       .setLayers(new int[] {2, 5, 2})
       .setBlockSize(1)
-      .setSeed(11L)
+      .setSeed(123L)
       .setMaxIter(100);
     MultilayerPerceptronClassificationModel model = mlpc.fit(dataFrame);
     Dataset<Row> result = model.transform(dataFrame);
diff --git a/mllib/src/test/scala/org/apache/spark/ml/ann/ANNSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/ann/ANNSuite.scala
index 1292e57d7c..dc91fc5f9e 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/ann/ANNSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/ann/ANNSuite.scala
@@ -42,7 +42,7 @@ class ANNSuite extends SparkFunSuite with MLlibTestSparkContext {
     val dataSample = rddData.first()
     val layerSizes = dataSample._1.size +: hiddenLayersTopology :+ dataSample._2.size
     val topology = FeedForwardTopology.multiLayerPerceptron(layerSizes, false)
-    val initialWeights = FeedForwardModel(topology, 23124).weights()
+    val initialWeights = FeedForwardModel(topology, 23124).weights
     val trainer = new FeedForwardTrainer(topology, 2, 1)
     trainer.setWeights(initialWeights)
     trainer.LBFGSOptimizer.setNumIterations(20)
@@ -76,10 +76,11 @@ class ANNSuite extends SparkFunSuite with MLlibTestSparkContext {
     val dataSample = rddData.first()
     val layerSizes = dataSample._1.size +: hiddenLayersTopology :+ dataSample._2.size
     val topology = FeedForwardTopology.multiLayerPerceptron(layerSizes, false)
-    val initialWeights = FeedForwardModel(topology, 23124).weights()
+    val initialWeights = FeedForwardModel(topology, 23124).weights
     val trainer = new FeedForwardTrainer(topology, 2, 2)
-    trainer.SGDOptimizer.setNumIterations(2000)
-    trainer.setWeights(initialWeights)
+    // TODO: add a test for SGD
+    trainer.LBFGSOptimizer.setConvergenceTol(1e-4).setNumIterations(20)
+    trainer.setWeights(initialWeights).setStackSize(1)
     val model = trainer.train(rddData)
     val predictionAndLabels = rddData.map { case (input, label) =>
       (model.predict(input), label)
diff --git a/mllib/src/test/scala/org/apache/spark/ml/ann/GradientSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/ann/GradientSuite.scala
new file mode 100644
index 0000000000..04cc426c40
--- /dev/null
+++ b/mllib/src/test/scala/org/apache/spark/ml/ann/GradientSuite.scala
@@ -0,0 +1,76 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.ann
+
+import breeze.linalg.{DenseMatrix => BDM}
+
+import org.apache.spark.SparkFunSuite
+import org.apache.spark.mllib.linalg.Vectors
+import org.apache.spark.mllib.util.MLlibTestSparkContext
+
+class GradientSuite extends SparkFunSuite with MLlibTestSparkContext {
+
+  test("Gradient computation against numerical differentiation") {
+    val input = new BDM[Double](3, 1, Array(1.0, 1.0, 1.0))
+    // output must contain zeros and one 1 for SoftMax
+    val target = new BDM[Double](2, 1, Array(0.0, 1.0))
+    val topology = FeedForwardTopology.multiLayerPerceptron(Array(3, 4, 2), softmaxOnTop = false)
+    val layersWithErrors = Seq(
+      new SigmoidLayerWithSquaredError(),
+      new SoftmaxLayerWithCrossEntropyLoss()
+    )
+    // check all layers that provide loss computation
+    // 1) compute loss and gradient given the model and initial weights
+    // 2) modify weights with small number epsilon (per dimension i)
+    // 3) compute new loss
+    // 4) ((newLoss - loss) / epsilon) should be close to the i-th component of the gradient
+    for (layerWithError <- layersWithErrors) {
+      topology.layers(topology.layers.length - 1) = layerWithError
+      val model = topology.model(seed = 12L)
+      val weights = model.weights.toArray
+      val numWeights = weights.size
+      val gradient = Vectors.dense(Array.fill[Double](numWeights)(0.0))
+      val loss = model.computeGradient(input, target, gradient, 1)
+      val eps = 1e-4
+      var i = 0
+      val tol = 1e-4
+      while (i < numWeights) {
+        val originalValue = weights(i)
+        weights(i) += eps
+        val newModel = topology.model(Vectors.dense(weights))
+        val newLoss = computeLoss(input, target, newModel)
+        val derivativeEstimate = (newLoss - loss) / eps
+        assert(math.abs(gradient(i) - derivativeEstimate) < tol, "Layer failed gradient check: " +
+          layerWithError.getClass)
+        weights(i) = originalValue
+        i += 1
+      }
+    }
+  }
+
+  private def computeLoss(input: BDM[Double], target: BDM[Double], model: TopologyModel): Double = {
+    val outputs = model.forward(input)
+    model.layerModels.last match {
+      case layerWithLoss: LossFunction =>
+        layerWithLoss.loss(outputs.last, target, new BDM[Double](target.rows, target.cols))
+      case _ =>
+        throw new UnsupportedOperationException("Top layer is required to have loss." +
+          " Failed layer:" + model.layerModels.last.getClass)
+    }
+  }
+}
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
index 53c7a559e3..43781385db 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
@@ -65,7 +65,7 @@ class MultilayerPerceptronClassifierSuite
     val trainer = new MultilayerPerceptronClassifier()
       .setLayers(layers)
       .setBlockSize(1)
-      .setSeed(11L)
+      .setSeed(123L)
       .setMaxIter(100)
     val model = trainer.fit(dataset)
     val result = model.transform(dataset)
@@ -75,7 +75,29 @@ class MultilayerPerceptronClassifierSuite
     }
   }
 
-  // TODO: implement a more rigorous test
+  test("Test setWeights by training restart") {
+    val dataFrame = sqlContext.createDataFrame(Seq(
+      (Vectors.dense(0.0, 0.0), 0.0),
+      (Vectors.dense(0.0, 1.0), 1.0),
+      (Vectors.dense(1.0, 0.0), 1.0),
+      (Vectors.dense(1.0, 1.0), 0.0))
+    ).toDF("features", "label")
+    val layers = Array[Int](2, 5, 2)
+    val trainer = new MultilayerPerceptronClassifier()
+      .setLayers(layers)
+      .setBlockSize(1)
+      .setSeed(12L)
+      .setMaxIter(1)
+      .setTol(1e-6)
+    val initialWeights = trainer.fit(dataFrame).weights
+    trainer.setWeights(initialWeights.copy)
+    val weights1 = trainer.fit(dataFrame).weights
+    trainer.setWeights(initialWeights.copy)
+    val weights2 = trainer.fit(dataFrame).weights
+    assert(weights1 ~== weights2 absTol 10e-5,
+      "Training should produce the same weights given equal initial weights and number of steps")
+  }
+
   test("3 class classification with 2 hidden layers") {
     val nPoints = 1000
 
diff --git a/project/MimaExcludes.scala b/project/MimaExcludes.scala
index 94621d7fa3..ff11775412 100644
--- a/project/MimaExcludes.scala
+++ b/project/MimaExcludes.scala
@@ -592,6 +592,11 @@ object MimaExcludes {
       ) ++ Seq(
         // [SPARK-14205][SQL] remove trait Queryable
         ProblemFilters.exclude[MissingTypesProblem]("org.apache.spark.sql.Dataset")
+      ) ++ Seq(
+        // [SPARK-11262][ML] Unit test for gradient, loss layers, memory management
+        // for multilayer perceptron.
+        // This class is marked as `private`.
+        ProblemFilters.exclude[MissingClassProblem]("org.apache.spark.ml.ann.SoftmaxFunction")
       )
     case v if v.startsWith("1.6") =>
       Seq(
diff --git a/python/pyspark/ml/classification.py b/python/pyspark/ml/classification.py
index f5335a3114..067009559b 100644
--- a/python/pyspark/ml/classification.py
+++ b/python/pyspark/ml/classification.py
@@ -788,7 +788,7 @@ class MultilayerPerceptronClassifier(JavaEstimator, HasFeaturesCol, HasLabelCol,
     ...     (1.0, Vectors.dense([0.0, 1.0])),
     ...     (1.0, Vectors.dense([1.0, 0.0])),
     ...     (0.0, Vectors.dense([1.0, 1.0]))], ["label", "features"])
-    >>> mlp = MultilayerPerceptronClassifier(maxIter=100, layers=[2, 5, 2], blockSize=1, seed=11)
+    >>> mlp = MultilayerPerceptronClassifier(maxIter=100, layers=[2, 5, 2], blockSize=1, seed=123)
     >>> model = mlp.fit(df)
     >>> model.layers
     [2, 5, 2]
-- 
cgit v1.2.3


From 22249afb4a932a82ff1f7a3befea9fda5a60a3f4 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Thu, 31 Mar 2016 23:49:58 -0700
Subject: [SPARK-14303][ML][SPARKR] Define and use KMeansWrapper for
 SparkR::kmeans

## What changes were proposed in this pull request?
Define and use ```KMeansWrapper``` for ```SparkR::kmeans```. It's only the code refactor for the original ```KMeans``` wrapper.

## How was this patch tested?
Existing tests.

cc mengxr

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12039 from yanboliang/spark-14059.
---
 R/pkg/R/mllib.R                                    | 91 +++++++++++++++-------
 .../org/apache/spark/ml/r/KMeansWrapper.scala      | 85 ++++++++++++++++++++
 .../org/apache/spark/ml/r/SparkRWrappers.scala     | 52 +------------
 3 files changed, 148 insertions(+), 80 deletions(-)
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/R/pkg/R/mllib.R b/R/pkg/R/mllib.R
index 33654d5216..f3152cc232 100644
--- a/R/pkg/R/mllib.R
+++ b/R/pkg/R/mllib.R
@@ -32,6 +32,11 @@ setClass("NaiveBayesModel", representation(jobj = "jobj"))
 #' @export
 setClass("AFTSurvivalRegressionModel", representation(jobj = "jobj"))
 
+#' @title S4 class that represents a KMeansModel
+#' @param jobj a Java object reference to the backing Scala KMeansModel
+#' @export
+setClass("KMeansModel", representation(jobj = "jobj"))
+
 #' Fits a generalized linear model
 #'
 #' Fits a generalized linear model, similarly to R's glm(). Also see the glmnet package.
@@ -154,17 +159,6 @@ setMethod("summary", signature(object = "PipelineModel"),
               colnames(coefficients) <- c("Estimate")
               rownames(coefficients) <- unlist(features)
               return(list(coefficients = coefficients))
-            } else if (modelName == "KMeansModel") {
-              modelSize <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                       "getKMeansModelSize", object@model)
-              cluster <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                     "getKMeansCluster", object@model, "classes")
-              k <- unlist(modelSize)[1]
-              size <- unlist(modelSize)[-1]
-              coefficients <- t(matrix(coefficients, ncol = k))
-              colnames(coefficients) <- unlist(features)
-              rownames(coefficients) <- 1:k
-              return(list(coefficients = coefficients, size = size, cluster = dataFrame(cluster)))
             } else {
               stop(paste("Unsupported model", modelName, sep = " "))
             }
@@ -213,21 +207,21 @@ setMethod("summary", signature(object = "NaiveBayesModel"),
 #' @examples
 #' \dontrun{
 #' model <- kmeans(x, centers = 2, algorithm="random")
-#'}
+#' }
 setMethod("kmeans", signature(x = "DataFrame"),
           function(x, centers, iter.max = 10, algorithm = c("random", "k-means||")) {
             columnNames <- as.array(colnames(x))
             algorithm <- match.arg(algorithm)
-            model <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers", "fitKMeans", x@sdf,
-                                 algorithm, iter.max, centers, columnNames)
-            return(new("PipelineModel", model = model))
+            jobj <- callJStatic("org.apache.spark.ml.r.KMeansWrapper", "fit", x@sdf,
+                                centers, iter.max, algorithm, columnNames)
+            return(new("KMeansModel", jobj = jobj))
          })
 
-#' Get fitted result from a model
+#' Get fitted result from a k-means model
 #'
-#' Get fitted result from a model, similarly to R's fitted().
+#' Get fitted result from a k-means model, similarly to R's fitted().
 #'
-#' @param object A fitted MLlib model
+#' @param object A fitted k-means model
 #' @return DataFrame containing fitted values
 #' @rdname fitted
 #' @export
@@ -237,19 +231,58 @@ setMethod("kmeans", signature(x = "DataFrame"),
 #' fitted.model <- fitted(model)
 #' showDF(fitted.model)
 #'}
-setMethod("fitted", signature(object = "PipelineModel"),
+setMethod("fitted", signature(object = "KMeansModel"),
           function(object, method = c("centers", "classes"), ...) {
-            modelName <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                     "getModelName", object@model)
+            method <- match.arg(method)
+            return(dataFrame(callJMethod(object@jobj, "fitted", method)))
+          })
 
-            if (modelName == "KMeansModel") {
-              method <- match.arg(method)
-              fittedResult <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                          "getKMeansCluster", object@model, method)
-              return(dataFrame(fittedResult))
-            } else {
-              stop(paste("Unsupported model", modelName, sep = " "))
-            }
+#' Get the summary of a k-means model
+#'
+#' Returns the summary of a k-means model produced by kmeans(),
+#' similarly to R's summary().
+#'
+#' @param object a fitted k-means model
+#' @return the model's coefficients, size and cluster
+#' @rdname summary
+#' @export
+#' @examples
+#' \dontrun{
+#' model <- kmeans(trainingData, 2)
+#' summary(model)
+#' }
+setMethod("summary", signature(object = "KMeansModel"),
+          function(object, ...) {
+            jobj <- object@jobj
+            features <- callJMethod(jobj, "features")
+            coefficients <- callJMethod(jobj, "coefficients")
+            cluster <- callJMethod(jobj, "cluster")
+            k <- callJMethod(jobj, "k")
+            size <- callJMethod(jobj, "size")
+            coefficients <- t(matrix(coefficients, ncol = k))
+            colnames(coefficients) <- unlist(features)
+            rownames(coefficients) <- 1:k
+            return(list(coefficients = coefficients, size = size, cluster = dataFrame(cluster)))
+          })
+
+#' Make predictions from a k-means model
+#'
+#' Make predictions from a model produced by kmeans().
+#'
+#' @param object A fitted k-means model
+#' @param newData DataFrame for testing
+#' @return DataFrame containing predicted labels in a column named "prediction"
+#' @rdname predict
+#' @export
+#' @examples
+#' \dontrun{
+#' model <- kmeans(trainingData, 2)
+#' predicted <- predict(model, testData)
+#' showDF(predicted)
+#' }
+setMethod("predict", signature(object = "KMeansModel"),
+          function(object, newData) {
+            return(dataFrame(callJMethod(object@jobj, "transform", newData@sdf)))
           })
 
 #' Fit a Bernoulli naive Bayes model
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
new file mode 100644
index 0000000000..d3a0df4063
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
@@ -0,0 +1,85 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.r
+
+import org.apache.spark.ml.{Pipeline, PipelineModel}
+import org.apache.spark.ml.attribute.AttributeGroup
+import org.apache.spark.ml.clustering.{KMeans, KMeansModel}
+import org.apache.spark.ml.feature.VectorAssembler
+import org.apache.spark.sql.DataFrame
+
+private[r] class KMeansWrapper private (
+    pipeline: PipelineModel) {
+
+  private val kMeansModel: KMeansModel = pipeline.stages(1).asInstanceOf[KMeansModel]
+
+  lazy val coefficients: Array[Double] = kMeansModel.clusterCenters.flatMap(_.toArray)
+
+  private lazy val attrs = AttributeGroup.fromStructField(
+    kMeansModel.summary.predictions.schema(kMeansModel.getFeaturesCol))
+
+  lazy val features: Array[String] = attrs.attributes.get.map(_.name.get)
+
+  lazy val k: Int = kMeansModel.getK
+
+  lazy val size: Array[Int] = kMeansModel.summary.size
+
+  lazy val cluster: DataFrame = kMeansModel.summary.cluster
+
+  def fitted(method: String): DataFrame = {
+    if (method == "centers") {
+      kMeansModel.summary.predictions.drop(kMeansModel.getFeaturesCol)
+    } else if (method == "classes") {
+      kMeansModel.summary.cluster
+    } else {
+      throw new UnsupportedOperationException(
+        s"Method (centers or classes) required but $method found.")
+    }
+  }
+
+  def transform(dataset: DataFrame): DataFrame = {
+    pipeline.transform(dataset).drop(kMeansModel.getFeaturesCol)
+  }
+
+}
+
+private[r] object KMeansWrapper {
+
+  def fit(
+      data: DataFrame,
+      k: Double,
+      maxIter: Double,
+      initMode: String,
+      columns: Array[String]): KMeansWrapper = {
+
+    val assembler = new VectorAssembler()
+      .setInputCols(columns)
+      .setOutputCol("features")
+
+    val kMeans = new KMeans()
+      .setK(k.toInt)
+      .setMaxIter(maxIter.toInt)
+      .setInitMode(initMode)
+
+    val pipeline = new Pipeline()
+      .setStages(Array(assembler, kMeans))
+      .fit(data)
+
+    new KMeansWrapper(pipeline)
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala b/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
index d23e4fc9d1..551e75dc0a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
@@ -20,8 +20,7 @@ package org.apache.spark.ml.api.r
 import org.apache.spark.ml.{Pipeline, PipelineModel}
 import org.apache.spark.ml.attribute._
 import org.apache.spark.ml.classification.{LogisticRegression, LogisticRegressionModel}
-import org.apache.spark.ml.clustering.{KMeans, KMeansModel}
-import org.apache.spark.ml.feature.{RFormula, VectorAssembler}
+import org.apache.spark.ml.feature.RFormula
 import org.apache.spark.ml.regression.{LinearRegression, LinearRegressionModel}
 import org.apache.spark.sql.DataFrame
 
@@ -52,22 +51,6 @@ private[r] object SparkRWrappers {
     pipeline.fit(df)
   }
 
-  def fitKMeans(
-      df: DataFrame,
-      initMode: String,
-      maxIter: Double,
-      k: Double,
-      columns: Array[String]): PipelineModel = {
-    val assembler = new VectorAssembler().setInputCols(columns)
-    val kMeans = new KMeans()
-      .setInitMode(initMode)
-      .setMaxIter(maxIter.toInt)
-      .setK(k.toInt)
-      .setFeaturesCol(assembler.getOutputCol)
-    val pipeline = new Pipeline().setStages(Array(assembler, kMeans))
-    pipeline.fit(df)
-  }
-
   def getModelCoefficients(model: PipelineModel): Array[Double] = {
     model.stages.last match {
       case m: LinearRegressionModel => {
@@ -89,8 +72,6 @@ private[r] object SparkRWrappers {
           m.coefficients.toArray
         }
       }
-      case m: KMeansModel =>
-        m.clusterCenters.flatMap(_.toArray)
     }
   }
 
@@ -104,31 +85,6 @@ private[r] object SparkRWrappers {
     }
   }
 
-  def getKMeansModelSize(model: PipelineModel): Array[Int] = {
-    model.stages.last match {
-      case m: KMeansModel => Array(m.getK) ++ m.summary.size
-      case other => throw new UnsupportedOperationException(
-        s"KMeansModel required but ${other.getClass.getSimpleName} found.")
-    }
-  }
-
-  def getKMeansCluster(model: PipelineModel, method: String): DataFrame = {
-    model.stages.last match {
-      case m: KMeansModel =>
-        if (method == "centers") {
-          // Drop the assembled vector for easy-print to R side.
-          m.summary.predictions.drop(m.summary.featuresCol)
-        } else if (method == "classes") {
-          m.summary.cluster
-        } else {
-          throw new UnsupportedOperationException(
-            s"Method (centers or classes) required but $method found.")
-        }
-      case other => throw new UnsupportedOperationException(
-        s"KMeansModel required but ${other.getClass.getSimpleName} found.")
-    }
-  }
-
   def getModelFeatures(model: PipelineModel): Array[String] = {
     model.stages.last match {
       case m: LinearRegressionModel =>
@@ -147,10 +103,6 @@ private[r] object SparkRWrappers {
         } else {
           attrs.attributes.get.map(_.name.get)
         }
-      case m: KMeansModel =>
-        val attrs = AttributeGroup.fromStructField(
-          m.summary.predictions.schema(m.summary.featuresCol))
-        attrs.attributes.get.map(_.name.get)
     }
   }
 
@@ -160,8 +112,6 @@ private[r] object SparkRWrappers {
         "LinearRegressionModel"
       case m: LogisticRegressionModel =>
         "LogisticRegressionModel"
-      case m: KMeansModel =>
-        "KMeansModel"
     }
   }
 }
-- 
cgit v1.2.3


From 3715ecdf417b47423ff07145a5623d8d817c45ef Mon Sep 17 00:00:00 2001
From: Cheng Lian <lian@databricks.com>
Date: Fri, 1 Apr 2016 17:02:48 +0800
Subject: [SPARK-14295][MLLIB][HOTFIX] Fixes Scala 2.10 compilation failure

## What changes were proposed in this pull request?

Fixes a compilation failure introduced in PR #12088 under Scala 2.10.

## How was this patch tested?

Compilation.

Author: Cheng Lian <lian@databricks.com>

Closes #12107 from liancheng/spark-14295-hotfix.
---
 mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala b/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
index 4b9d77949f..774170ff40 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
@@ -120,7 +120,7 @@ object MLUtils {
       i += 1
     }
 
-    (label, indices, values)
+    (label, indices.toArray, values.toArray)
   }
 
   /**
-- 
cgit v1.2.3


From 36e8fb8005eccea67a9dea8cf68ec3105aa43351 Mon Sep 17 00:00:00 2001
From: BenFradet <benjamin.fradet@gmail.com>
Date: Fri, 1 Apr 2016 18:25:43 -0700
Subject: [SPARK-7425][ML] spark.ml Predictor should support other numeric
 types for label

Currently, the Predictor abstraction expects the input labelCol type to be DoubleType, but we should support other numeric types. This will involve updating the PredictorParams.validateAndTransformSchema method.

Author: BenFradet <benjamin.fradet@gmail.com>

Closes #10355 from BenFradet/SPARK-7425.
---
 .../main/scala/org/apache/spark/ml/Predictor.scala |  9 +--
 .../ml/classification/LogisticRegression.scala     |  7 +-
 .../apache/spark/ml/classification/OneVsRest.scala |  4 +-
 .../ml/regression/AFTSurvivalRegression.scala      | 11 +--
 .../regression/GeneralizedLinearRegression.scala   | 13 ++--
 .../spark/ml/regression/IsotonicRegression.scala   |  4 +-
 .../spark/ml/regression/LinearRegression.scala     | 11 +--
 .../org/apache/spark/ml/util/SchemaUtils.scala     | 24 ++++--
 .../DecisionTreeClassifierSuite.scala              | 15 +++-
 .../ml/classification/GBTClassifierSuite.scala     |  9 ++-
 .../classification/LogisticRegressionSuite.scala   | 11 ++-
 .../MultilayerPerceptronClassifierSuite.scala      | 12 +++
 .../spark/ml/classification/NaiveBayesSuite.scala  | 14 +++-
 .../spark/ml/classification/OneVsRestSuite.scala   | 16 +++-
 .../RandomForestClassifierSuite.scala              |  8 ++
 .../ml/regression/AFTSurvivalRegressionSuite.scala |  9 +++
 .../ml/regression/DecisionTreeRegressorSuite.scala |  8 ++
 .../spark/ml/regression/GBTRegressorSuite.scala    |  8 +-
 .../GeneralizedLinearRegressionSuite.scala         | 12 ++-
 .../ml/regression/IsotonicRegressionSuite.scala    |  9 +++
 .../ml/regression/LinearRegressionSuite.scala      | 17 ++++-
 .../ml/regression/RandomForestRegressorSuite.scala |  8 ++
 .../org/apache/spark/ml/tree/impl/TreeTests.scala  | 18 +++++
 .../org/apache/spark/ml/util/MLTestingUtils.scala  | 86 +++++++++++++++++++++-
 24 files changed, 294 insertions(+), 49 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala b/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
index ebe48700f8..d23ae6f794 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
@@ -36,6 +36,7 @@ private[ml] trait PredictorParams extends Params
 
   /**
    * Validates and transforms the input schema with the provided param map.
+   *
    * @param schema input schema
    * @param fitting whether this is in fitting
    * @param featuresDataType  SQL DataType for FeaturesType.
@@ -49,8 +50,7 @@ private[ml] trait PredictorParams extends Params
     // TODO: Support casting Array[Double] and Array[Float] to Vector when FeaturesType = Vector
     SchemaUtils.checkColumnType(schema, $(featuresCol), featuresDataType)
     if (fitting) {
-      // TODO: Allow other numeric types
-      SchemaUtils.checkColumnType(schema, $(labelCol), DoubleType)
+      SchemaUtils.checkNumericType(schema, $(labelCol))
     }
     SchemaUtils.appendColumn(schema, $(predictionCol), DoubleType)
   }
@@ -121,9 +121,8 @@ abstract class Predictor[
    * and put it in an RDD with strong types.
    */
   protected def extractLabeledPoints(dataset: DataFrame): RDD[LabeledPoint] = {
-    dataset.select($(labelCol), $(featuresCol)).rdd.map {
-      case Row(label: Double, features: Vector) =>
-        LabeledPoint(label, features)
+    dataset.select(col($(labelCol)).cast(DoubleType), col($(featuresCol))).rdd.map {
+      case Row(label: Double, features: Vector) => LabeledPoint(label, features)
     }
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
index 3d1d5b6892..aeb94a6600 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
@@ -38,6 +38,7 @@ import org.apache.spark.mllib.util.MLUtils
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Row}
 import org.apache.spark.sql.functions.{col, lit}
+import org.apache.spark.sql.types.DoubleType
 import org.apache.spark.storage.StorageLevel
 
 /**
@@ -265,7 +266,7 @@ class LogisticRegression @Since("1.2.0") (
       LogisticRegressionModel = {
     val w = if ($(weightCol).isEmpty) lit(1.0) else col($(weightCol))
     val instances: RDD[Instance] =
-      dataset.select(col($(labelCol)), w, col($(featuresCol))).rdd.map {
+      dataset.select(col($(labelCol)).cast(DoubleType), w, col($(featuresCol))).rdd.map {
         case Row(label: Double, weight: Double, features: Vector) =>
           Instance(label, weight, features)
       }
@@ -361,7 +362,7 @@ class LogisticRegression @Since("1.2.0") (
         if (optInitialModel.isDefined && optInitialModel.get.coefficients.size != numFeatures) {
           val vec = optInitialModel.get.coefficients
           logWarning(
-            s"Initial coefficients provided ${vec} did not match the expected size ${numFeatures}")
+            s"Initial coefficients provided $vec did not match the expected size $numFeatures")
         }
 
         if (optInitialModel.isDefined && optInitialModel.get.coefficients.size == numFeatures) {
@@ -522,7 +523,7 @@ class LogisticRegressionModel private[spark] (
       (LogisticRegressionModel, String) = {
     $(probabilityCol) match {
       case "" =>
-        val probabilityColName = "probability_" + java.util.UUID.randomUUID.toString()
+        val probabilityColName = "probability_" + java.util.UUID.randomUUID.toString
         (copy(ParamMap.empty).setProbabilityCol(probabilityColName), probabilityColName)
       case p => (this, p)
     }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
index 98b99a3485..263d54ce4d 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
@@ -295,10 +295,12 @@ final class OneVsRest @Since("1.4.0") (
 
   @Since("1.4.0")
   override def fit(dataset: DataFrame): OneVsRestModel = {
+    transformSchema(dataset.schema)
+
     // determine number of classes either from metadata if provided, or via computation.
     val labelSchema = dataset.schema($(labelCol))
     val computeNumClasses: () => Int = () => {
-      val Row(maxLabelIndex: Double) = dataset.agg(max($(labelCol))).head()
+      val Row(maxLabelIndex: Double) = dataset.agg(max(col($(labelCol)).cast(DoubleType))).head()
       // classes are assumed to be numbered from 0,...,maxLabelIndex
       maxLabelIndex.toInt + 1
     }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
index ba5708ab8d..3278974954 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
@@ -103,7 +103,7 @@ private[regression] trait AFTSurvivalRegressionParams extends Params
     SchemaUtils.checkColumnType(schema, $(featuresCol), new VectorUDT)
     if (fitting) {
       SchemaUtils.checkColumnType(schema, $(censorCol), DoubleType)
-      SchemaUtils.checkColumnType(schema, $(labelCol), DoubleType)
+      SchemaUtils.checkNumericType(schema, $(labelCol))
     }
     if (hasQuantilesCol) {
       SchemaUtils.appendColumn(schema, $(quantilesCol), new VectorUDT)
@@ -184,10 +184,11 @@ class AFTSurvivalRegression @Since("1.6.0") (@Since("1.6.0") override val uid: S
    * and put it in an RDD with strong types.
    */
   protected[ml] def extractAFTPoints(dataset: DataFrame): RDD[AFTPoint] = {
-    dataset.select($(featuresCol), $(labelCol), $(censorCol)).rdd.map {
-      case Row(features: Vector, label: Double, censor: Double) =>
-        AFTPoint(features, label, censor)
-    }
+    dataset.select(col($(featuresCol)), col($(labelCol)).cast(DoubleType), col($(censorCol)))
+      .rdd.map {
+        case Row(features: Vector, label: Double, censor: Double) =>
+          AFTPoint(features, label, censor)
+      }
   }
 
   @Since("1.6.0")
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
index 0e71e8d8e1..a40d3731cb 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
@@ -33,7 +33,7 @@ import org.apache.spark.mllib.linalg.{BLAS, Vector}
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Row}
 import org.apache.spark.sql.functions._
-import org.apache.spark.sql.types.{DataType, StructType}
+import org.apache.spark.sql.types.{DataType, DoubleType, StructType}
 
 /**
  * Params for Generalized Linear Regression.
@@ -47,6 +47,7 @@ private[regression] trait GeneralizedLinearRegressionBase extends PredictorParam
    * to be used in the model.
    * Supported options: "gaussian", "binomial", "poisson" and "gamma".
    * Default is "gaussian".
+   *
    * @group param
    */
   @Since("2.0.0")
@@ -63,6 +64,7 @@ private[regression] trait GeneralizedLinearRegressionBase extends PredictorParam
    * Param for the name of link function which provides the relationship
    * between the linear predictor and the mean of the distribution function.
    * Supported options: "identity", "log", "inverse", "logit", "probit", "cloglog" and "sqrt".
+   *
    * @group param
    */
   @Since("2.0.0")
@@ -210,9 +212,10 @@ class GeneralizedLinearRegression @Since("2.0.0") (@Since("2.0.0") override val
     }
 
     val w = if ($(weightCol).isEmpty) lit(1.0) else col($(weightCol))
-    val instances: RDD[Instance] = dataset.select(col($(labelCol)), w, col($(featuresCol))).rdd
-      .map { case Row(label: Double, weight: Double, features: Vector) =>
-        Instance(label, weight, features)
+    val instances: RDD[Instance] =
+      dataset.select(col($(labelCol)).cast(DoubleType), w, col($(featuresCol))).rdd.map {
+        case Row(label: Double, weight: Double, features: Vector) =>
+          Instance(label, weight, features)
       }
 
     if (familyObj == Gaussian && linkObj == Identity) {
@@ -698,7 +701,7 @@ class GeneralizedLinearRegressionModel private[ml] (
     : (GeneralizedLinearRegressionModel, String) = {
     $(predictionCol) match {
       case "" =>
-        val predictionColName = "prediction_" + java.util.UUID.randomUUID.toString()
+        val predictionColName = "prediction_" + java.util.UUID.randomUUID.toString
         (copy(ParamMap.empty).setPredictionCol(predictionColName), predictionColName)
       case p => (this, p)
     }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala
index fb733f9a34..bd0b631d89 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala
@@ -90,7 +90,7 @@ private[regression] trait IsotonicRegressionBase extends Params with HasFeatures
     } else {
       lit(1.0)
     }
-    dataset.select(col($(labelCol)), f, w).rdd.map {
+    dataset.select(col($(labelCol)).cast(DoubleType), f, w).rdd.map {
       case Row(label: Double, feature: Double, weight: Double) =>
         (label, feature, weight)
     }
@@ -106,7 +106,7 @@ private[regression] trait IsotonicRegressionBase extends Params with HasFeatures
       schema: StructType,
       fitting: Boolean): StructType = {
     if (fitting) {
-      SchemaUtils.checkColumnType(schema, $(labelCol), DoubleType)
+      SchemaUtils.checkNumericType(schema, $(labelCol))
       if (hasWeightCol) {
         SchemaUtils.checkColumnType(schema, $(weightCol), DoubleType)
       } else {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
index 5ec02135cc..ba5ad4c072 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
@@ -40,6 +40,7 @@ import org.apache.spark.mllib.stat.MultivariateOnlineSummarizer
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Row}
 import org.apache.spark.sql.functions._
+import org.apache.spark.sql.types.DoubleType
 import org.apache.spark.storage.StorageLevel
 
 /**
@@ -171,7 +172,7 @@ class LinearRegression @Since("1.3.0") (@Since("1.3.0") override val uid: String
       // For low dimensional data, WeightedLeastSquares is more efficiently since the
       // training algorithm only requires one pass through the data. (SPARK-10668)
       val instances: RDD[Instance] = dataset.select(
-        col($(labelCol)), w, col($(featuresCol))).rdd.map {
+        col($(labelCol)).cast(DoubleType), w, col($(featuresCol))).rdd.map {
           case Row(label: Double, weight: Double, features: Vector) =>
             Instance(label, weight, features)
       }
@@ -431,7 +432,7 @@ class LinearRegressionModel private[ml] (
   private[regression] def findSummaryModelAndPredictionCol(): (LinearRegressionModel, String) = {
     $(predictionCol) match {
       case "" =>
-        val predictionColName = "prediction_" + java.util.UUID.randomUUID.toString()
+        val predictionColName = "prediction_" + java.util.UUID.randomUUID.toString
         (copy(ParamMap.empty).setPredictionCol(predictionColName), predictionColName)
       case p => (this, p)
     }
@@ -550,7 +551,7 @@ class LinearRegressionSummary private[regression] (
 
   @transient private val metrics = new RegressionMetrics(
     predictions
-      .select(predictionCol, labelCol)
+      .select(col(predictionCol), col(labelCol).cast(DoubleType))
       .rdd
       .map { case Row(pred: Double, label: Double) => (pred, label) },
     !model.getFitIntercept)
@@ -653,7 +654,7 @@ class LinearRegressionSummary private[regression] (
           col(model.getWeightCol)).as("wse")).agg(sum(col("wse"))).first().getDouble(0)
       }
       val sigma2 = rss / degreesOfFreedom
-      diagInvAtWA.map(_ * sigma2).map(math.sqrt(_))
+      diagInvAtWA.map(_ * sigma2).map(math.sqrt)
     }
   }
 
@@ -826,7 +827,7 @@ private class LeastSquaresAggregator(
     instance match { case Instance(label, weight, features) =>
       require(dim == features.size, s"Dimensions mismatch when adding new sample." +
         s" Expecting $dim but got ${features.size}.")
-      require(weight >= 0.0, s"instance weight, ${weight} has to be >= 0.0")
+      require(weight >= 0.0, s"instance weight, $weight has to be >= 0.0")
 
       if (weight == 0.0) return this
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/util/SchemaUtils.scala b/mllib/src/main/scala/org/apache/spark/ml/util/SchemaUtils.scala
index 76021ad8f4..334410c962 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/util/SchemaUtils.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/util/SchemaUtils.scala
@@ -17,7 +17,7 @@
 
 package org.apache.spark.ml.util
 
-import org.apache.spark.sql.types.{DataType, StructField, StructType}
+import org.apache.spark.sql.types.{DataType, NumericType, StructField, StructType}
 
 
 /**
@@ -44,10 +44,10 @@ private[spark] object SchemaUtils {
   }
 
   /**
-    * Check whether the given schema contains a column of one of the require data types.
-    * @param colName  column name
-    * @param dataTypes  required column data types
-    */
+   * Check whether the given schema contains a column of one of the require data types.
+   * @param colName  column name
+   * @param dataTypes  required column data types
+   */
   def checkColumnTypes(
       schema: StructType,
       colName: String,
@@ -60,6 +60,20 @@ private[spark] object SchemaUtils {
         s"${dataTypes.mkString("[", ", ", "]")} but was actually of type $actualDataType.$message")
   }
 
+  /**
+   * Check whether the given schema contains a column of the numeric data type.
+   * @param colName  column name
+   */
+  def checkNumericType(
+      schema: StructType,
+      colName: String,
+      msg: String = ""): Unit = {
+    val actualDataType = schema(colName).dataType
+    val message = if (msg != null && msg.trim.length > 0) " " + msg else ""
+    require(actualDataType.isInstanceOf[NumericType], s"Column $colName must be of type " +
+      s"NumericType but was actually of type $actualDataType.$message")
+  }
+
   /**
    * Appends a new column to the input schema. This fails if the given output column already exists.
    * @param schema input schema
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/DecisionTreeClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/DecisionTreeClassifierSuite.scala
index 2b07524815..fe839e15e9 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/DecisionTreeClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/DecisionTreeClassifierSuite.scala
@@ -27,8 +27,7 @@ import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.{DecisionTree => OldDecisionTree, DecisionTreeSuite => OldDecisionTreeSuite}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
-import org.apache.spark.sql.Row
+import org.apache.spark.sql.{DataFrame, Row}
 
 class DecisionTreeClassifierSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
@@ -176,7 +175,7 @@ class DecisionTreeClassifierSuite
   }
 
   test("Multiclass classification tree with 10-ary (ordered) categorical features," +
-      " with just enough bins") {
+    " with just enough bins") {
     val rdd = categoricalDataPointsForMulticlassForOrderedFeaturesRDD
     val dt = new DecisionTreeClassifier()
       .setImpurity("Gini")
@@ -273,7 +272,7 @@ class DecisionTreeClassifierSuite
     ))
     val df = TreeTests.setMetadata(data, Map(0 -> 1), 2)
     val dt = new DecisionTreeClassifier().setMaxDepth(3)
-    val model = dt.fit(df)
+    dt.fit(df)
   }
 
   test("Use soft prediction for binary classification with ordered categorical features") {
@@ -335,6 +334,14 @@ class DecisionTreeClassifierSuite
     assert(importances.toArray.forall(_ >= 0.0))
   }
 
+  test("should support all NumericType labels and not support other types") {
+    val dt = new DecisionTreeClassifier().setMaxDepth(1)
+    MLTestingUtils.checkNumericTypes[DecisionTreeClassificationModel, DecisionTreeClassifier](
+      dt, isClassification = true, sqlContext) { (expected, actual) =>
+        TreeTests.checkEqual(expected, actual)
+      }
+  }
+
   /////////////////////////////////////////////////////////////////////////////
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
index bf7481e8a3..76d8c9372e 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
@@ -31,7 +31,6 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.util.Utils
 
-
 /**
  * Test suite for [[GBTClassifier]].
  */
@@ -102,6 +101,14 @@ class GBTClassifierSuite extends SparkFunSuite with MLlibTestSparkContext {
     Utils.deleteRecursively(tempDir)
   }
 
+  test("should support all NumericType labels and not support other types") {
+    val gbt = new GBTClassifier().setMaxDepth(1)
+    MLTestingUtils.checkNumericTypes[GBTClassificationModel, GBTClassifier](
+      gbt, isClassification = true, sqlContext) { (expected, actual) =>
+        TreeTests.checkEqual(expected, actual)
+      }
+  }
+
   // TODO: Reinstate test once runWithValidation is implemented   SPARK-7132
   /*
   test("runWithValidation stops early and performs better on a validation dataset") {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
index afeeaf7fb5..7eefaf2346 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
@@ -103,7 +103,7 @@ class LogisticRegressionSuite
     assert(model.hasSummary)
     // Validate that we re-insert a probability column for evaluation
     val fieldNames = model.summary.predictions.schema.fieldNames
-    assert((dataset.schema.fieldNames.toSet).subsetOf(
+    assert(dataset.schema.fieldNames.toSet.subsetOf(
       fieldNames.toSet))
     assert(fieldNames.exists(s => s.startsWith("probability_")))
   }
@@ -934,6 +934,15 @@ class LogisticRegressionSuite
     testEstimatorAndModelReadWrite(lr, dataset, LogisticRegressionSuite.allParamSettings,
       checkModelData)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val lr = new LogisticRegression().setMaxIter(1)
+    MLTestingUtils.checkNumericTypes[LogisticRegressionModel, LogisticRegression](
+      lr, isClassification = true, sqlContext) { (expected, actual) =>
+        assert(expected.intercept === actual.intercept)
+        assert(expected.coefficients.toArray === actual.coefficients.toArray)
+      }
+  }
 }
 
 object LogisticRegressionSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
index 43781385db..06ff049b48 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
@@ -19,6 +19,7 @@ package org.apache.spark.ml.classification
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
+import org.apache.spark.ml.util.MLTestingUtils
 import org.apache.spark.mllib.classification.LogisticRegressionSuite._
 import org.apache.spark.mllib.classification.LogisticRegressionWithLBFGS
 import org.apache.spark.mllib.evaluation.MulticlassMetrics
@@ -162,4 +163,15 @@ class MultilayerPerceptronClassifierSuite
     assert(newMlpModel.layers === mlpModel.layers)
     assert(newMlpModel.weights === mlpModel.weights)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val layers = Array(3, 2)
+    val mpc = new MultilayerPerceptronClassifier().setLayers(layers).setMaxIter(1)
+    MLTestingUtils.checkNumericTypes[
+        MultilayerPerceptronClassificationModel, MultilayerPerceptronClassifier](
+      mpc, isClassification = true, sqlContext) { (expected, actual) =>
+        assert(expected.layers === actual.layers)
+        assert(expected.weights === actual.weights)
+      }
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala
index 082a6bcd21..4727cd436f 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala
@@ -21,7 +21,7 @@ import breeze.linalg.{Vector => BV}
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.param.ParamsSuite
-import org.apache.spark.ml.util.DefaultReadWriteTest
+import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.classification.NaiveBayes.{Bernoulli, Multinomial}
 import org.apache.spark.mllib.classification.NaiveBayesSuite._
 import org.apache.spark.mllib.linalg._
@@ -86,7 +86,7 @@ class NaiveBayesSuite extends SparkFunSuite with MLlibTestSparkContext with Defa
       model: NaiveBayesModel,
       modelType: String): Unit = {
     featureAndProbabilities.collect().foreach {
-      case Row(features: Vector, probability: Vector) => {
+      case Row(features: Vector, probability: Vector) =>
         assert(probability.toArray.sum ~== 1.0 relTol 1.0e-10)
         val expected = modelType match {
           case Multinomial =>
@@ -97,7 +97,6 @@ class NaiveBayesSuite extends SparkFunSuite with MLlibTestSparkContext with Defa
             throw new UnknownError(s"Invalid modelType: $modelType.")
         }
         assert(probability ~== expected relTol 1.0e-10)
-      }
     }
   }
 
@@ -185,6 +184,15 @@ class NaiveBayesSuite extends SparkFunSuite with MLlibTestSparkContext with Defa
     val nb = new NaiveBayes()
     testEstimatorAndModelReadWrite(nb, dataset, NaiveBayesSuite.allParamSettings, checkModelData)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val nb = new NaiveBayes()
+    MLTestingUtils.checkNumericTypes[NaiveBayesModel, NaiveBayes](
+      nb, isClassification = true, sqlContext) { (expected, actual) =>
+        assert(expected.pi === actual.pi)
+        assert(expected.theta === actual.theta)
+      }
+  }
 }
 
 object NaiveBayesSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
index 51c1baf682..4131396726 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
@@ -74,7 +74,7 @@ class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
     // copied model must have the same parent.
     MLTestingUtils.checkCopy(ovaModel)
 
-    assert(ovaModel.models.size === numClasses)
+    assert(ovaModel.models.length === numClasses)
     val transformedDataset = ovaModel.transform(dataset)
 
     // check for label metadata in prediction col
@@ -224,6 +224,20 @@ class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
     val newOvaModel = testDefaultReadWrite(ovaModel, testParams = false)
     checkModelData(ovaModel, newOvaModel)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val ovr = new OneVsRest().setClassifier(new LogisticRegression().setMaxIter(1))
+    MLTestingUtils.checkNumericTypes[OneVsRestModel, OneVsRest](
+      ovr, isClassification = true, sqlContext) { (expected, actual) =>
+        val expectedModels = expected.models.map(m => m.asInstanceOf[LogisticRegressionModel])
+        val actualModels = actual.models.map(m => m.asInstanceOf[LogisticRegressionModel])
+        assert(expectedModels.length === actualModels.length)
+        expectedModels.zip(actualModels).foreach { case (e, a) =>
+          assert(e.intercept === a.intercept)
+          assert(e.coefficients.toArray === a.coefficients.toArray)
+        }
+      }
+  }
 }
 
 private class MockLogisticRegression(uid: String) extends LogisticRegression(uid) {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala
index b896099e31..052bc83c38 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala
@@ -178,6 +178,14 @@ class RandomForestClassifierSuite extends SparkFunSuite with MLlibTestSparkConte
     assert(importances.toArray.forall(_ >= 0.0))
   }
 
+  test("should support all NumericType labels and not support other types") {
+    val rf = new RandomForestClassifier().setMaxDepth(1)
+    MLTestingUtils.checkNumericTypes[RandomForestClassificationModel, RandomForestClassifier](
+      rf, isClassification = true, sqlContext) { (expected, actual) =>
+        TreeTests.checkEqual(expected, actual)
+      }
+  }
+
   /////////////////////////////////////////////////////////////////////////////
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala
index dbd752d2aa..f4844cc671 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala
@@ -347,6 +347,15 @@ class AFTSurvivalRegressionSuite
     }
   }
 
+  test("should support all NumericType labels") {
+    val aft = new AFTSurvivalRegression().setMaxIter(1)
+    MLTestingUtils.checkNumericTypes[AFTSurvivalRegressionModel, AFTSurvivalRegression](
+      aft, isClassification = false, sqlContext) { (expected, actual) =>
+        assert(expected.intercept === actual.intercept)
+        assert(expected.coefficients === actual.coefficients)
+      }
+  }
+
   test("read/write") {
     def checkModelData(
         model: AFTSurvivalRegressionModel,
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/DecisionTreeRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/DecisionTreeRegressorSuite.scala
index 662e3fc679..e9fb2677b2 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/DecisionTreeRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/DecisionTreeRegressorSuite.scala
@@ -117,6 +117,14 @@ class DecisionTreeRegressorSuite
     assert(importances.toArray.forall(_ >= 0.0))
   }
 
+  test("should support all NumericType labels and not support other types") {
+    val dt = new DecisionTreeRegressor().setMaxDepth(1)
+    MLTestingUtils.checkNumericTypes[DecisionTreeRegressionModel, DecisionTreeRegressor](
+      dt, isClassification = false, sqlContext) { (expected, actual) =>
+        TreeTests.checkEqual(expected, actual)
+      }
+  }
+
   /////////////////////////////////////////////////////////////////////////////
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
index dfb8418086..914818f41f 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
@@ -29,7 +29,6 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.DataFrame
 import org.apache.spark.util.Utils
 
-
 /**
  * Test suite for [[GBTRegressor]].
  */
@@ -110,7 +109,14 @@ class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
 
     sc.checkpointDir = None
     Utils.deleteRecursively(tempDir)
+  }
 
+  test("should support all NumericType labels and not support other types") {
+    val gbt = new GBTRegressor().setMaxDepth(1)
+    MLTestingUtils.checkNumericTypes[GBTRegressionModel, GBTRegressor](
+      gbt, isClassification = false, sqlContext) { (expected, actual) =>
+        TreeTests.checkEqual(expected, actual)
+      }
   }
 
   // TODO: Reinstate test once runWithValidation is implemented  SPARK-7132
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
index 4ebdbf2213..2265464b51 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
@@ -982,6 +982,16 @@ class GeneralizedLinearRegressionSuite
     testEstimatorAndModelReadWrite(glr, datasetPoissonLog,
       GeneralizedLinearRegressionSuite.allParamSettings, checkModelData)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val glr = new GeneralizedLinearRegression().setMaxIter(1)
+    MLTestingUtils.checkNumericTypes[
+        GeneralizedLinearRegressionModel, GeneralizedLinearRegression](
+      glr, isClassification = false, sqlContext) { (expected, actual) =>
+        assert(expected.intercept === actual.intercept)
+        assert(expected.coefficients === actual.coefficients)
+      }
+  }
 }
 
 object GeneralizedLinearRegressionSuite {
@@ -1023,7 +1033,7 @@ object GeneralizedLinearRegressionSuite {
     generator.setSeed(seed)
 
     (0 until nPoints).map { _ =>
-      val features = Vectors.dense(coefficients.indices.map { rndElement(_) }.toArray)
+      val features = Vectors.dense(coefficients.indices.map(rndElement).toArray)
       val eta = BLAS.dot(Vectors.dense(coefficients), features) + intercept
       val mu = link match {
         case "identity" => eta
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/IsotonicRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/IsotonicRegressionSuite.scala
index b8874b4cd3..3a10ad7ed0 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/IsotonicRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/IsotonicRegressionSuite.scala
@@ -180,6 +180,15 @@ class IsotonicRegressionSuite
     testEstimatorAndModelReadWrite(ir, dataset, IsotonicRegressionSuite.allParamSettings,
       checkModelData)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val ir = new IsotonicRegression()
+    MLTestingUtils.checkNumericTypes[IsotonicRegressionModel, IsotonicRegression](
+      ir, isClassification = false, sqlContext) { (expected, actual) =>
+        assert(expected.boundaries === actual.boundaries)
+        assert(expected.predictions === actual.predictions)
+      }
+  }
 }
 
 object IsotonicRegressionSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala
index bd45d21e8d..cccb7f8d1b 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala
@@ -61,9 +61,9 @@ class LinearRegressionSuite
     val featureSize = 4100
     datasetWithSparseFeature = sqlContext.createDataFrame(
       sc.parallelize(LinearDataGenerator.generateLinearInput(
-        intercept = 0.0, weights = Seq.fill(featureSize)(r.nextDouble).toArray,
-        xMean = Seq.fill(featureSize)(r.nextDouble).toArray,
-        xVariance = Seq.fill(featureSize)(r.nextDouble).toArray, nPoints = 200,
+        intercept = 0.0, weights = Seq.fill(featureSize)(r.nextDouble()).toArray,
+        xMean = Seq.fill(featureSize)(r.nextDouble()).toArray,
+        xVariance = Seq.fill(featureSize)(r.nextDouble()).toArray, nPoints = 200,
         seed, eps = 0.1, sparsity = 0.7), 2))
 
     /*
@@ -687,7 +687,7 @@ class LinearRegressionSuite
       // Validate that we re-insert a prediction column for evaluation
       val modelNoPredictionColFieldNames
       = modelNoPredictionCol.summary.predictions.schema.fieldNames
-      assert((datasetWithDenseFeature.schema.fieldNames.toSet).subsetOf(
+      assert(datasetWithDenseFeature.schema.fieldNames.toSet.subsetOf(
         modelNoPredictionColFieldNames.toSet))
       assert(modelNoPredictionColFieldNames.exists(s => s.startsWith("prediction_")))
 
@@ -1006,6 +1006,15 @@ class LinearRegressionSuite
     testEstimatorAndModelReadWrite(lr, datasetWithWeight, LinearRegressionSuite.allParamSettings,
       checkModelData)
   }
+
+  test("should support all NumericType labels and not support other types") {
+    val lr = new LinearRegression().setMaxIter(1)
+    MLTestingUtils.checkNumericTypes[LinearRegressionModel, LinearRegression](
+      lr, isClassification = false, sqlContext) { (expected, actual) =>
+        assert(expected.intercept === actual.intercept)
+        assert(expected.coefficients === actual.coefficients)
+      }
+  }
 }
 
 object LinearRegressionSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala
index 6be0c8bca0..2ab4f1b146 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala
@@ -94,6 +94,14 @@ class RandomForestRegressorSuite extends SparkFunSuite with MLlibTestSparkContex
     assert(importances.toArray.forall(_ >= 0.0))
   }
 
+  test("should support all NumericType labels and not support other types") {
+    val rf = new RandomForestRegressor().setMaxDepth(1)
+    MLTestingUtils.checkNumericTypes[RandomForestRegressionModel, RandomForestRegressor](
+      rf, isClassification = false, sqlContext) { (expected, actual) =>
+        TreeTests.checkEqual(expected, actual)
+      }
+  }
+
   /////////////////////////////////////////////////////////////////////////////
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala
index 12808b0305..bd5bd17147 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala
@@ -73,6 +73,24 @@ private[ml] object TreeTests extends SparkFunSuite {
       numClasses)
   }
 
+  /**
+   * Set label metadata (particularly the number of classes) on a DataFrame.
+   * @param data  Dataset.  Categorical features and labels must already have 0-based indices.
+   *              This must be non-empty.
+   * @param numClasses  Number of classes label can take. If 0, mark as continuous.
+   * @param labelColName  Name of the label column on which to set the metadata.
+   * @return DataFrame with metadata
+   */
+  def setMetadata(data: DataFrame, numClasses: Int, labelColName: String): DataFrame = {
+    val labelAttribute = if (numClasses == 0) {
+      NumericAttribute.defaultAttr.withName(labelColName)
+    } else {
+      NominalAttribute.defaultAttr.withName(labelColName).withNumValues(numClasses)
+    }
+    val labelMetadata = labelAttribute.toMetadata()
+    data.select(data("features"), data(labelColName).as(labelColName, labelMetadata))
+  }
+
   /**
    * Check if the two trees are exactly the same.
    * Note: I hesitate to override Node.equals since it could cause problems if users
diff --git a/mllib/src/test/scala/org/apache/spark/ml/util/MLTestingUtils.scala b/mllib/src/test/scala/org/apache/spark/ml/util/MLTestingUtils.scala
index d290cc9b06..8108460518 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/util/MLTestingUtils.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/util/MLTestingUtils.scala
@@ -17,14 +17,96 @@
 
 package org.apache.spark.ml.util
 
-import org.apache.spark.ml.Model
+import org.apache.spark.SparkFunSuite
+import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.param.ParamMap
+import org.apache.spark.ml.tree.impl.TreeTests
+import org.apache.spark.mllib.linalg.Vectors
+import org.apache.spark.sql.{DataFrame, SQLContext}
+import org.apache.spark.sql.functions._
+import org.apache.spark.sql.types._
 
-object MLTestingUtils {
+object MLTestingUtils extends SparkFunSuite {
   def checkCopy(model: Model[_]): Unit = {
     val copied = model.copy(ParamMap.empty)
       .asInstanceOf[Model[_]]
     assert(copied.parent.uid == model.parent.uid)
     assert(copied.parent == model.parent)
   }
+
+  def checkNumericTypes[M <: Model[M], T <: Estimator[M]](
+      estimator: T,
+      isClassification: Boolean,
+      sqlContext: SQLContext)(check: (M, M) => Unit): Unit = {
+    val dfs = if (isClassification) {
+      genClassifDFWithNumericLabelCol(sqlContext)
+    } else {
+      genRegressionDFWithNumericLabelCol(sqlContext)
+    }
+    val expected = estimator.fit(dfs(DoubleType))
+    val actuals = dfs.keys.filter(_ != DoubleType).map(t => estimator.fit(dfs(t)))
+    actuals.foreach(actual => check(expected, actual))
+
+    val dfWithStringLabels = generateDFWithStringLabelCol(sqlContext)
+    val thrown = intercept[IllegalArgumentException] {
+      estimator.fit(dfWithStringLabels)
+    }
+    assert(thrown.getMessage contains
+      "Column label must be of type NumericType but was actually of type StringType")
+  }
+
+  def genClassifDFWithNumericLabelCol(
+      sqlContext: SQLContext,
+      labelColName: String = "label",
+      featuresColName: String = "features"): Map[NumericType, DataFrame] = {
+    val df = sqlContext.createDataFrame(Seq(
+      (0, Vectors.dense(0, 2, 3)),
+      (1, Vectors.dense(0, 3, 1)),
+      (0, Vectors.dense(0, 2, 2)),
+      (1, Vectors.dense(0, 3, 9)),
+      (0, Vectors.dense(0, 2, 6))
+    )).toDF(labelColName, featuresColName)
+
+    val types =
+      Seq(ShortType, LongType, IntegerType, FloatType, ByteType, DoubleType, DecimalType(10, 0))
+    types.map(t => t -> df.select(col(labelColName).cast(t), col(featuresColName)))
+      .map { case (t, d) => t -> TreeTests.setMetadata(d, 2, labelColName) }
+      .toMap
+  }
+
+  def genRegressionDFWithNumericLabelCol(
+      sqlContext: SQLContext,
+      labelColName: String = "label",
+      featuresColName: String = "features",
+      censorColName: String = "censor"): Map[NumericType, DataFrame] = {
+    val df = sqlContext.createDataFrame(Seq(
+      (0, Vectors.dense(0)),
+      (1, Vectors.dense(1)),
+      (2, Vectors.dense(2)),
+      (3, Vectors.dense(3)),
+      (4, Vectors.dense(4))
+    )).toDF(labelColName, featuresColName)
+
+    val types =
+      Seq(ShortType, LongType, IntegerType, FloatType, ByteType, DoubleType, DecimalType(10, 0))
+    types
+      .map(t => t -> df.select(col(labelColName).cast(t), col(featuresColName)))
+      .map { case (t, d) =>
+        t -> TreeTests.setMetadata(d, 0, labelColName).withColumn(censorColName, lit(0.0))
+      }
+      .toMap
+  }
+
+  def generateDFWithStringLabelCol(
+      sqlContext: SQLContext,
+      labelColName: String = "label",
+      featuresColName: String = "features",
+      censorColName: String = "censor"): DataFrame =
+    sqlContext.createDataFrame(Seq(
+      ("0", Vectors.dense(0, 2, 3), 0.0),
+      ("1", Vectors.dense(0, 3, 1), 1.0),
+      ("0", Vectors.dense(0, 2, 2), 0.0),
+      ("1", Vectors.dense(0, 3, 9), 1.0),
+      ("0", Vectors.dense(0, 2, 6), 0.0)
+    )).toDF(labelColName, featuresColName, censorColName)
 }
-- 
cgit v1.2.3


From 4fc35e6f5c590feb47cbcb5b1136f2e985677b3f Mon Sep 17 00:00:00 2001
From: sethah <seth.hendrickson16@gmail.com>
Date: Fri, 1 Apr 2016 21:23:35 -0700
Subject: [SPARK-14308][ML][MLLIB] Remove unused mllib tree classes and move
 private classes to ML

## What changes were proposed in this pull request?

Decision tree helper classes will be migrated to ML. This patch moves those internal classes that are not part of the public API and removes ones that are no longer used, after [SPARK-12183](https://github.com/apache/spark/pull/11855). No functional changes are made.

Details:
* Bin.scala is removed as the ML implementation does not require bins
* mllib NodeIdCache is removed. It was only used by the mllib implementation previously, which no longer exists
* mllib TreePoint is removed. It was only used by the mllib implementation previously, which no longer exists
* BaggedPoint, DTStatsAggregator, DecisionTreeMetadata, BaggedPointSuite and TimeTracker are all moved to ML.

## How was this patch tested?

No functional changes are made. Existing unit tests ensure behavior is unchanged.

Author: sethah <seth.hendrickson16@gmail.com>

Closes #12097 from sethah/cleanup_mllib_tree.
---
 .../apache/spark/ml/tree/impl/BaggedPoint.scala    | 125 ++++++++++++
 .../spark/ml/tree/impl/DTStatsAggregator.scala     | 181 +++++++++++++++++
 .../spark/ml/tree/impl/DecisionTreeMetadata.scala  | 217 +++++++++++++++++++++
 .../spark/ml/tree/impl/GradientBoostedTrees.scala  |   1 -
 .../apache/spark/ml/tree/impl/NodeIdCache.scala    |   1 -
 .../apache/spark/ml/tree/impl/RandomForest.scala   |   4 +-
 .../apache/spark/ml/tree/impl/TimeTracker.scala    |  70 +++++++
 .../org/apache/spark/ml/tree/impl/TreePoint.scala  |   1 -
 .../spark/mllib/tree/GradientBoostedTrees.scala    |   3 +-
 .../apache/spark/mllib/tree/impl/BaggedPoint.scala | 125 ------------
 .../spark/mllib/tree/impl/DTStatsAggregator.scala  | 178 -----------------
 .../mllib/tree/impl/DecisionTreeMetadata.scala     | 217 ---------------------
 .../apache/spark/mllib/tree/impl/NodeIdCache.scala | 195 ------------------
 .../apache/spark/mllib/tree/impl/TimeTracker.scala |  70 -------
 .../apache/spark/mllib/tree/impl/TreePoint.scala   | 150 --------------
 .../apache/spark/mllib/tree/impurity/Entropy.scala |   2 +-
 .../apache/spark/mllib/tree/impurity/Gini.scala    |   2 +-
 .../spark/mllib/tree/impurity/Variance.scala       |   2 +-
 .../org/apache/spark/mllib/tree/model/Bin.scala    |  47 -----
 .../spark/ml/tree/impl/BaggedPointSuite.scala      |  99 ++++++++++
 .../spark/ml/tree/impl/RandomForestSuite.scala     |   1 -
 .../spark/mllib/tree/DecisionTreeSuite.scala       |   2 +-
 .../spark/mllib/tree/impl/BaggedPointSuite.scala   |  99 ----------
 23 files changed, 699 insertions(+), 1093 deletions(-)
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/tree/impl/BaggedPoint.scala
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/tree/impl/DTStatsAggregator.scala
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/tree/impl/TimeTracker.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/impl/BaggedPoint.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DTStatsAggregator.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TimeTracker.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TreePoint.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/mllib/tree/model/Bin.scala
 create mode 100644 mllib/src/test/scala/org/apache/spark/ml/tree/impl/BaggedPointSuite.scala
 delete mode 100644 mllib/src/test/scala/org/apache/spark/mllib/tree/impl/BaggedPointSuite.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/BaggedPoint.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/BaggedPoint.scala
new file mode 100644
index 0000000000..4e372702f0
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/BaggedPoint.scala
@@ -0,0 +1,125 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.tree.impl
+
+import org.apache.commons.math3.distribution.PoissonDistribution
+
+import org.apache.spark.rdd.RDD
+import org.apache.spark.util.Utils
+import org.apache.spark.util.random.XORShiftRandom
+
+/**
+ * Internal representation of a datapoint which belongs to several subsamples of the same dataset,
+ * particularly for bagging (e.g., for random forests).
+ *
+ * This holds one instance, as well as an array of weights which represent the (weighted)
+ * number of times which this instance appears in each subsamplingRate.
+ * E.g., (datum, [1, 0, 4]) indicates that there are 3 subsamples of the dataset and that
+ * this datum has 1 copy, 0 copies, and 4 copies in the 3 subsamples, respectively.
+ *
+ * @param datum  Data instance
+ * @param subsampleWeights  Weight of this instance in each subsampled dataset.
+ *
+ * TODO: This does not currently support (Double) weighted instances.  Once MLlib has weighted
+ *       dataset support, update.  (We store subsampleWeights as Double for this future extension.)
+ */
+private[spark] class BaggedPoint[Datum](val datum: Datum, val subsampleWeights: Array[Double])
+  extends Serializable
+
+private[spark] object BaggedPoint {
+
+  /**
+   * Convert an input dataset into its BaggedPoint representation,
+   * choosing subsamplingRate counts for each instance.
+   * Each subsamplingRate has the same number of instances as the original dataset,
+   * and is created by subsampling without replacement.
+   * @param input Input dataset.
+   * @param subsamplingRate Fraction of the training data used for learning decision tree.
+   * @param numSubsamples Number of subsamples of this RDD to take.
+   * @param withReplacement Sampling with/without replacement.
+   * @param seed Random seed.
+   * @return BaggedPoint dataset representation.
+   */
+  def convertToBaggedRDD[Datum] (
+      input: RDD[Datum],
+      subsamplingRate: Double,
+      numSubsamples: Int,
+      withReplacement: Boolean,
+      seed: Long = Utils.random.nextLong()): RDD[BaggedPoint[Datum]] = {
+    if (withReplacement) {
+      convertToBaggedRDDSamplingWithReplacement(input, subsamplingRate, numSubsamples, seed)
+    } else {
+      if (numSubsamples == 1 && subsamplingRate == 1.0) {
+        convertToBaggedRDDWithoutSampling(input)
+      } else {
+        convertToBaggedRDDSamplingWithoutReplacement(input, subsamplingRate, numSubsamples, seed)
+      }
+    }
+  }
+
+  private def convertToBaggedRDDSamplingWithoutReplacement[Datum] (
+      input: RDD[Datum],
+      subsamplingRate: Double,
+      numSubsamples: Int,
+      seed: Long): RDD[BaggedPoint[Datum]] = {
+    input.mapPartitionsWithIndex { (partitionIndex, instances) =>
+      // Use random seed = seed + partitionIndex + 1 to make generation reproducible.
+      val rng = new XORShiftRandom
+      rng.setSeed(seed + partitionIndex + 1)
+      instances.map { instance =>
+        val subsampleWeights = new Array[Double](numSubsamples)
+        var subsampleIndex = 0
+        while (subsampleIndex < numSubsamples) {
+          val x = rng.nextDouble()
+          subsampleWeights(subsampleIndex) = {
+            if (x < subsamplingRate) 1.0 else 0.0
+          }
+          subsampleIndex += 1
+        }
+        new BaggedPoint(instance, subsampleWeights)
+      }
+    }
+  }
+
+  private def convertToBaggedRDDSamplingWithReplacement[Datum] (
+      input: RDD[Datum],
+      subsample: Double,
+      numSubsamples: Int,
+      seed: Long): RDD[BaggedPoint[Datum]] = {
+    input.mapPartitionsWithIndex { (partitionIndex, instances) =>
+      // Use random seed = seed + partitionIndex + 1 to make generation reproducible.
+      val poisson = new PoissonDistribution(subsample)
+      poisson.reseedRandomGenerator(seed + partitionIndex + 1)
+      instances.map { instance =>
+        val subsampleWeights = new Array[Double](numSubsamples)
+        var subsampleIndex = 0
+        while (subsampleIndex < numSubsamples) {
+          subsampleWeights(subsampleIndex) = poisson.sample()
+          subsampleIndex += 1
+        }
+        new BaggedPoint(instance, subsampleWeights)
+      }
+    }
+  }
+
+  private def convertToBaggedRDDWithoutSampling[Datum] (
+      input: RDD[Datum]): RDD[BaggedPoint[Datum]] = {
+    input.map(datum => new BaggedPoint(datum, Array(1.0)))
+  }
+
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DTStatsAggregator.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DTStatsAggregator.scala
new file mode 100644
index 0000000000..61091bb803
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DTStatsAggregator.scala
@@ -0,0 +1,181 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.tree.impl
+
+import org.apache.spark.mllib.tree.impurity._
+
+
+
+/**
+ * DecisionTree statistics aggregator for a node.
+ * This holds a flat array of statistics for a set of (features, bins)
+ * and helps with indexing.
+ * This class is abstract to support learning with and without feature subsampling.
+ */
+private[spark] class DTStatsAggregator(
+    val metadata: DecisionTreeMetadata,
+    featureSubset: Option[Array[Int]]) extends Serializable {
+
+  /**
+   * [[ImpurityAggregator]] instance specifying the impurity type.
+   */
+  val impurityAggregator: ImpurityAggregator = metadata.impurity match {
+    case Gini => new GiniAggregator(metadata.numClasses)
+    case Entropy => new EntropyAggregator(metadata.numClasses)
+    case Variance => new VarianceAggregator()
+    case _ => throw new IllegalArgumentException(s"Bad impurity parameter: ${metadata.impurity}")
+  }
+
+  /**
+   * Number of elements (Double values) used for the sufficient statistics of each bin.
+   */
+  private val statsSize: Int = impurityAggregator.statsSize
+
+  /**
+   * Number of bins for each feature.  This is indexed by the feature index.
+   */
+  private val numBins: Array[Int] = {
+    if (featureSubset.isDefined) {
+      featureSubset.get.map(metadata.numBins(_))
+    } else {
+      metadata.numBins
+    }
+  }
+
+  /**
+   * Offset for each feature for calculating indices into the [[allStats]] array.
+   */
+  private val featureOffsets: Array[Int] = {
+    numBins.scanLeft(0)((total, nBins) => total + statsSize * nBins)
+  }
+
+  /**
+   * Total number of elements stored in this aggregator
+   */
+  private val allStatsSize: Int = featureOffsets.last
+
+  /**
+   * Flat array of elements.
+   * Index for start of stats for a (feature, bin) is:
+   *   index = featureOffsets(featureIndex) + binIndex * statsSize
+   */
+  private val allStats: Array[Double] = new Array[Double](allStatsSize)
+
+  /**
+   * Array of parent node sufficient stats.
+   *
+   * Note: this is necessary because stats for the parent node are not available
+   *       on the first iteration of tree learning.
+   */
+  private val parentStats: Array[Double] = new Array[Double](statsSize)
+
+  /**
+   * Get an [[ImpurityCalculator]] for a given (node, feature, bin).
+   *
+   * @param featureOffset  This is a pre-computed (node, feature) offset
+   *                           from [[getFeatureOffset]].
+   */
+  def getImpurityCalculator(featureOffset: Int, binIndex: Int): ImpurityCalculator = {
+    impurityAggregator.getCalculator(allStats, featureOffset + binIndex * statsSize)
+  }
+
+  /**
+   * Get an [[ImpurityCalculator]] for the parent node.
+   */
+  def getParentImpurityCalculator(): ImpurityCalculator = {
+    impurityAggregator.getCalculator(parentStats, 0)
+  }
+
+  /**
+   * Update the stats for a given (feature, bin) for ordered features, using the given label.
+   */
+  def update(featureIndex: Int, binIndex: Int, label: Double, instanceWeight: Double): Unit = {
+    val i = featureOffsets(featureIndex) + binIndex * statsSize
+    impurityAggregator.update(allStats, i, label, instanceWeight)
+  }
+
+  /**
+   * Update the parent node stats using the given label.
+   */
+  def updateParent(label: Double, instanceWeight: Double): Unit = {
+    impurityAggregator.update(parentStats, 0, label, instanceWeight)
+  }
+
+  /**
+   * Faster version of [[update]].
+   * Update the stats for a given (feature, bin), using the given label.
+   *
+   * @param featureOffset  This is a pre-computed feature offset
+   *                           from [[getFeatureOffset]].
+   */
+  def featureUpdate(
+      featureOffset: Int,
+      binIndex: Int,
+      label: Double,
+      instanceWeight: Double): Unit = {
+    impurityAggregator.update(allStats, featureOffset + binIndex * statsSize,
+      label, instanceWeight)
+  }
+
+  /**
+   * Pre-compute feature offset for use with [[featureUpdate]].
+   * For ordered features only.
+   */
+  def getFeatureOffset(featureIndex: Int): Int = featureOffsets(featureIndex)
+
+  /**
+   * For a given feature, merge the stats for two bins.
+   *
+   * @param featureOffset  This is a pre-computed feature offset
+   *                           from [[getFeatureOffset]].
+   * @param binIndex  The other bin is merged into this bin.
+   * @param otherBinIndex  This bin is not modified.
+   */
+  def mergeForFeature(featureOffset: Int, binIndex: Int, otherBinIndex: Int): Unit = {
+    impurityAggregator.merge(allStats, featureOffset + binIndex * statsSize,
+      featureOffset + otherBinIndex * statsSize)
+  }
+
+  /**
+   * Merge this aggregator with another, and returns this aggregator.
+   * This method modifies this aggregator in-place.
+   */
+  def merge(other: DTStatsAggregator): DTStatsAggregator = {
+    require(allStatsSize == other.allStatsSize,
+      s"DTStatsAggregator.merge requires that both aggregators have the same length stats vectors."
+        + s" This aggregator is of length $allStatsSize, but the other is ${other.allStatsSize}.")
+    var i = 0
+    // TODO: Test BLAS.axpy
+    while (i < allStatsSize) {
+      allStats(i) += other.allStats(i)
+      i += 1
+    }
+
+    require(statsSize == other.statsSize,
+      s"DTStatsAggregator.merge requires that both aggregators have the same length parent " +
+        s"stats vectors. This aggregator's parent stats are length $statsSize, " +
+        s"but the other is ${other.statsSize}.")
+    var j = 0
+    while (j < statsSize) {
+      parentStats(j) += other.parentStats(j)
+      j += 1
+    }
+
+    this
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala
new file mode 100644
index 0000000000..df8eb5d1f9
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala
@@ -0,0 +1,217 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.tree.impl
+
+import scala.collection.mutable
+
+import org.apache.spark.internal.Logging
+import org.apache.spark.mllib.regression.LabeledPoint
+import org.apache.spark.mllib.tree.configuration.Algo._
+import org.apache.spark.mllib.tree.configuration.QuantileStrategy._
+import org.apache.spark.mllib.tree.configuration.Strategy
+import org.apache.spark.mllib.tree.impurity.Impurity
+import org.apache.spark.rdd.RDD
+
+/**
+ * Learning and dataset metadata for DecisionTree.
+ *
+ * @param numClasses    For classification: labels can take values {0, ..., numClasses - 1}.
+ *                      For regression: fixed at 0 (no meaning).
+ * @param maxBins  Maximum number of bins, for all features.
+ * @param featureArity  Map: categorical feature index --> arity.
+ *                      I.e., the feature takes values in {0, ..., arity - 1}.
+ * @param numBins  Number of bins for each feature.
+ */
+private[spark] class DecisionTreeMetadata(
+    val numFeatures: Int,
+    val numExamples: Long,
+    val numClasses: Int,
+    val maxBins: Int,
+    val featureArity: Map[Int, Int],
+    val unorderedFeatures: Set[Int],
+    val numBins: Array[Int],
+    val impurity: Impurity,
+    val quantileStrategy: QuantileStrategy,
+    val maxDepth: Int,
+    val minInstancesPerNode: Int,
+    val minInfoGain: Double,
+    val numTrees: Int,
+    val numFeaturesPerNode: Int) extends Serializable {
+
+  def isUnordered(featureIndex: Int): Boolean = unorderedFeatures.contains(featureIndex)
+
+  def isClassification: Boolean = numClasses >= 2
+
+  def isMulticlass: Boolean = numClasses > 2
+
+  def isMulticlassWithCategoricalFeatures: Boolean = isMulticlass && (featureArity.size > 0)
+
+  def isCategorical(featureIndex: Int): Boolean = featureArity.contains(featureIndex)
+
+  def isContinuous(featureIndex: Int): Boolean = !featureArity.contains(featureIndex)
+
+  /**
+   * Number of splits for the given feature.
+   * For unordered features, there is 1 bin per split.
+   * For ordered features, there is 1 more bin than split.
+   */
+  def numSplits(featureIndex: Int): Int = if (isUnordered(featureIndex)) {
+    numBins(featureIndex)
+  } else {
+    numBins(featureIndex) - 1
+  }
+
+
+  /**
+   * Set number of splits for a continuous feature.
+   * For a continuous feature, number of bins is number of splits plus 1.
+   */
+  def setNumSplits(featureIndex: Int, numSplits: Int) {
+    require(isContinuous(featureIndex),
+      s"Only number of bin for a continuous feature can be set.")
+    numBins(featureIndex) = numSplits + 1
+  }
+
+  /**
+   * Indicates if feature subsampling is being used.
+   */
+  def subsamplingFeatures: Boolean = numFeatures != numFeaturesPerNode
+
+}
+
+private[spark] object DecisionTreeMetadata extends Logging {
+
+  /**
+   * Construct a [[DecisionTreeMetadata]] instance for this dataset and parameters.
+   * This computes which categorical features will be ordered vs. unordered,
+   * as well as the number of splits and bins for each feature.
+   */
+  def buildMetadata(
+      input: RDD[LabeledPoint],
+      strategy: Strategy,
+      numTrees: Int,
+      featureSubsetStrategy: String): DecisionTreeMetadata = {
+
+    val numFeatures = input.map(_.features.size).take(1).headOption.getOrElse {
+      throw new IllegalArgumentException(s"DecisionTree requires size of input RDD > 0, " +
+        s"but was given by empty one.")
+    }
+    val numExamples = input.count()
+    val numClasses = strategy.algo match {
+      case Classification => strategy.numClasses
+      case Regression => 0
+    }
+
+    val maxPossibleBins = math.min(strategy.maxBins, numExamples).toInt
+    if (maxPossibleBins < strategy.maxBins) {
+      logWarning(s"DecisionTree reducing maxBins from ${strategy.maxBins} to $maxPossibleBins" +
+        s" (= number of training instances)")
+    }
+
+    // We check the number of bins here against maxPossibleBins.
+    // This needs to be checked here instead of in Strategy since maxPossibleBins can be modified
+    // based on the number of training examples.
+    if (strategy.categoricalFeaturesInfo.nonEmpty) {
+      val maxCategoriesPerFeature = strategy.categoricalFeaturesInfo.values.max
+      val maxCategory =
+        strategy.categoricalFeaturesInfo.find(_._2 == maxCategoriesPerFeature).get._1
+      require(maxCategoriesPerFeature <= maxPossibleBins,
+        s"DecisionTree requires maxBins (= $maxPossibleBins) to be at least as large as the " +
+        s"number of values in each categorical feature, but categorical feature $maxCategory " +
+        s"has $maxCategoriesPerFeature values. Considering remove this and other categorical " +
+        "features with a large number of values, or add more training examples.")
+    }
+
+    val unorderedFeatures = new mutable.HashSet[Int]()
+    val numBins = Array.fill[Int](numFeatures)(maxPossibleBins)
+    if (numClasses > 2) {
+      // Multiclass classification
+      val maxCategoriesForUnorderedFeature =
+        ((math.log(maxPossibleBins / 2 + 1) / math.log(2.0)) + 1).floor.toInt
+      strategy.categoricalFeaturesInfo.foreach { case (featureIndex, numCategories) =>
+        // Hack: If a categorical feature has only 1 category, we treat it as continuous.
+        // TODO(SPARK-9957): Handle this properly by filtering out those features.
+        if (numCategories > 1) {
+          // Decide if some categorical features should be treated as unordered features,
+          //  which require 2 * ((1 << numCategories - 1) - 1) bins.
+          // We do this check with log values to prevent overflows in case numCategories is large.
+          // The next check is equivalent to: 2 * ((1 << numCategories - 1) - 1) <= maxBins
+          if (numCategories <= maxCategoriesForUnorderedFeature) {
+            unorderedFeatures.add(featureIndex)
+            numBins(featureIndex) = numUnorderedBins(numCategories)
+          } else {
+            numBins(featureIndex) = numCategories
+          }
+        }
+      }
+    } else {
+      // Binary classification or regression
+      strategy.categoricalFeaturesInfo.foreach { case (featureIndex, numCategories) =>
+        // If a categorical feature has only 1 category, we treat it as continuous: SPARK-9957
+        if (numCategories > 1) {
+          numBins(featureIndex) = numCategories
+        }
+      }
+    }
+
+    // Set number of features to use per node (for random forests).
+    val _featureSubsetStrategy = featureSubsetStrategy match {
+      case "auto" =>
+        if (numTrees == 1) {
+          "all"
+        } else {
+          if (strategy.algo == Classification) {
+            "sqrt"
+          } else {
+            "onethird"
+          }
+        }
+      case _ => featureSubsetStrategy
+    }
+    val numFeaturesPerNode: Int = _featureSubsetStrategy match {
+      case "all" => numFeatures
+      case "sqrt" => math.sqrt(numFeatures).ceil.toInt
+      case "log2" => math.max(1, (math.log(numFeatures) / math.log(2)).ceil.toInt)
+      case "onethird" => (numFeatures / 3.0).ceil.toInt
+    }
+
+    new DecisionTreeMetadata(numFeatures, numExamples, numClasses, numBins.max,
+      strategy.categoricalFeaturesInfo, unorderedFeatures.toSet, numBins,
+      strategy.impurity, strategy.quantileCalculationStrategy, strategy.maxDepth,
+      strategy.minInstancesPerNode, strategy.minInfoGain, numTrees, numFeaturesPerNode)
+  }
+
+  /**
+   * Version of [[DecisionTreeMetadata#buildMetadata]] for DecisionTree.
+   */
+  def buildMetadata(
+      input: RDD[LabeledPoint],
+      strategy: Strategy): DecisionTreeMetadata = {
+    buildMetadata(input, strategy, numTrees = 1, featureSubsetStrategy = "all")
+  }
+
+    /**
+   * Given the arity of a categorical feature (arity = number of categories),
+   * return the number of bins for the feature if it is to be treated as an unordered feature.
+   * There is 1 split for every partitioning of categories into 2 disjoint, non-empty sets;
+   * there are math.pow(2, arity - 1) - 1 such splits.
+   * Each split has 2 corresponding bins.
+   */
+  def numUnorderedBins(arity: Int): Int = (1 << arity - 1) - 1
+
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
index b37f4e891e..0749d93b7d 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
@@ -25,7 +25,6 @@ import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
 import org.apache.spark.mllib.tree.configuration.{BoostingStrategy => OldBoostingStrategy}
-import org.apache.spark.mllib.tree.impl.TimeTracker
 import org.apache.spark.mllib.tree.impurity.{Variance => OldVariance}
 import org.apache.spark.mllib.tree.loss.{Loss => OldLoss}
 import org.apache.spark.rdd.RDD
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/NodeIdCache.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/NodeIdCache.scala
index 2c8286766f..9d697a36b6 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/NodeIdCache.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/NodeIdCache.scala
@@ -26,7 +26,6 @@ import org.apache.hadoop.fs.{FileSystem, Path}
 import org.apache.spark.annotation.DeveloperApi
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.tree.{LearningNode, Split}
-import org.apache.spark.mllib.tree.impl.BaggedPoint
 import org.apache.spark.rdd.RDD
 import org.apache.spark.storage.StorageLevel
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala
index cccf052b3e..7b1fd089f2 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/RandomForest.scala
@@ -28,8 +28,6 @@ import org.apache.spark.ml.regression.DecisionTreeRegressionModel
 import org.apache.spark.ml.tree._
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, Strategy => OldStrategy}
-import org.apache.spark.mllib.tree.impl.{BaggedPoint, DecisionTreeMetadata, DTStatsAggregator,
-  TimeTracker}
 import org.apache.spark.mllib.tree.impurity.ImpurityCalculator
 import org.apache.spark.mllib.tree.model.ImpurityStats
 import org.apache.spark.rdd.RDD
@@ -330,7 +328,7 @@ private[spark] object RandomForest extends Logging {
   /**
    * Given a group of nodes, this finds the best split for each node.
    *
-   * @param input Training data: RDD of [[org.apache.spark.mllib.tree.impl.TreePoint]]
+   * @param input Training data: RDD of [[org.apache.spark.ml.tree.impl.TreePoint]]
    * @param metadata Learning and dataset metadata
    * @param topNodes Root node for each tree.  Used for matching instances with nodes.
    * @param nodesForGroup Mapping: treeIndex --> nodes to be split in tree
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TimeTracker.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TimeTracker.scala
new file mode 100644
index 0000000000..4cc250aa46
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TimeTracker.scala
@@ -0,0 +1,70 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.tree.impl
+
+import scala.collection.mutable.{HashMap => MutableHashMap}
+
+/**
+ * Time tracker implementation which holds labeled timers.
+ */
+private[spark] class TimeTracker extends Serializable {
+
+  private val starts: MutableHashMap[String, Long] = new MutableHashMap[String, Long]()
+
+  private val totals: MutableHashMap[String, Long] = new MutableHashMap[String, Long]()
+
+  /**
+   * Starts a new timer, or re-starts a stopped timer.
+   */
+  def start(timerLabel: String): Unit = {
+    val currentTime = System.nanoTime()
+    if (starts.contains(timerLabel)) {
+      throw new RuntimeException(s"TimeTracker.start(timerLabel) called again on" +
+        s" timerLabel = $timerLabel before that timer was stopped.")
+    }
+    starts(timerLabel) = currentTime
+  }
+
+  /**
+   * Stops a timer and returns the elapsed time in seconds.
+   */
+  def stop(timerLabel: String): Double = {
+    val currentTime = System.nanoTime()
+    if (!starts.contains(timerLabel)) {
+      throw new RuntimeException(s"TimeTracker.stop(timerLabel) called on" +
+        s" timerLabel = $timerLabel, but that timer was not started.")
+    }
+    val elapsed = currentTime - starts(timerLabel)
+    starts.remove(timerLabel)
+    if (totals.contains(timerLabel)) {
+      totals(timerLabel) += elapsed
+    } else {
+      totals(timerLabel) = elapsed
+    }
+    elapsed / 1e9
+  }
+
+  /**
+   * Print all timing results in seconds.
+   */
+  override def toString: String = {
+    totals.map { case (label, elapsed) =>
+        s"  $label: ${elapsed / 1e9}"
+      }.mkString("\n")
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TreePoint.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TreePoint.scala
index 9fa27e5e1f..3a2bf3c725 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TreePoint.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/TreePoint.scala
@@ -19,7 +19,6 @@ package org.apache.spark.ml.tree.impl
 
 import org.apache.spark.ml.tree.{ContinuousSplit, Split}
 import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.mllib.tree.impl.DecisionTreeMetadata
 import org.apache.spark.rdd.RDD
 
 
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala
index d166dc7905..0f0c6b466d 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala
@@ -20,11 +20,11 @@ package org.apache.spark.mllib.tree
 import org.apache.spark.annotation.Since
 import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.internal.Logging
+import org.apache.spark.ml.tree.impl.TimeTracker
 import org.apache.spark.mllib.impl.PeriodicRDDCheckpointer
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.Algo._
 import org.apache.spark.mllib.tree.configuration.BoostingStrategy
-import org.apache.spark.mllib.tree.impl.TimeTracker
 import org.apache.spark.mllib.tree.impurity.Variance
 import org.apache.spark.mllib.tree.model.{DecisionTreeModel, GradientBoostedTreesModel}
 import org.apache.spark.rdd.RDD
@@ -165,6 +165,7 @@ object GradientBoostedTrees extends Logging {
 
   /**
    * Internal method for performing regression using trees as base learners.
+   *
    * @param input Training dataset.
    * @param validationInput Validation dataset, ignored if validate is set to false.
    * @param boostingStrategy Boosting parameters.
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/BaggedPoint.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/BaggedPoint.scala
deleted file mode 100644
index 572815df0b..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/BaggedPoint.scala
+++ /dev/null
@@ -1,125 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import org.apache.commons.math3.distribution.PoissonDistribution
-
-import org.apache.spark.rdd.RDD
-import org.apache.spark.util.Utils
-import org.apache.spark.util.random.XORShiftRandom
-
-/**
- * Internal representation of a datapoint which belongs to several subsamples of the same dataset,
- * particularly for bagging (e.g., for random forests).
- *
- * This holds one instance, as well as an array of weights which represent the (weighted)
- * number of times which this instance appears in each subsamplingRate.
- * E.g., (datum, [1, 0, 4]) indicates that there are 3 subsamples of the dataset and that
- * this datum has 1 copy, 0 copies, and 4 copies in the 3 subsamples, respectively.
- *
- * @param datum  Data instance
- * @param subsampleWeights  Weight of this instance in each subsampled dataset.
- *
- * TODO: This does not currently support (Double) weighted instances.  Once MLlib has weighted
- *       dataset support, update.  (We store subsampleWeights as Double for this future extension.)
- */
-private[spark] class BaggedPoint[Datum](val datum: Datum, val subsampleWeights: Array[Double])
-  extends Serializable
-
-private[spark] object BaggedPoint {
-
-  /**
-   * Convert an input dataset into its BaggedPoint representation,
-   * choosing subsamplingRate counts for each instance.
-   * Each subsamplingRate has the same number of instances as the original dataset,
-   * and is created by subsampling without replacement.
-   * @param input Input dataset.
-   * @param subsamplingRate Fraction of the training data used for learning decision tree.
-   * @param numSubsamples Number of subsamples of this RDD to take.
-   * @param withReplacement Sampling with/without replacement.
-   * @param seed Random seed.
-   * @return BaggedPoint dataset representation.
-   */
-  def convertToBaggedRDD[Datum] (
-      input: RDD[Datum],
-      subsamplingRate: Double,
-      numSubsamples: Int,
-      withReplacement: Boolean,
-      seed: Long = Utils.random.nextLong()): RDD[BaggedPoint[Datum]] = {
-    if (withReplacement) {
-      convertToBaggedRDDSamplingWithReplacement(input, subsamplingRate, numSubsamples, seed)
-    } else {
-      if (numSubsamples == 1 && subsamplingRate == 1.0) {
-        convertToBaggedRDDWithoutSampling(input)
-      } else {
-        convertToBaggedRDDSamplingWithoutReplacement(input, subsamplingRate, numSubsamples, seed)
-      }
-    }
-  }
-
-  private def convertToBaggedRDDSamplingWithoutReplacement[Datum] (
-      input: RDD[Datum],
-      subsamplingRate: Double,
-      numSubsamples: Int,
-      seed: Long): RDD[BaggedPoint[Datum]] = {
-    input.mapPartitionsWithIndex { (partitionIndex, instances) =>
-      // Use random seed = seed + partitionIndex + 1 to make generation reproducible.
-      val rng = new XORShiftRandom
-      rng.setSeed(seed + partitionIndex + 1)
-      instances.map { instance =>
-        val subsampleWeights = new Array[Double](numSubsamples)
-        var subsampleIndex = 0
-        while (subsampleIndex < numSubsamples) {
-          val x = rng.nextDouble()
-          subsampleWeights(subsampleIndex) = {
-            if (x < subsamplingRate) 1.0 else 0.0
-          }
-          subsampleIndex += 1
-        }
-        new BaggedPoint(instance, subsampleWeights)
-      }
-    }
-  }
-
-  private def convertToBaggedRDDSamplingWithReplacement[Datum] (
-      input: RDD[Datum],
-      subsample: Double,
-      numSubsamples: Int,
-      seed: Long): RDD[BaggedPoint[Datum]] = {
-    input.mapPartitionsWithIndex { (partitionIndex, instances) =>
-      // Use random seed = seed + partitionIndex + 1 to make generation reproducible.
-      val poisson = new PoissonDistribution(subsample)
-      poisson.reseedRandomGenerator(seed + partitionIndex + 1)
-      instances.map { instance =>
-        val subsampleWeights = new Array[Double](numSubsamples)
-        var subsampleIndex = 0
-        while (subsampleIndex < numSubsamples) {
-          subsampleWeights(subsampleIndex) = poisson.sample()
-          subsampleIndex += 1
-        }
-        new BaggedPoint(instance, subsampleWeights)
-      }
-    }
-  }
-
-  private def convertToBaggedRDDWithoutSampling[Datum] (
-      input: RDD[Datum]): RDD[BaggedPoint[Datum]] = {
-    input.map(datum => new BaggedPoint(datum, Array(1.0)))
-  }
-
-}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DTStatsAggregator.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DTStatsAggregator.scala
deleted file mode 100644
index c745e9f8db..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DTStatsAggregator.scala
+++ /dev/null
@@ -1,178 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import org.apache.spark.mllib.tree.impurity._
-
-
-
-/**
- * DecisionTree statistics aggregator for a node.
- * This holds a flat array of statistics for a set of (features, bins)
- * and helps with indexing.
- * This class is abstract to support learning with and without feature subsampling.
- */
-private[spark] class DTStatsAggregator(
-    val metadata: DecisionTreeMetadata,
-    featureSubset: Option[Array[Int]]) extends Serializable {
-
-  /**
-   * [[ImpurityAggregator]] instance specifying the impurity type.
-   */
-  val impurityAggregator: ImpurityAggregator = metadata.impurity match {
-    case Gini => new GiniAggregator(metadata.numClasses)
-    case Entropy => new EntropyAggregator(metadata.numClasses)
-    case Variance => new VarianceAggregator()
-    case _ => throw new IllegalArgumentException(s"Bad impurity parameter: ${metadata.impurity}")
-  }
-
-  /**
-   * Number of elements (Double values) used for the sufficient statistics of each bin.
-   */
-  private val statsSize: Int = impurityAggregator.statsSize
-
-  /**
-   * Number of bins for each feature.  This is indexed by the feature index.
-   */
-  private val numBins: Array[Int] = {
-    if (featureSubset.isDefined) {
-      featureSubset.get.map(metadata.numBins(_))
-    } else {
-      metadata.numBins
-    }
-  }
-
-  /**
-   * Offset for each feature for calculating indices into the [[allStats]] array.
-   */
-  private val featureOffsets: Array[Int] = {
-    numBins.scanLeft(0)((total, nBins) => total + statsSize * nBins)
-  }
-
-  /**
-   * Total number of elements stored in this aggregator
-   */
-  private val allStatsSize: Int = featureOffsets.last
-
-  /**
-   * Flat array of elements.
-   * Index for start of stats for a (feature, bin) is:
-   *   index = featureOffsets(featureIndex) + binIndex * statsSize
-   */
-  private val allStats: Array[Double] = new Array[Double](allStatsSize)
-
-  /**
-   * Array of parent node sufficient stats.
-   *
-   * Note: this is necessary because stats for the parent node are not available
-   *       on the first iteration of tree learning.
-   */
-  private val parentStats: Array[Double] = new Array[Double](statsSize)
-
-  /**
-   * Get an [[ImpurityCalculator]] for a given (node, feature, bin).
-   * @param featureOffset  This is a pre-computed (node, feature) offset
-   *                           from [[getFeatureOffset]].
-   */
-  def getImpurityCalculator(featureOffset: Int, binIndex: Int): ImpurityCalculator = {
-    impurityAggregator.getCalculator(allStats, featureOffset + binIndex * statsSize)
-  }
-
-  /**
-   * Get an [[ImpurityCalculator]] for the parent node.
-   */
-  def getParentImpurityCalculator(): ImpurityCalculator = {
-    impurityAggregator.getCalculator(parentStats, 0)
-  }
-
-  /**
-   * Update the stats for a given (feature, bin) for ordered features, using the given label.
-   */
-  def update(featureIndex: Int, binIndex: Int, label: Double, instanceWeight: Double): Unit = {
-    val i = featureOffsets(featureIndex) + binIndex * statsSize
-    impurityAggregator.update(allStats, i, label, instanceWeight)
-  }
-
-  /**
-   * Update the parent node stats using the given label.
-   */
-  def updateParent(label: Double, instanceWeight: Double): Unit = {
-    impurityAggregator.update(parentStats, 0, label, instanceWeight)
-  }
-
-  /**
-   * Faster version of [[update]].
-   * Update the stats for a given (feature, bin), using the given label.
-   * @param featureOffset  This is a pre-computed feature offset
-   *                           from [[getFeatureOffset]].
-   */
-  def featureUpdate(
-      featureOffset: Int,
-      binIndex: Int,
-      label: Double,
-      instanceWeight: Double): Unit = {
-    impurityAggregator.update(allStats, featureOffset + binIndex * statsSize,
-      label, instanceWeight)
-  }
-
-  /**
-   * Pre-compute feature offset for use with [[featureUpdate]].
-   * For ordered features only.
-   */
-  def getFeatureOffset(featureIndex: Int): Int = featureOffsets(featureIndex)
-
-  /**
-   * For a given feature, merge the stats for two bins.
-   * @param featureOffset  This is a pre-computed feature offset
-   *                           from [[getFeatureOffset]].
-   * @param binIndex  The other bin is merged into this bin.
-   * @param otherBinIndex  This bin is not modified.
-   */
-  def mergeForFeature(featureOffset: Int, binIndex: Int, otherBinIndex: Int): Unit = {
-    impurityAggregator.merge(allStats, featureOffset + binIndex * statsSize,
-      featureOffset + otherBinIndex * statsSize)
-  }
-
-  /**
-   * Merge this aggregator with another, and returns this aggregator.
-   * This method modifies this aggregator in-place.
-   */
-  def merge(other: DTStatsAggregator): DTStatsAggregator = {
-    require(allStatsSize == other.allStatsSize,
-      s"DTStatsAggregator.merge requires that both aggregators have the same length stats vectors."
-        + s" This aggregator is of length $allStatsSize, but the other is ${other.allStatsSize}.")
-    var i = 0
-    // TODO: Test BLAS.axpy
-    while (i < allStatsSize) {
-      allStats(i) += other.allStats(i)
-      i += 1
-    }
-
-    require(statsSize == other.statsSize,
-      s"DTStatsAggregator.merge requires that both aggregators have the same length parent " +
-        s"stats vectors. This aggregator's parent stats are length $statsSize, " +
-        s"but the other is ${other.statsSize}.")
-    var j = 0
-    while (j < statsSize) {
-      parentStats(j) += other.parentStats(j)
-      j += 1
-    }
-
-    this
-  }
-}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala
deleted file mode 100644
index 4f27dc44ef..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala
+++ /dev/null
@@ -1,217 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import scala.collection.mutable
-
-import org.apache.spark.internal.Logging
-import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.mllib.tree.configuration.Algo._
-import org.apache.spark.mllib.tree.configuration.QuantileStrategy._
-import org.apache.spark.mllib.tree.configuration.Strategy
-import org.apache.spark.mllib.tree.impurity.Impurity
-import org.apache.spark.rdd.RDD
-
-/**
- * Learning and dataset metadata for DecisionTree.
- *
- * @param numClasses    For classification: labels can take values {0, ..., numClasses - 1}.
- *                      For regression: fixed at 0 (no meaning).
- * @param maxBins  Maximum number of bins, for all features.
- * @param featureArity  Map: categorical feature index --> arity.
- *                      I.e., the feature takes values in {0, ..., arity - 1}.
- * @param numBins  Number of bins for each feature.
- */
-private[spark] class DecisionTreeMetadata(
-    val numFeatures: Int,
-    val numExamples: Long,
-    val numClasses: Int,
-    val maxBins: Int,
-    val featureArity: Map[Int, Int],
-    val unorderedFeatures: Set[Int],
-    val numBins: Array[Int],
-    val impurity: Impurity,
-    val quantileStrategy: QuantileStrategy,
-    val maxDepth: Int,
-    val minInstancesPerNode: Int,
-    val minInfoGain: Double,
-    val numTrees: Int,
-    val numFeaturesPerNode: Int) extends Serializable {
-
-  def isUnordered(featureIndex: Int): Boolean = unorderedFeatures.contains(featureIndex)
-
-  def isClassification: Boolean = numClasses >= 2
-
-  def isMulticlass: Boolean = numClasses > 2
-
-  def isMulticlassWithCategoricalFeatures: Boolean = isMulticlass && (featureArity.size > 0)
-
-  def isCategorical(featureIndex: Int): Boolean = featureArity.contains(featureIndex)
-
-  def isContinuous(featureIndex: Int): Boolean = !featureArity.contains(featureIndex)
-
-  /**
-   * Number of splits for the given feature.
-   * For unordered features, there is 1 bin per split.
-   * For ordered features, there is 1 more bin than split.
-   */
-  def numSplits(featureIndex: Int): Int = if (isUnordered(featureIndex)) {
-    numBins(featureIndex)
-  } else {
-    numBins(featureIndex) - 1
-  }
-
-
-  /**
-   * Set number of splits for a continuous feature.
-   * For a continuous feature, number of bins is number of splits plus 1.
-   */
-  def setNumSplits(featureIndex: Int, numSplits: Int) {
-    require(isContinuous(featureIndex),
-      s"Only number of bin for a continuous feature can be set.")
-    numBins(featureIndex) = numSplits + 1
-  }
-
-  /**
-   * Indicates if feature subsampling is being used.
-   */
-  def subsamplingFeatures: Boolean = numFeatures != numFeaturesPerNode
-
-}
-
-private[spark] object DecisionTreeMetadata extends Logging {
-
-  /**
-   * Construct a [[DecisionTreeMetadata]] instance for this dataset and parameters.
-   * This computes which categorical features will be ordered vs. unordered,
-   * as well as the number of splits and bins for each feature.
-   */
-  def buildMetadata(
-      input: RDD[LabeledPoint],
-      strategy: Strategy,
-      numTrees: Int,
-      featureSubsetStrategy: String): DecisionTreeMetadata = {
-
-    val numFeatures = input.map(_.features.size).take(1).headOption.getOrElse {
-      throw new IllegalArgumentException(s"DecisionTree requires size of input RDD > 0, " +
-        s"but was given by empty one.")
-    }
-    val numExamples = input.count()
-    val numClasses = strategy.algo match {
-      case Classification => strategy.numClasses
-      case Regression => 0
-    }
-
-    val maxPossibleBins = math.min(strategy.maxBins, numExamples).toInt
-    if (maxPossibleBins < strategy.maxBins) {
-      logWarning(s"DecisionTree reducing maxBins from ${strategy.maxBins} to $maxPossibleBins" +
-        s" (= number of training instances)")
-    }
-
-    // We check the number of bins here against maxPossibleBins.
-    // This needs to be checked here instead of in Strategy since maxPossibleBins can be modified
-    // based on the number of training examples.
-    if (strategy.categoricalFeaturesInfo.nonEmpty) {
-      val maxCategoriesPerFeature = strategy.categoricalFeaturesInfo.values.max
-      val maxCategory =
-        strategy.categoricalFeaturesInfo.find(_._2 == maxCategoriesPerFeature).get._1
-      require(maxCategoriesPerFeature <= maxPossibleBins,
-        s"DecisionTree requires maxBins (= $maxPossibleBins) to be at least as large as the " +
-        s"number of values in each categorical feature, but categorical feature $maxCategory " +
-        s"has $maxCategoriesPerFeature values. Considering remove this and other categorical " +
-        "features with a large number of values, or add more training examples.")
-    }
-
-    val unorderedFeatures = new mutable.HashSet[Int]()
-    val numBins = Array.fill[Int](numFeatures)(maxPossibleBins)
-    if (numClasses > 2) {
-      // Multiclass classification
-      val maxCategoriesForUnorderedFeature =
-        ((math.log(maxPossibleBins / 2 + 1) / math.log(2.0)) + 1).floor.toInt
-      strategy.categoricalFeaturesInfo.foreach { case (featureIndex, numCategories) =>
-        // Hack: If a categorical feature has only 1 category, we treat it as continuous.
-        // TODO(SPARK-9957): Handle this properly by filtering out those features.
-        if (numCategories > 1) {
-          // Decide if some categorical features should be treated as unordered features,
-          //  which require 2 * ((1 << numCategories - 1) - 1) bins.
-          // We do this check with log values to prevent overflows in case numCategories is large.
-          // The next check is equivalent to: 2 * ((1 << numCategories - 1) - 1) <= maxBins
-          if (numCategories <= maxCategoriesForUnorderedFeature) {
-            unorderedFeatures.add(featureIndex)
-            numBins(featureIndex) = numUnorderedBins(numCategories)
-          } else {
-            numBins(featureIndex) = numCategories
-          }
-        }
-      }
-    } else {
-      // Binary classification or regression
-      strategy.categoricalFeaturesInfo.foreach { case (featureIndex, numCategories) =>
-        // If a categorical feature has only 1 category, we treat it as continuous: SPARK-9957
-        if (numCategories > 1) {
-          numBins(featureIndex) = numCategories
-        }
-      }
-    }
-
-    // Set number of features to use per node (for random forests).
-    val _featureSubsetStrategy = featureSubsetStrategy match {
-      case "auto" =>
-        if (numTrees == 1) {
-          "all"
-        } else {
-          if (strategy.algo == Classification) {
-            "sqrt"
-          } else {
-            "onethird"
-          }
-        }
-      case _ => featureSubsetStrategy
-    }
-    val numFeaturesPerNode: Int = _featureSubsetStrategy match {
-      case "all" => numFeatures
-      case "sqrt" => math.sqrt(numFeatures).ceil.toInt
-      case "log2" => math.max(1, (math.log(numFeatures) / math.log(2)).ceil.toInt)
-      case "onethird" => (numFeatures / 3.0).ceil.toInt
-    }
-
-    new DecisionTreeMetadata(numFeatures, numExamples, numClasses, numBins.max,
-      strategy.categoricalFeaturesInfo, unorderedFeatures.toSet, numBins,
-      strategy.impurity, strategy.quantileCalculationStrategy, strategy.maxDepth,
-      strategy.minInstancesPerNode, strategy.minInfoGain, numTrees, numFeaturesPerNode)
-  }
-
-  /**
-   * Version of [[DecisionTreeMetadata#buildMetadata]] for DecisionTree.
-   */
-  def buildMetadata(
-      input: RDD[LabeledPoint],
-      strategy: Strategy): DecisionTreeMetadata = {
-    buildMetadata(input, strategy, numTrees = 1, featureSubsetStrategy = "all")
-  }
-
-    /**
-   * Given the arity of a categorical feature (arity = number of categories),
-   * return the number of bins for the feature if it is to be treated as an unordered feature.
-   * There is 1 split for every partitioning of categories into 2 disjoint, non-empty sets;
-   * there are math.pow(2, arity - 1) - 1 such splits.
-   * Each split has 2 corresponding bins.
-   */
-  def numUnorderedBins(arity: Int): Int = (1 << arity - 1) - 1
-
-}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala
deleted file mode 100644
index dc7e969f7b..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala
+++ /dev/null
@@ -1,195 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import scala.collection.mutable
-
-import org.apache.hadoop.fs.{FileSystem, Path}
-
-import org.apache.spark.annotation.DeveloperApi
-import org.apache.spark.mllib.tree.configuration.FeatureType._
-import org.apache.spark.mllib.tree.model.{Bin, Node, Split}
-import org.apache.spark.rdd.RDD
-import org.apache.spark.storage.StorageLevel
-
-/**
- * :: DeveloperApi ::
- * This is used by the node id cache to find the child id that a data point would belong to.
- * @param split Split information.
- * @param nodeIndex The current node index of a data point that this will update.
- */
-@DeveloperApi
-private[tree] case class NodeIndexUpdater(
-    split: Split,
-    nodeIndex: Int) {
-  /**
-   * Determine a child node index based on the feature value and the split.
-   * @param binnedFeatures Binned feature values.
-   * @param bins Bin information to convert the bin indices to approximate feature values.
-   * @return Child node index to update to.
-   */
-  def updateNodeIndex(binnedFeatures: Array[Int], bins: Array[Array[Bin]]): Int = {
-    if (split.featureType == Continuous) {
-      val featureIndex = split.feature
-      val binIndex = binnedFeatures(featureIndex)
-      val featureValueUpperBound = bins(featureIndex)(binIndex).highSplit.threshold
-      if (featureValueUpperBound <= split.threshold) {
-        Node.leftChildIndex(nodeIndex)
-      } else {
-        Node.rightChildIndex(nodeIndex)
-      }
-    } else {
-      if (split.categories.contains(binnedFeatures(split.feature).toDouble)) {
-        Node.leftChildIndex(nodeIndex)
-      } else {
-        Node.rightChildIndex(nodeIndex)
-      }
-    }
-  }
-}
-
-/**
- * :: DeveloperApi ::
- * A given TreePoint would belong to a particular node per tree.
- * Each row in the nodeIdsForInstances RDD is an array over trees of the node index
- * in each tree. Initially, values should all be 1 for root node.
- * The nodeIdsForInstances RDD needs to be updated at each iteration.
- * @param nodeIdsForInstances The initial values in the cache
- *            (should be an Array of all 1's (meaning the root nodes)).
- * @param checkpointInterval The checkpointing interval
- *                           (how often should the cache be checkpointed.).
- */
-@DeveloperApi
-private[spark] class NodeIdCache(
-  var nodeIdsForInstances: RDD[Array[Int]],
-  val checkpointInterval: Int) {
-
-  // Keep a reference to a previous node Ids for instances.
-  // Because we will keep on re-persisting updated node Ids,
-  // we want to unpersist the previous RDD.
-  private var prevNodeIdsForInstances: RDD[Array[Int]] = null
-
-  // To keep track of the past checkpointed RDDs.
-  private val checkpointQueue = mutable.Queue[RDD[Array[Int]]]()
-  private var rddUpdateCount = 0
-
-  /**
-   * Update the node index values in the cache.
-   * This updates the RDD and its lineage.
-   * TODO: Passing bin information to executors seems unnecessary and costly.
-   * @param data The RDD of training rows.
-   * @param nodeIdUpdaters A map of node index updaters.
-   *                       The key is the indices of nodes that we want to update.
-   * @param bins Bin information needed to find child node indices.
-   */
-  def updateNodeIndices(
-      data: RDD[BaggedPoint[TreePoint]],
-      nodeIdUpdaters: Array[mutable.Map[Int, NodeIndexUpdater]],
-      bins: Array[Array[Bin]]): Unit = {
-    if (prevNodeIdsForInstances != null) {
-      // Unpersist the previous one if one exists.
-      prevNodeIdsForInstances.unpersist()
-    }
-
-    prevNodeIdsForInstances = nodeIdsForInstances
-    nodeIdsForInstances = data.zip(nodeIdsForInstances).map {
-      case (point, node) => {
-        var treeId = 0
-        while (treeId < nodeIdUpdaters.length) {
-          val nodeIdUpdater = nodeIdUpdaters(treeId).getOrElse(node(treeId), null)
-          if (nodeIdUpdater != null) {
-            val newNodeIndex = nodeIdUpdater.updateNodeIndex(
-              binnedFeatures = point.datum.binnedFeatures,
-              bins = bins)
-            node(treeId) = newNodeIndex
-          }
-
-          treeId += 1
-        }
-
-        node
-      }
-    }
-
-    // Keep on persisting new ones.
-    nodeIdsForInstances.persist(StorageLevel.MEMORY_AND_DISK)
-    rddUpdateCount += 1
-
-    // Handle checkpointing if the directory is not None.
-    if (nodeIdsForInstances.sparkContext.getCheckpointDir.nonEmpty &&
-      (rddUpdateCount % checkpointInterval) == 0) {
-      // Let's see if we can delete previous checkpoints.
-      var canDelete = true
-      while (checkpointQueue.size > 1 && canDelete) {
-        // We can delete the oldest checkpoint iff
-        // the next checkpoint actually exists in the file system.
-        if (checkpointQueue.get(1).get.getCheckpointFile.isDefined) {
-          val old = checkpointQueue.dequeue()
-
-          // Since the old checkpoint is not deleted by Spark,
-          // we'll manually delete it here.
-          val fs = FileSystem.get(old.sparkContext.hadoopConfiguration)
-          fs.delete(new Path(old.getCheckpointFile.get), true)
-        } else {
-          canDelete = false
-        }
-      }
-
-      nodeIdsForInstances.checkpoint()
-      checkpointQueue.enqueue(nodeIdsForInstances)
-    }
-  }
-
-  /**
-   * Call this after training is finished to delete any remaining checkpoints.
-   */
-  def deleteAllCheckpoints(): Unit = {
-    while (checkpointQueue.nonEmpty) {
-      val old = checkpointQueue.dequeue()
-      for (checkpointFile <- old.getCheckpointFile) {
-        val fs = FileSystem.get(old.sparkContext.hadoopConfiguration)
-        fs.delete(new Path(checkpointFile), true)
-      }
-    }
-    if (prevNodeIdsForInstances != null) {
-      // Unpersist the previous one if one exists.
-      prevNodeIdsForInstances.unpersist()
-    }
-  }
-}
-
-private[spark] object NodeIdCache {
-  /**
-   * Initialize the node Id cache with initial node Id values.
-   * @param data The RDD of training rows.
-   * @param numTrees The number of trees that we want to create cache for.
-   * @param checkpointInterval The checkpointing interval
-   *                           (how often should the cache be checkpointed.).
-   * @param initVal The initial values in the cache.
-   * @return A node Id cache containing an RDD of initial root node Indices.
-   */
-  def init(
-      data: RDD[BaggedPoint[TreePoint]],
-      numTrees: Int,
-      checkpointInterval: Int,
-      initVal: Int = 1): NodeIdCache = {
-    new NodeIdCache(
-      data.map(_ => Array.fill[Int](numTrees)(initVal)),
-      checkpointInterval)
-  }
-}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TimeTracker.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TimeTracker.scala
deleted file mode 100644
index 70afaa162b..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TimeTracker.scala
+++ /dev/null
@@ -1,70 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import scala.collection.mutable.{HashMap => MutableHashMap}
-
-/**
- * Time tracker implementation which holds labeled timers.
- */
-private[spark] class TimeTracker extends Serializable {
-
-  private val starts: MutableHashMap[String, Long] = new MutableHashMap[String, Long]()
-
-  private val totals: MutableHashMap[String, Long] = new MutableHashMap[String, Long]()
-
-  /**
-   * Starts a new timer, or re-starts a stopped timer.
-   */
-  def start(timerLabel: String): Unit = {
-    val currentTime = System.nanoTime()
-    if (starts.contains(timerLabel)) {
-      throw new RuntimeException(s"TimeTracker.start(timerLabel) called again on" +
-        s" timerLabel = $timerLabel before that timer was stopped.")
-    }
-    starts(timerLabel) = currentTime
-  }
-
-  /**
-   * Stops a timer and returns the elapsed time in seconds.
-   */
-  def stop(timerLabel: String): Double = {
-    val currentTime = System.nanoTime()
-    if (!starts.contains(timerLabel)) {
-      throw new RuntimeException(s"TimeTracker.stop(timerLabel) called on" +
-        s" timerLabel = $timerLabel, but that timer was not started.")
-    }
-    val elapsed = currentTime - starts(timerLabel)
-    starts.remove(timerLabel)
-    if (totals.contains(timerLabel)) {
-      totals(timerLabel) += elapsed
-    } else {
-      totals(timerLabel) = elapsed
-    }
-    elapsed / 1e9
-  }
-
-  /**
-   * Print all timing results in seconds.
-   */
-  override def toString: String = {
-    totals.map { case (label, elapsed) =>
-        s"  $label: ${elapsed / 1e9}"
-      }.mkString("\n")
-  }
-}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TreePoint.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TreePoint.scala
deleted file mode 100644
index 21919d69a3..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/TreePoint.scala
+++ /dev/null
@@ -1,150 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.mllib.tree.model.Bin
-import org.apache.spark.rdd.RDD
-
-
-/**
- * Internal representation of LabeledPoint for DecisionTree.
- * This bins feature values based on a subsampled of data as follows:
- *  (a) Continuous features are binned into ranges.
- *  (b) Unordered categorical features are binned based on subsets of feature values.
- *      "Unordered categorical features" are categorical features with low arity used in
- *      multiclass classification.
- *  (c) Ordered categorical features are binned based on feature values.
- *      "Ordered categorical features" are categorical features with high arity,
- *      or any categorical feature used in regression or binary classification.
- *
- * @param label  Label from LabeledPoint
- * @param binnedFeatures  Binned feature values.
- *                        Same length as LabeledPoint.features, but values are bin indices.
- */
-private[spark] class TreePoint(val label: Double, val binnedFeatures: Array[Int])
-  extends Serializable {
-}
-
-private[spark] object TreePoint {
-
-  /**
-   * Convert an input dataset into its TreePoint representation,
-   * binning feature values in preparation for DecisionTree training.
-   * @param input     Input dataset.
-   * @param bins      Bins for features, of size (numFeatures, numBins).
-   * @param metadata  Learning and dataset metadata
-   * @return  TreePoint dataset representation
-   */
-  def convertToTreeRDD(
-      input: RDD[LabeledPoint],
-      bins: Array[Array[Bin]],
-      metadata: DecisionTreeMetadata): RDD[TreePoint] = {
-    // Construct arrays for featureArity for efficiency in the inner loop.
-    val featureArity: Array[Int] = new Array[Int](metadata.numFeatures)
-    var featureIndex = 0
-    while (featureIndex < metadata.numFeatures) {
-      featureArity(featureIndex) = metadata.featureArity.getOrElse(featureIndex, 0)
-      featureIndex += 1
-    }
-    input.map { x =>
-      TreePoint.labeledPointToTreePoint(x, bins, featureArity)
-    }
-  }
-
-  /**
-   * Convert one LabeledPoint into its TreePoint representation.
-   * @param bins      Bins for features, of size (numFeatures, numBins).
-   * @param featureArity  Array indexed by feature, with value 0 for continuous and numCategories
-   *                      for categorical features.
-   */
-  private def labeledPointToTreePoint(
-      labeledPoint: LabeledPoint,
-      bins: Array[Array[Bin]],
-      featureArity: Array[Int]): TreePoint = {
-    val numFeatures = labeledPoint.features.size
-    val arr = new Array[Int](numFeatures)
-    var featureIndex = 0
-    while (featureIndex < numFeatures) {
-      arr(featureIndex) = findBin(featureIndex, labeledPoint, featureArity(featureIndex),
-        bins)
-      featureIndex += 1
-    }
-    new TreePoint(labeledPoint.label, arr)
-  }
-
-  /**
-   * Find bin for one (labeledPoint, feature).
-   *
-   * @param featureArity  0 for continuous features; number of categories for categorical features.
-   * @param bins   Bins for features, of size (numFeatures, numBins).
-   */
-  private def findBin(
-      featureIndex: Int,
-      labeledPoint: LabeledPoint,
-      featureArity: Int,
-      bins: Array[Array[Bin]]): Int = {
-
-    /**
-     * Binary search helper method for continuous feature.
-     */
-    def binarySearchForBins(): Int = {
-      val binForFeatures = bins(featureIndex)
-      val feature = labeledPoint.features(featureIndex)
-      var left = 0
-      var right = binForFeatures.length - 1
-      while (left <= right) {
-        val mid = left + (right - left) / 2
-        val bin = binForFeatures(mid)
-        val lowThreshold = bin.lowSplit.threshold
-        val highThreshold = bin.highSplit.threshold
-        if ((lowThreshold < feature) && (highThreshold >= feature)) {
-          return mid
-        } else if (lowThreshold >= feature) {
-          right = mid - 1
-        } else {
-          left = mid + 1
-        }
-      }
-      -1
-    }
-
-    if (featureArity == 0) {
-      // Perform binary search for finding bin for continuous features.
-      val binIndex = binarySearchForBins()
-      if (binIndex == -1) {
-        throw new RuntimeException("No bin was found for continuous feature." +
-          " This error can occur when given invalid data values (such as NaN)." +
-          s" Feature index: $featureIndex.  Feature value: ${labeledPoint.features(featureIndex)}")
-      }
-      binIndex
-    } else {
-      // Categorical feature bins are indexed by feature values.
-      val featureValue = labeledPoint.features(featureIndex)
-      if (featureValue < 0 || featureValue >= featureArity) {
-        throw new IllegalArgumentException(
-          s"DecisionTree given invalid data:" +
-            s" Feature $featureIndex is categorical with values in" +
-            s" {0,...,${featureArity - 1}," +
-            s" but a data point gives it value $featureValue.\n" +
-            "  Bad data point: " + labeledPoint.toString)
-      }
-      featureValue.toInt
-    }
-  }
-}
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Entropy.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Entropy.scala
index 13aff11007..ff7700d2d1 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Entropy.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Entropy.scala
@@ -85,7 +85,7 @@ object Entropy extends Impurity {
  * Note: Instances of this class do not hold the data; they operate on views of the data.
  * @param numClasses  Number of classes for label.
  */
-private[tree] class EntropyAggregator(numClasses: Int)
+private[spark] class EntropyAggregator(numClasses: Int)
   extends ImpurityAggregator(numClasses) with Serializable {
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Gini.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Gini.scala
index 39c7f9c3be..58dc79b739 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Gini.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Gini.scala
@@ -81,7 +81,7 @@ object Gini extends Impurity {
  * Note: Instances of this class do not hold the data; they operate on views of the data.
  * @param numClasses  Number of classes for label.
  */
-private[tree] class GiniAggregator(numClasses: Int)
+private[spark] class GiniAggregator(numClasses: Int)
   extends ImpurityAggregator(numClasses) with Serializable {
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Variance.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Variance.scala
index 92d74a1b83..2423516123 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Variance.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Variance.scala
@@ -71,7 +71,7 @@ object Variance extends Impurity {
  * in order to compute impurity from a sample.
  * Note: Instances of this class do not hold the data; they operate on views of the data.
  */
-private[tree] class VarianceAggregator()
+private[spark] class VarianceAggregator()
   extends ImpurityAggregator(statsSize = 3) with Serializable {
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Bin.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Bin.scala
deleted file mode 100644
index 0cad473782..0000000000
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/model/Bin.scala
+++ /dev/null
@@ -1,47 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.model
-
-import org.apache.spark.mllib.tree.configuration.FeatureType._
-
-/**
- * Used for "binning" the feature values for faster best split calculation.
- *
- * For a continuous feature, the bin is determined by a low and a high split,
- *  where an example with featureValue falls into the bin s.t.
- *  lowSplit.threshold < featureValue <= highSplit.threshold.
- *
- * For ordered categorical features, there is a 1-1-1 correspondence between
- *  bins, splits, and feature values.  The bin is determined by category/feature value.
- *  However, the bins are not necessarily ordered by feature value;
- *  they are ordered using impurity.
- *
- * For unordered categorical features, there is a 1-1 correspondence between bins, splits,
- *  where bins and splits correspond to subsets of feature values (in highSplit.categories).
- *  An unordered feature with k categories uses (1 << k - 1) - 1 bins, corresponding to all
- *  partitionings of categories into 2 disjoint, non-empty sets.
- *
- * @param lowSplit signifying the lower threshold for the continuous feature to be
- *                 accepted in the bin
- * @param highSplit signifying the upper threshold for the continuous feature to be
- *                  accepted in the bin
- * @param featureType type of feature -- categorical or continuous
- * @param category categorical label value accepted in the bin for ordered features
- */
-private[tree]
-case class Bin(lowSplit: Split, highSplit: Split, featureType: FeatureType, category: Double)
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/BaggedPointSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/BaggedPointSuite.scala
new file mode 100644
index 0000000000..77ab3d8bb7
--- /dev/null
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/BaggedPointSuite.scala
@@ -0,0 +1,99 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.tree.impl
+
+import org.apache.spark.SparkFunSuite
+import org.apache.spark.mllib.tree.EnsembleTestHelper
+import org.apache.spark.mllib.util.MLlibTestSparkContext
+
+/**
+ * Test suite for [[BaggedPoint]].
+ */
+class BaggedPointSuite extends SparkFunSuite with MLlibTestSparkContext  {
+
+  test("BaggedPoint RDD: without subsampling") {
+    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
+    val rdd = sc.parallelize(arr)
+    val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, 1.0, 1, false, 42)
+    baggedRDD.collect().foreach { baggedPoint =>
+      assert(baggedPoint.subsampleWeights.size == 1 && baggedPoint.subsampleWeights(0) == 1)
+    }
+  }
+
+  test("BaggedPoint RDD: with subsampling with replacement (fraction = 1.0)") {
+    val numSubsamples = 100
+    val (expectedMean, expectedStddev) = (1.0, 1.0)
+
+    val seeds = Array(123, 5354, 230, 349867, 23987)
+    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
+    val rdd = sc.parallelize(arr)
+    seeds.foreach { seed =>
+      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, 1.0, numSubsamples, true, seed)
+      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
+      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
+        expectedStddev, epsilon = 0.01)
+    }
+  }
+
+  test("BaggedPoint RDD: with subsampling with replacement (fraction = 0.5)") {
+    val numSubsamples = 100
+    val subsample = 0.5
+    val (expectedMean, expectedStddev) = (subsample, math.sqrt(subsample))
+
+    val seeds = Array(123, 5354, 230, 349867, 23987)
+    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
+    val rdd = sc.parallelize(arr)
+    seeds.foreach { seed =>
+      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, subsample, numSubsamples, true, seed)
+      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
+      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
+        expectedStddev, epsilon = 0.01)
+    }
+  }
+
+  test("BaggedPoint RDD: with subsampling without replacement (fraction = 1.0)") {
+    val numSubsamples = 100
+    val (expectedMean, expectedStddev) = (1.0, 0)
+
+    val seeds = Array(123, 5354, 230, 349867, 23987)
+    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
+    val rdd = sc.parallelize(arr)
+    seeds.foreach { seed =>
+      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, 1.0, numSubsamples, false, seed)
+      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
+      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
+        expectedStddev, epsilon = 0.01)
+    }
+  }
+
+  test("BaggedPoint RDD: with subsampling without replacement (fraction = 0.5)") {
+    val numSubsamples = 100
+    val subsample = 0.5
+    val (expectedMean, expectedStddev) = (subsample, math.sqrt(subsample * (1 - subsample)))
+
+    val seeds = Array(123, 5354, 230, 349867, 23987)
+    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
+    val rdd = sc.parallelize(arr)
+    seeds.foreach { seed =>
+      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, subsample, numSubsamples, false, seed)
+      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
+      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
+        expectedStddev, epsilon = 0.01)
+    }
+  }
+}
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
index 441338e74e..e64551f03c 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
@@ -26,7 +26,6 @@ import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.{DecisionTreeSuite => OldDTSuite, EnsembleTestHelper}
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, QuantileStrategy, Strategy => OldStrategy}
-import org.apache.spark.mllib.tree.impl.{BaggedPoint, DecisionTreeMetadata}
 import org.apache.spark.mllib.tree.impurity.{Entropy, Gini, GiniCalculator}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala
index bb1041b109..49cb7e1f24 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala
@@ -20,12 +20,12 @@ package org.apache.spark.mllib.tree
 import scala.collection.JavaConverters._
 
 import org.apache.spark.SparkFunSuite
+import org.apache.spark.ml.tree.impl.DecisionTreeMetadata
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.Algo._
 import org.apache.spark.mllib.tree.configuration.FeatureType._
 import org.apache.spark.mllib.tree.configuration.Strategy
-import org.apache.spark.mllib.tree.impl.DecisionTreeMetadata
 import org.apache.spark.mllib.tree.impurity.{Entropy, Gini, Variance}
 import org.apache.spark.mllib.tree.model._
 import org.apache.spark.mllib.util.MLlibTestSparkContext
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/impl/BaggedPointSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/impl/BaggedPointSuite.scala
deleted file mode 100644
index 9d756da410..0000000000
--- a/mllib/src/test/scala/org/apache/spark/mllib/tree/impl/BaggedPointSuite.scala
+++ /dev/null
@@ -1,99 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.mllib.tree.impl
-
-import org.apache.spark.SparkFunSuite
-import org.apache.spark.mllib.tree.EnsembleTestHelper
-import org.apache.spark.mllib.util.MLlibTestSparkContext
-
-/**
- * Test suite for [[BaggedPoint]].
- */
-class BaggedPointSuite extends SparkFunSuite with MLlibTestSparkContext  {
-
-  test("BaggedPoint RDD: without subsampling") {
-    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
-    val rdd = sc.parallelize(arr)
-    val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, 1.0, 1, false, 42)
-    baggedRDD.collect().foreach { baggedPoint =>
-      assert(baggedPoint.subsampleWeights.size == 1 && baggedPoint.subsampleWeights(0) == 1)
-    }
-  }
-
-  test("BaggedPoint RDD: with subsampling with replacement (fraction = 1.0)") {
-    val numSubsamples = 100
-    val (expectedMean, expectedStddev) = (1.0, 1.0)
-
-    val seeds = Array(123, 5354, 230, 349867, 23987)
-    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
-    val rdd = sc.parallelize(arr)
-    seeds.foreach { seed =>
-      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, 1.0, numSubsamples, true, seed)
-      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
-      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
-        expectedStddev, epsilon = 0.01)
-    }
-  }
-
-  test("BaggedPoint RDD: with subsampling with replacement (fraction = 0.5)") {
-    val numSubsamples = 100
-    val subsample = 0.5
-    val (expectedMean, expectedStddev) = (subsample, math.sqrt(subsample))
-
-    val seeds = Array(123, 5354, 230, 349867, 23987)
-    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
-    val rdd = sc.parallelize(arr)
-    seeds.foreach { seed =>
-      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, subsample, numSubsamples, true, seed)
-      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
-      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
-        expectedStddev, epsilon = 0.01)
-    }
-  }
-
-  test("BaggedPoint RDD: with subsampling without replacement (fraction = 1.0)") {
-    val numSubsamples = 100
-    val (expectedMean, expectedStddev) = (1.0, 0)
-
-    val seeds = Array(123, 5354, 230, 349867, 23987)
-    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
-    val rdd = sc.parallelize(arr)
-    seeds.foreach { seed =>
-      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, 1.0, numSubsamples, false, seed)
-      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
-      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
-        expectedStddev, epsilon = 0.01)
-    }
-  }
-
-  test("BaggedPoint RDD: with subsampling without replacement (fraction = 0.5)") {
-    val numSubsamples = 100
-    val subsample = 0.5
-    val (expectedMean, expectedStddev) = (subsample, math.sqrt(subsample * (1 - subsample)))
-
-    val seeds = Array(123, 5354, 230, 349867, 23987)
-    val arr = EnsembleTestHelper.generateOrderedLabeledPoints(1, 1000)
-    val rdd = sc.parallelize(arr)
-    seeds.foreach { seed =>
-      val baggedRDD = BaggedPoint.convertToBaggedRDD(rdd, subsample, numSubsamples, false, seed)
-      val subsampleCounts: Array[Array[Double]] = baggedRDD.map(_.subsampleWeights).collect()
-      EnsembleTestHelper.testRandomArrays(subsampleCounts, numSubsamples, expectedMean,
-        expectedStddev, epsilon = 0.01)
-    }
-  }
-}
-- 
cgit v1.2.3


From 06694f1c68cb752ea311144f0dbe50e92e1393cf Mon Sep 17 00:00:00 2001
From: Jacek Laskowski <jacek@japila.pl>
Date: Sat, 2 Apr 2016 08:12:04 -0700
Subject: [MINOR] Typo fixes

## What changes were proposed in this pull request?

Typo fixes. No functional changes.

## How was this patch tested?

Built the sources and ran with samples.

Author: Jacek Laskowski <jacek@japila.pl>

Closes #11802 from jaceklaskowski/typo-fixes.
---
 .../streaming/RecoverableNetworkWordCount.scala    |  2 +-
 .../main/scala/org/apache/spark/ml/Pipeline.scala  |  2 +-
 .../spark/ml/regression/LinearRegression.scala     |  2 +-
 .../sql/catalyst/plans/logical/LogicalPlan.scala   |  4 ++--
 .../org/apache/spark/sql/ExperimentalMethods.scala |  2 +-
 .../sql/execution/joins/BroadcastHashJoin.scala    |  2 +-
 .../scala/org/apache/spark/sql/functions.scala     | 12 ++++++------
 .../apache/spark/streaming/StreamingContext.scala  | 13 +++++++------
 .../streaming/dstream/ConstantInputDStream.scala   |  2 +-
 .../apache/spark/streaming/dstream/DStream.scala   |  8 ++++----
 .../streaming/dstream/DStreamCheckpointData.scala  |  6 +++---
 .../spark/streaming/dstream/InputDStream.scala     |  6 +++---
 .../streaming/dstream/ReducedWindowedDStream.scala |  2 +-
 .../spark/streaming/dstream/StateDStream.scala     | 12 ++++++------
 .../streaming/scheduler/ReceivedBlockTracker.scala |  4 ++--
 .../streaming/scheduler/rate/RateEstimator.scala   | 22 ++++++++++++----------
 16 files changed, 52 insertions(+), 49 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/examples/src/main/scala/org/apache/spark/examples/streaming/RecoverableNetworkWordCount.scala b/examples/src/main/scala/org/apache/spark/examples/streaming/RecoverableNetworkWordCount.scala
index 05f8e65d65..b6b8bc33f7 100644
--- a/examples/src/main/scala/org/apache/spark/examples/streaming/RecoverableNetworkWordCount.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/streaming/RecoverableNetworkWordCount.scala
@@ -141,7 +141,7 @@ object RecoverableNetworkWordCount {
 
   def main(args: Array[String]) {
     if (args.length != 4) {
-      System.err.println("You arguments were " + args.mkString("[", ", ", "]"))
+      System.err.println("Your arguments were " + args.mkString("[", ", ", "]"))
       System.err.println(
         """
           |Usage: RecoverableNetworkWordCount <hostname> <port> <checkpoint-directory>
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala b/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
index 3a99979a88..afefaaa883 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
@@ -147,7 +147,7 @@ class Pipeline @Since("1.4.0") (
             t
           case _ =>
             throw new IllegalArgumentException(
-              s"Do not support stage $stage of type ${stage.getClass}")
+              s"Does not support stage $stage of type ${stage.getClass}")
         }
         if (index < indexOfLastEstimator) {
           curDataset = transformer.transform(curDataset)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
index ba5ad4c072..2633c06f40 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
@@ -58,7 +58,7 @@ private[regression] trait LinearRegressionParams extends PredictorParams
  * The specific squared error loss function used is:
  *   L = 1/2n ||A coefficients - y||^2^
  *
- * This support multiple types of regularization:
+ * This supports multiple types of regularization:
  *  - none (a.k.a. ordinary least squares)
  *  - L2 (ridge regression)
  *  - L1 (Lasso)
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala
index ecf4285c46..aceeb8aadc 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala
@@ -79,13 +79,13 @@ abstract class LogicalPlan extends QueryPlan[LogicalPlan] with Logging {
 
   /**
    * Computes [[Statistics]] for this plan. The default implementation assumes the output
-   * cardinality is the product of of all child plan's cardinality, i.e. applies in the case
+   * cardinality is the product of all child plan's cardinality, i.e. applies in the case
    * of cartesian joins.
    *
    * [[LeafNode]]s must override this.
    */
   def statistics: Statistics = {
-    if (children.size == 0) {
+    if (children.isEmpty) {
       throw new UnsupportedOperationException(s"LeafNode $nodeName must implement statistics.")
     }
     Statistics(sizeInBytes = children.map(_.statistics.sizeInBytes).product)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/ExperimentalMethods.scala b/sql/core/src/main/scala/org/apache/spark/sql/ExperimentalMethods.scala
index d7cd84fd24..c5df028485 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/ExperimentalMethods.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/ExperimentalMethods.scala
@@ -37,7 +37,7 @@ class ExperimentalMethods private[sql]() {
 
   /**
    * Allows extra strategies to be injected into the query planner at runtime.  Note this API
-   * should be consider experimental and is not intended to be stable across releases.
+   * should be considered experimental and is not intended to be stable across releases.
    *
    * @since 1.3.0
    */
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
index f5b083c216..0ed1ed41b0 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
@@ -32,7 +32,7 @@ import org.apache.spark.util.collection.CompactBuffer
 /**
  * Performs an inner hash join of two child relations.  When the output RDD of this operator is
  * being constructed, a Spark job is asynchronously started to calculate the values for the
- * broadcasted relation.  This data is then placed in a Spark broadcast variable.  The streamed
+ * broadcast relation.  This data is then placed in a Spark broadcast variable.  The streamed
  * relation is not shuffled.
  */
 case class BroadcastHashJoin(
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
index 74906050ac..baf947d037 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
@@ -2232,7 +2232,7 @@ object functions {
 
   /**
    * Splits str around pattern (pattern is a regular expression).
-   * NOTE: pattern is a string represent the regular expression.
+   * NOTE: pattern is a string representation of the regular expression.
    *
    * @group string_funcs
    * @since 1.5.0
@@ -2267,9 +2267,9 @@ object functions {
 
   /**
    * Translate any character in the src by a character in replaceString.
-   * The characters in replaceString is corresponding to the characters in matchingString.
-   * The translate will happen when any character in the string matching with the character
-   * in the matchingString.
+   * The characters in replaceString correspond to the characters in matchingString.
+   * The translate will happen when any character in the string matches the character
+   * in the `matchingString`.
    *
    * @group string_funcs
    * @since 1.5.0
@@ -2692,7 +2692,7 @@ object functions {
   //////////////////////////////////////////////////////////////////////////////////////////////
 
   /**
-   * Returns true if the array contain the value
+   * Returns true if the array contains `value`
    * @group collection_funcs
    * @since 1.5.0
    */
@@ -2920,7 +2920,7 @@ object functions {
 
   /**
    * Defines a user-defined function (UDF) using a Scala closure. For this variant, the caller must
-   * specifcy the output data type, and there is no automatic input type coercion.
+   * specify the output data type, and there is no automatic input type coercion.
    *
    * @param f  A closure in Scala
    * @param dataType  The output data type of the UDF
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala b/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala
index 3a664c4f5c..c1e151d08b 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala
@@ -132,7 +132,7 @@ class StreamingContext private[streaming] (
       "both SparkContext and checkpoint as null")
   }
 
-  private[streaming] val isCheckpointPresent = (_cp != null)
+  private[streaming] val isCheckpointPresent: Boolean = _cp != null
 
   private[streaming] val sc: SparkContext = {
     if (_sc != null) {
@@ -213,8 +213,8 @@ class StreamingContext private[streaming] (
   def sparkContext: SparkContext = sc
 
   /**
-   * Set each DStreams in this context to remember RDDs it generated in the last given duration.
-   * DStreams remember RDDs only for a limited duration of time and releases them for garbage
+   * Set each DStream in this context to remember RDDs it generated in the last given duration.
+   * DStreams remember RDDs only for a limited duration of time and release them for garbage
    * collection. This method allows the developer to specify how long to remember the RDDs (
    * if the developer wishes to query old data outside the DStream computation).
    * @param duration Minimum duration that each DStream should remember its RDDs
@@ -282,13 +282,14 @@ class StreamingContext private[streaming] (
   }
 
   /**
-   * Create a input stream from TCP source hostname:port. Data is received using
+   * Creates an input stream from TCP source hostname:port. Data is received using
    * a TCP socket and the receive bytes is interpreted as UTF8 encoded `\n` delimited
    * lines.
    * @param hostname      Hostname to connect to for receiving data
    * @param port          Port to connect to for receiving data
    * @param storageLevel  Storage level to use for storing the received objects
    *                      (default: StorageLevel.MEMORY_AND_DISK_SER_2)
+   * @see [[socketStream]]
    */
   def socketTextStream(
       hostname: String,
@@ -299,7 +300,7 @@ class StreamingContext private[streaming] (
   }
 
   /**
-   * Create a input stream from TCP source hostname:port. Data is received using
+   * Creates an input stream from TCP source hostname:port. Data is received using
    * a TCP socket and the receive bytes it interpreted as object using the given
    * converter.
    * @param hostname      Hostname to connect to for receiving data
@@ -860,7 +861,7 @@ private class StreamingContextPythonHelper {
    */
   def tryRecoverFromCheckpoint(checkpointPath: String): Option[StreamingContext] = {
     val checkpointOption = CheckpointReader.read(
-      checkpointPath, new SparkConf(), SparkHadoopUtil.get.conf, false)
+      checkpointPath, new SparkConf(), SparkHadoopUtil.get.conf, ignoreReadError = false)
     checkpointOption.map(new StreamingContext(null, _, null))
   }
 }
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/ConstantInputDStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/ConstantInputDStream.scala
index b5f86fe779..995470ec8d 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/ConstantInputDStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/ConstantInputDStream.scala
@@ -23,7 +23,7 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.streaming.{StreamingContext, Time}
 
 /**
- * An input stream that always returns the same RDD on each timestep. Useful for testing.
+ * An input stream that always returns the same RDD on each time step. Useful for testing.
  */
 class ConstantInputDStream[T: ClassTag](_ssc: StreamingContext, rdd: RDD[T])
   extends InputDStream[T](_ssc) {
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala
index eb7b64eaf4..c40beeff97 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala
@@ -83,7 +83,7 @@ abstract class DStream[T: ClassTag] (
 
   // RDDs generated, marked as private[streaming] so that testsuites can access it
   @transient
-  private[streaming] var generatedRDDs = new HashMap[Time, RDD[T]] ()
+  private[streaming] var generatedRDDs = new HashMap[Time, RDD[T]]()
 
   // Time zero for the DStream
   private[streaming] var zeroTime: Time = null
@@ -269,7 +269,7 @@ abstract class DStream[T: ClassTag] (
       checkpointDuration == null || rememberDuration > checkpointDuration,
       s"The remember duration for ${this.getClass.getSimpleName} has been set to " +
         s" $rememberDuration which is not more than the checkpoint interval" +
-        s" ($checkpointDuration). Please set it to higher than $checkpointDuration."
+        s" ($checkpointDuration). Please set it to a value higher than $checkpointDuration."
     )
 
     dependencies.foreach(_.validateAtStart())
@@ -277,7 +277,7 @@ abstract class DStream[T: ClassTag] (
     logInfo(s"Slide time = $slideDuration")
     logInfo(s"Storage level = ${storageLevel.description}")
     logInfo(s"Checkpoint interval = $checkpointDuration")
-    logInfo(s"Remember duration = $rememberDuration")
+    logInfo(s"Remember interval = $rememberDuration")
     logInfo(s"Initialized and validated $this")
   }
 
@@ -535,7 +535,7 @@ abstract class DStream[T: ClassTag] (
   private def readObject(ois: ObjectInputStream): Unit = Utils.tryOrIOException {
     logDebug(s"${this.getClass().getSimpleName}.readObject used")
     ois.defaultReadObject()
-    generatedRDDs = new HashMap[Time, RDD[T]] ()
+    generatedRDDs = new HashMap[Time, RDD[T]]()
   }
 
   // =======================================================================
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala
index 365a6bc417..431c9dbe2c 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala
@@ -29,7 +29,7 @@ import org.apache.spark.streaming.Time
 import org.apache.spark.util.Utils
 
 private[streaming]
-class DStreamCheckpointData[T: ClassTag] (dstream: DStream[T])
+class DStreamCheckpointData[T: ClassTag](dstream: DStream[T])
   extends Serializable with Logging {
   protected val data = new HashMap[Time, AnyRef]()
 
@@ -45,7 +45,7 @@ class DStreamCheckpointData[T: ClassTag] (dstream: DStream[T])
   /**
    * Updates the checkpoint data of the DStream. This gets called every time
    * the graph checkpoint is initiated. Default implementation records the
-   * checkpoint files to which the generate RDDs of the DStream has been saved.
+   * checkpoint files at which the generated RDDs of the DStream have been saved.
    */
   def update(time: Time) {
 
@@ -103,7 +103,7 @@ class DStreamCheckpointData[T: ClassTag] (dstream: DStream[T])
 
   /**
    * Restore the checkpoint data. This gets called once when the DStream graph
-   * (along with its DStreams) are being restored from a graph checkpoint file.
+   * (along with its output DStreams) is being restored from a graph checkpoint file.
    * Default implementation restores the RDDs from their checkpoint files.
    */
   def restore() {
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/InputDStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/InputDStream.scala
index 0b6b191dbe..dc88349db5 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/InputDStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/InputDStream.scala
@@ -39,7 +39,7 @@ import org.apache.spark.util.Utils
  *
  * @param _ssc Streaming context that will execute this input stream
  */
-abstract class InputDStream[T: ClassTag] (_ssc: StreamingContext)
+abstract class InputDStream[T: ClassTag](_ssc: StreamingContext)
   extends DStream[T](_ssc) {
 
   private[streaming] var lastValidTime: Time = null
@@ -90,8 +90,8 @@ abstract class InputDStream[T: ClassTag] (_ssc: StreamingContext)
     } else {
       // Time is valid, but check it it is more than lastValidTime
       if (lastValidTime != null && time < lastValidTime) {
-        logWarning("isTimeValid called with " + time + " where as last valid time is " +
-          lastValidTime)
+        logWarning(s"isTimeValid called with $time whereas the last valid time " +
+          s"is $lastValidTime")
       }
       lastValidTime = time
       true
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/ReducedWindowedDStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/ReducedWindowedDStream.scala
index a9be2f213f..a9e93838b8 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/ReducedWindowedDStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/ReducedWindowedDStream.scala
@@ -87,7 +87,7 @@ class ReducedWindowedDStream[K: ClassTag, V: ClassTag](
 
     logDebug("Window time = " + windowDuration)
     logDebug("Slide time = " + slideDuration)
-    logDebug("ZeroTime = " + zeroTime)
+    logDebug("Zero time = " + zeroTime)
     logDebug("Current window = " + currentWindow)
     logDebug("Previous window = " + previousWindow)
 
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala
index 68eff89030..0379957e58 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala
@@ -70,7 +70,7 @@ class StateDStream[K: ClassTag, V: ClassTag, S: ClassTag](
         // Try to get the parent RDD
         parent.getOrCompute(validTime) match {
           case Some(parentRDD) => {   // If parent RDD exists, then compute as usual
-            computeUsingPreviousRDD (parentRDD, prevStateRDD)
+            computeUsingPreviousRDD(parentRDD, prevStateRDD)
           }
           case None => {    // If parent RDD does not exist
 
@@ -98,15 +98,15 @@ class StateDStream[K: ClassTag, V: ClassTag, S: ClassTag](
                 // and then apply the update function
                 val updateFuncLocal = updateFunc
                 val finalFunc = (iterator: Iterator[(K, Iterable[V])]) => {
-                  updateFuncLocal (iterator.map (tuple => (tuple._1, tuple._2.toSeq, None)))
+                  updateFuncLocal(iterator.map(tuple => (tuple._1, tuple._2.toSeq, None)))
                 }
 
-                val groupedRDD = parentRDD.groupByKey (partitioner)
-                val sessionRDD = groupedRDD.mapPartitions (finalFunc, preservePartitioning)
+                val groupedRDD = parentRDD.groupByKey(partitioner)
+                val sessionRDD = groupedRDD.mapPartitions(finalFunc, preservePartitioning)
                 // logDebug("Generating state RDD for time " + validTime + " (first)")
-                Some (sessionRDD)
+                Some(sessionRDD)
               }
-              case Some (initialStateRDD) => {
+              case Some(initialStateRDD) => {
                 computeUsingPreviousRDD(parentRDD, initialStateRDD)
               }
             }
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/scheduler/ReceivedBlockTracker.scala b/streaming/src/main/scala/org/apache/spark/streaming/scheduler/ReceivedBlockTracker.scala
index 9c8e68b03d..5d9a8ac0d9 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/scheduler/ReceivedBlockTracker.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/scheduler/ReceivedBlockTracker.scala
@@ -119,7 +119,7 @@ private[streaming] class ReceivedBlockTracker(
         timeToAllocatedBlocks.put(batchTime, allocatedBlocks)
         lastAllocatedBatchTime = batchTime
       } else {
-        logInfo(s"Possibly processed batch $batchTime need to be processed again in WAL recovery")
+        logInfo(s"Possibly processed batch $batchTime needs to be processed again in WAL recovery")
       }
     } else {
       // This situation occurs when:
@@ -129,7 +129,7 @@ private[streaming] class ReceivedBlockTracker(
       // 2. Slow checkpointing makes recovered batch time older than WAL recovered
       // lastAllocatedBatchTime.
       // This situation will only occurs in recovery time.
-      logInfo(s"Possibly processed batch $batchTime need to be processed again in WAL recovery")
+      logInfo(s"Possibly processed batch $batchTime needs to be processed again in WAL recovery")
     }
   }
 
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/scheduler/rate/RateEstimator.scala b/streaming/src/main/scala/org/apache/spark/streaming/scheduler/rate/RateEstimator.scala
index d7210f64fc..7b2ef6881d 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/scheduler/rate/RateEstimator.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/scheduler/rate/RateEstimator.scala
@@ -21,18 +21,20 @@ import org.apache.spark.SparkConf
 import org.apache.spark.streaming.Duration
 
 /**
- * A component that estimates the rate at wich an InputDStream should ingest
- * elements, based on updates at every batch completion.
+ * A component that estimates the rate at which an `InputDStream` should ingest
+ * records, based on updates at every batch completion.
+ *
+ * @see [[org.apache.spark.streaming.scheduler.RateController]]
  */
 private[streaming] trait RateEstimator extends Serializable {
 
   /**
-   * Computes the number of elements the stream attached to this `RateEstimator`
+   * Computes the number of records the stream attached to this `RateEstimator`
    * should ingest per second, given an update on the size and completion
    * times of the latest batch.
    *
-   * @param time The timetamp of the current batch interval that just finished
-   * @param elements The number of elements that were processed in this batch
+   * @param time The timestamp of the current batch interval that just finished
+   * @param elements The number of records that were processed in this batch
    * @param processingDelay The time in ms that took for the job to complete
    * @param schedulingDelay The time in ms that the job spent in the scheduling queue
    */
@@ -46,13 +48,13 @@ private[streaming] trait RateEstimator extends Serializable {
 object RateEstimator {
 
   /**
-   * Return a new RateEstimator based on the value of `spark.streaming.RateEstimator`.
+   * Return a new `RateEstimator` based on the value of
+   * `spark.streaming.backpressure.rateEstimator`.
    *
-   * The only known estimator right now is `pid`.
+   * The only known and acceptable estimator right now is `pid`.
    *
    * @return An instance of RateEstimator
-   * @throws IllegalArgumentException if there is a configured RateEstimator that doesn't match any
-   *         known estimators.
+   * @throws IllegalArgumentException if the configured RateEstimator is not `pid`.
    */
   def create(conf: SparkConf, batchInterval: Duration): RateEstimator =
     conf.get("spark.streaming.backpressure.rateEstimator", "pid") match {
@@ -64,6 +66,6 @@ object RateEstimator {
         new PIDRateEstimator(batchInterval.milliseconds, proportional, integral, derived, minRate)
 
       case estimator =>
-        throw new IllegalArgumentException(s"Unkown rate estimator: $estimator")
+        throw new IllegalArgumentException(s"Unknown rate estimator: $estimator")
     }
 }
-- 
cgit v1.2.3


From 4a6e78abd9d5edc4a5092738dff0006bbe202a89 Mon Sep 17 00:00:00 2001
From: Dongjoon Hyun <dongjoon@apache.org>
Date: Sat, 2 Apr 2016 17:50:40 -0700
Subject: [MINOR][DOCS] Use multi-line JavaDoc comments in Scala code.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## What changes were proposed in this pull request?

This PR aims to fix all Scala-Style multiline comments into Java-Style multiline comments in Scala codes.
(All comment-only changes over 77 files: +786 lines, −747 lines)

## How was this patch tested?

Manual.

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #12130 from dongjoon-hyun/use_multiine_javadoc_comments.
---
 .../main/scala/org/apache/spark/FutureAction.scala |  14 +-
 .../main/scala/org/apache/spark/SSLOptions.scala   |  57 +++---
 .../main/scala/org/apache/spark/SparkContext.scala |  42 +++--
 .../org/apache/spark/api/java/JavaPairRDD.scala    |   8 +-
 .../apache/spark/api/java/JavaSparkContext.scala   |  60 ++++---
 .../apache/spark/deploy/worker/CommandUtils.scala  |   2 +-
 .../scala/org/apache/spark/rdd/CoGroupedRDD.scala  |  10 +-
 core/src/main/scala/org/apache/spark/rdd/RDD.scala |   4 +-
 .../mesos/CoarseMesosSchedulerBackend.scala        |  24 +--
 .../cluster/mesos/MesosClusterScheduler.scala      |  12 +-
 .../cluster/mesos/MesosSchedulerUtils.scala        |   4 +-
 .../org/apache/spark/shuffle/ShuffleManager.scala  |   6 +-
 .../apache/spark/storage/memory/MemoryStore.scala  |  20 +--
 .../main/scala/org/apache/spark/util/Utils.scala   |  17 +-
 core/src/test/scala/org/apache/spark/Smuggle.scala |  46 ++---
 .../apache/spark/memory/MemoryManagerSuite.scala   |  24 +--
 .../org/apache/spark/examples/BroadcastTest.scala  |   4 +-
 .../apache/spark/examples/DFSReadWriteTest.scala   |  20 +--
 .../org/apache/spark/examples/GroupByTest.scala    |   4 +-
 .../apache/spark/examples/MultiBroadcastTest.scala |   4 +-
 .../spark/examples/SimpleSkewedGroupByTest.scala   |   4 +-
 .../apache/spark/examples/SkewedGroupByTest.scala  |   4 +-
 .../streaming/clickstream/PageViewGenerator.scala  |  23 +--
 .../streaming/clickstream/PageViewStream.scala     |  21 +--
 .../spark/streaming/flume/FlumeInputDStream.scala  |  15 +-
 .../spark/streaming/kafka/KafkaRDDPartition.scala  |  15 +-
 .../scala/org/apache/spark/graphx/GraphOps.scala   |  10 +-
 .../spark/graphx/lib/ConnectedComponents.scala     |  18 +-
 .../spark/ml/feature/ElementwiseProduct.scala      |   6 +-
 .../api/python/GaussianMixtureModelWrapper.scala   |   8 +-
 .../mllib/api/python/Word2VecModelWrapper.scala    |   4 +-
 .../org/apache/spark/mllib/linalg/Matrices.scala   |  16 +-
 .../StreamingLinearRegressionWithSGD.scala         |   4 +-
 .../scala/org/apache/spark/repl/SparkILoop.scala   |  21 ++-
 .../scala/org/apache/spark/repl/SparkImports.scala |   5 +-
 .../main/scala/org/apache/spark/sql/Encoder.scala  |  24 +--
 .../spark/sql/catalyst/analysis/Analyzer.scala     |  20 +--
 .../sql/catalyst/expressions/Projection.scala      |   6 +-
 .../expressions/codegen/CodeGenerator.scala        |  26 +--
 .../spark/sql/catalyst/expressions/grouping.scala  |  18 +-
 .../spark/sql/catalyst/expressions/misc.scala      |   4 +-
 .../spark/sql/catalyst/optimizer/Optimizer.scala   |  40 ++---
 .../spark/sql/catalyst/parser/AstBuilder.scala     |   4 +-
 .../spark/sql/catalyst/planning/patterns.scala     |  28 +--
 .../spark/sql/catalyst/plans/QueryPlan.scala       |   4 +-
 .../sql/catalyst/plans/physical/partitioning.scala |   6 +-
 .../optimizer/OptimizerExtendableSuite.scala       |  14 +-
 .../scala/org/apache/spark/sql/SQLContext.scala    |  14 +-
 .../apache/spark/sql/execution/CacheManager.scala  |   7 +-
 .../org/apache/spark/sql/execution/SparkPlan.scala |   4 +-
 .../spark/sql/execution/WholeStageCodegen.scala    | 172 +++++++++----------
 .../org/apache/spark/sql/execution/Window.scala    |  36 ++--
 .../execution/aggregate/AggregationIterator.scala  |  22 +--
 .../aggregate/SortBasedAggregationIterator.scala   |   6 +-
 .../execution/aggregate/TungstenAggregate.scala    |  16 +-
 .../execution/datasources/SqlNewHadoopRDD.scala    |   8 +-
 .../execution/datasources/csv/CSVInferSchema.scala |  22 +--
 .../execution/datasources/csv/DefaultSource.scala  |   4 +-
 .../spark/sql/execution/datasources/ddl.scala      |   8 +-
 .../sql/execution/joins/BroadcastHashJoin.scala    |  22 +--
 .../sql/execution/joins/CartesianProduct.scala     |   8 +-
 .../spark/sql/execution/joins/HashJoin.scala       |   8 +-
 .../spark/sql/execution/joins/HashedRelation.scala |  76 ++++----
 .../spark/sql/execution/joins/SortMergeJoin.scala  |  36 ++--
 .../state/HDFSBackedStateStoreProvider.scala       |   8 +-
 .../spark/sql/execution/ui/SparkPlanGraph.scala    |   4 +-
 .../scala/org/apache/spark/sql/functions.scala     | 191 ++++++++++-----------
 .../org/apache/spark/sql/sources/interfaces.scala  |  26 +--
 .../scala/org/apache/spark/sql/QueryTest.scala     |   4 +-
 .../sql/execution/BenchmarkWholeStageCodegen.scala |   8 +-
 .../execution/datasources/csv/CSVParserSuite.scala |   4 +-
 .../apache/spark/streaming/StreamingContext.scala  |   7 +-
 .../streaming/api/java/JavaStreamingContext.scala  |   7 +-
 .../spark/streaming/receiver/RateLimiter.scala     |  23 +--
 .../apache/spark/streaming/scheduler/JobSet.scala  |   7 +-
 .../apache/spark/tools/GenerateMIMAIgnore.scala    |   9 +-
 .../spark/deploy/yarn/YarnSparkHadoopUtil.scala    |  16 +-
 77 files changed, 786 insertions(+), 747 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/core/src/main/scala/org/apache/spark/FutureAction.scala b/core/src/main/scala/org/apache/spark/FutureAction.scala
index 2a8220ff40..ce11772a6d 100644
--- a/core/src/main/scala/org/apache/spark/FutureAction.scala
+++ b/core/src/main/scala/org/apache/spark/FutureAction.scala
@@ -146,16 +146,16 @@ class SimpleFutureAction[T] private[spark](jobWaiter: JobWaiter[_], resultFunc:
 
 
 /**
-  * Handle via which a "run" function passed to a [[ComplexFutureAction]]
-  * can submit jobs for execution.
-  */
+ * Handle via which a "run" function passed to a [[ComplexFutureAction]]
+ * can submit jobs for execution.
+ */
 @DeveloperApi
 trait JobSubmitter {
   /**
-    * Submit a job for execution and return a FutureAction holding the result.
-    * This is a wrapper around the same functionality provided by SparkContext
-    * to enable cancellation.
-    */
+   * Submit a job for execution and return a FutureAction holding the result.
+   * This is a wrapper around the same functionality provided by SparkContext
+   * to enable cancellation.
+   */
   def submitJob[T, U, R](
     rdd: RDD[T],
     processPartition: Iterator[T] => U,
diff --git a/core/src/main/scala/org/apache/spark/SSLOptions.scala b/core/src/main/scala/org/apache/spark/SSLOptions.scala
index 30db6ccbf4..719905a2c9 100644
--- a/core/src/main/scala/org/apache/spark/SSLOptions.scala
+++ b/core/src/main/scala/org/apache/spark/SSLOptions.scala
@@ -132,34 +132,35 @@ private[spark] case class SSLOptions(
 
 private[spark] object SSLOptions extends Logging {
 
-  /** Resolves SSLOptions settings from a given Spark configuration object at a given namespace.
-    *
-    * The following settings are allowed:
-    * $ - `[ns].enabled` - `true` or `false`, to enable or disable SSL respectively
-    * $ - `[ns].keyStore` - a path to the key-store file; can be relative to the current directory
-    * $ - `[ns].keyStorePassword` - a password to the key-store file
-    * $ - `[ns].keyPassword` - a password to the private key
-    * $ - `[ns].keyStoreType` - the type of the key-store
-    * $ - `[ns].needClientAuth` - whether SSL needs client authentication
-    * $ - `[ns].trustStore` - a path to the trust-store file; can be relative to the current
-    *                         directory
-    * $ - `[ns].trustStorePassword` - a password to the trust-store file
-    * $ - `[ns].trustStoreType` - the type of trust-store
-    * $ - `[ns].protocol` - a protocol name supported by a particular Java version
-    * $ - `[ns].enabledAlgorithms` - a comma separated list of ciphers
-    *
-    * For a list of protocols and ciphers supported by particular Java versions, you may go to
-    * [[https://blogs.oracle.com/java-platform-group/entry/diagnosing_tls_ssl_and_https Oracle
-    * blog page]].
-    *
-    * You can optionally specify the default configuration. If you do, for each setting which is
-    * missing in SparkConf, the corresponding setting is used from the default configuration.
-    *
-    * @param conf Spark configuration object where the settings are collected from
-    * @param ns the namespace name
-    * @param defaults the default configuration
-    * @return [[org.apache.spark.SSLOptions]] object
-    */
+  /**
+   * Resolves SSLOptions settings from a given Spark configuration object at a given namespace.
+   *
+   * The following settings are allowed:
+   * $ - `[ns].enabled` - `true` or `false`, to enable or disable SSL respectively
+   * $ - `[ns].keyStore` - a path to the key-store file; can be relative to the current directory
+   * $ - `[ns].keyStorePassword` - a password to the key-store file
+   * $ - `[ns].keyPassword` - a password to the private key
+   * $ - `[ns].keyStoreType` - the type of the key-store
+   * $ - `[ns].needClientAuth` - whether SSL needs client authentication
+   * $ - `[ns].trustStore` - a path to the trust-store file; can be relative to the current
+   *                         directory
+   * $ - `[ns].trustStorePassword` - a password to the trust-store file
+   * $ - `[ns].trustStoreType` - the type of trust-store
+   * $ - `[ns].protocol` - a protocol name supported by a particular Java version
+   * $ - `[ns].enabledAlgorithms` - a comma separated list of ciphers
+   *
+   * For a list of protocols and ciphers supported by particular Java versions, you may go to
+   * [[https://blogs.oracle.com/java-platform-group/entry/diagnosing_tls_ssl_and_https Oracle
+   * blog page]].
+   *
+   * You can optionally specify the default configuration. If you do, for each setting which is
+   * missing in SparkConf, the corresponding setting is used from the default configuration.
+   *
+   * @param conf Spark configuration object where the settings are collected from
+   * @param ns the namespace name
+   * @param defaults the default configuration
+   * @return [[org.apache.spark.SSLOptions]] object
+   */
   def parse(conf: SparkConf, ns: String, defaults: Option[SSLOptions] = None): SSLOptions = {
     val enabled = conf.getBoolean(s"$ns.enabled", defaultValue = defaults.exists(_.enabled))
 
diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala
index d7cb253d69..4b3264cbf5 100644
--- a/core/src/main/scala/org/apache/spark/SparkContext.scala
+++ b/core/src/main/scala/org/apache/spark/SparkContext.scala
@@ -773,9 +773,11 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
     parallelize(seq, numSlices)
   }
 
-  /** Distribute a local Scala collection to form an RDD, with one or more
-    * location preferences (hostnames of Spark nodes) for each object.
-    * Create a new partition for each collection item. */
+  /**
+   * Distribute a local Scala collection to form an RDD, with one or more
+   * location preferences (hostnames of Spark nodes) for each object.
+   * Create a new partition for each collection item.
+   */
   def makeRDD[T: ClassTag](seq: Seq[(T, Seq[String])]): RDD[T] = withScope {
     assertNotStopped()
     val indexToPrefs = seq.zipWithIndex.map(t => (t._2, t._1._2)).toMap
@@ -1095,14 +1097,15 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
     new NewHadoopRDD(this, fClass, kClass, vClass, jconf)
   }
 
-  /** Get an RDD for a Hadoop SequenceFile with given key and value types.
-    *
-    * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
-    * record, directly caching the returned RDD or directly passing it to an aggregation or shuffle
-    * operation will create many references to the same object.
-    * If you plan to directly cache, sort, or aggregate Hadoop writable objects, you should first
-    * copy them using a `map` function.
-    */
+  /**
+   * Get an RDD for a Hadoop SequenceFile with given key and value types.
+   *
+   * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
+   * record, directly caching the returned RDD or directly passing it to an aggregation or shuffle
+   * operation will create many references to the same object.
+   * If you plan to directly cache, sort, or aggregate Hadoop writable objects, you should first
+   * copy them using a `map` function.
+   */
   def sequenceFile[K, V](path: String,
       keyClass: Class[K],
       valueClass: Class[V],
@@ -1113,14 +1116,15 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
     hadoopFile(path, inputFormatClass, keyClass, valueClass, minPartitions)
   }
 
-  /** Get an RDD for a Hadoop SequenceFile with given key and value types.
-    *
-    * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
-    * record, directly caching the returned RDD or directly passing it to an aggregation or shuffle
-    * operation will create many references to the same object.
-    * If you plan to directly cache, sort, or aggregate Hadoop writable objects, you should first
-    * copy them using a `map` function.
-    * */
+  /**
+   * Get an RDD for a Hadoop SequenceFile with given key and value types.
+   *
+   * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
+   * record, directly caching the returned RDD or directly passing it to an aggregation or shuffle
+   * operation will create many references to the same object.
+   * If you plan to directly cache, sort, or aggregate Hadoop writable objects, you should first
+   * copy them using a `map` function.
+   */
   def sequenceFile[K, V](
       path: String,
       keyClass: Class[K],
diff --git a/core/src/main/scala/org/apache/spark/api/java/JavaPairRDD.scala b/core/src/main/scala/org/apache/spark/api/java/JavaPairRDD.scala
index e080f91f50..2897272a8b 100644
--- a/core/src/main/scala/org/apache/spark/api/java/JavaPairRDD.scala
+++ b/core/src/main/scala/org/apache/spark/api/java/JavaPairRDD.scala
@@ -461,10 +461,10 @@ class JavaPairRDD[K, V](val rdd: RDD[(K, V)])
     fromRDD(rdd.partitionBy(partitioner))
 
   /**
-    * Return an RDD containing all pairs of elements with matching keys in `this` and `other`. Each
-    * pair of elements will be returned as a (k, (v1, v2)) tuple, where (k, v1) is in `this` and
-    * (k, v2) is in `other`. Uses the given Partitioner to partition the output RDD.
-    */
+   * Return an RDD containing all pairs of elements with matching keys in `this` and `other`. Each
+   * pair of elements will be returned as a (k, (v1, v2)) tuple, where (k, v1) is in `this` and
+   * (k, v2) is in `other`. Uses the given Partitioner to partition the output RDD.
+   */
   def join[W](other: JavaPairRDD[K, W], partitioner: Partitioner): JavaPairRDD[K, (V, W)] =
     fromRDD(rdd.join(other, partitioner))
 
diff --git a/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala b/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala
index d362c40b7a..dfd91ae338 100644
--- a/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala
+++ b/core/src/main/scala/org/apache/spark/api/java/JavaSparkContext.scala
@@ -295,13 +295,14 @@ class JavaSparkContext(val sc: SparkContext)
     new JavaRDD(sc.binaryRecords(path, recordLength))
   }
 
-  /** Get an RDD for a Hadoop SequenceFile with given key and value types.
-    *
-    * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
-    * record, directly caching the returned RDD will create many references to the same object.
-    * If you plan to directly cache Hadoop writable objects, you should first copy them using
-    * a `map` function.
-    * */
+  /**
+   * Get an RDD for a Hadoop SequenceFile with given key and value types.
+   *
+   * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
+   * record, directly caching the returned RDD will create many references to the same object.
+   * If you plan to directly cache Hadoop writable objects, you should first copy them using
+   * a `map` function.
+   */
   def sequenceFile[K, V](path: String,
     keyClass: Class[K],
     valueClass: Class[V],
@@ -312,13 +313,14 @@ class JavaSparkContext(val sc: SparkContext)
     new JavaPairRDD(sc.sequenceFile(path, keyClass, valueClass, minPartitions))
   }
 
-  /** Get an RDD for a Hadoop SequenceFile.
-    *
-    * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
-    * record, directly caching the returned RDD will create many references to the same object.
-    * If you plan to directly cache Hadoop writable objects, you should first copy them using
-    * a `map` function.
-    */
+  /**
+   * Get an RDD for a Hadoop SequenceFile.
+   *
+   * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
+   * record, directly caching the returned RDD will create many references to the same object.
+   * If you plan to directly cache Hadoop writable objects, you should first copy them using
+   * a `map` function.
+   */
   def sequenceFile[K, V](path: String, keyClass: Class[K], valueClass: Class[V]):
   JavaPairRDD[K, V] = {
     implicit val ctagK: ClassTag[K] = ClassTag(keyClass)
@@ -411,13 +413,14 @@ class JavaSparkContext(val sc: SparkContext)
     new JavaHadoopRDD(rdd.asInstanceOf[HadoopRDD[K, V]])
   }
 
-  /** Get an RDD for a Hadoop file with an arbitrary InputFormat.
-    *
-    * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
-    * record, directly caching the returned RDD will create many references to the same object.
-    * If you plan to directly cache Hadoop writable objects, you should first copy them using
-    * a `map` function.
-    */
+  /**
+   * Get an RDD for a Hadoop file with an arbitrary InputFormat.
+   *
+   * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
+   * record, directly caching the returned RDD will create many references to the same object.
+   * If you plan to directly cache Hadoop writable objects, you should first copy them using
+   * a `map` function.
+   */
   def hadoopFile[K, V, F <: InputFormat[K, V]](
     path: String,
     inputFormatClass: Class[F],
@@ -431,13 +434,14 @@ class JavaSparkContext(val sc: SparkContext)
     new JavaHadoopRDD(rdd.asInstanceOf[HadoopRDD[K, V]])
   }
 
-  /** Get an RDD for a Hadoop file with an arbitrary InputFormat
-    *
-    * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
-    * record, directly caching the returned RDD will create many references to the same object.
-    * If you plan to directly cache Hadoop writable objects, you should first copy them using
-    * a `map` function.
-    */
+  /**
+   * Get an RDD for a Hadoop file with an arbitrary InputFormat
+   *
+   * '''Note:''' Because Hadoop's RecordReader class re-uses the same Writable object for each
+   * record, directly caching the returned RDD will create many references to the same object.
+   * If you plan to directly cache Hadoop writable objects, you should first copy them using
+   * a `map` function.
+   */
   def hadoopFile[K, V, F <: InputFormat[K, V]](
     path: String,
     inputFormatClass: Class[F],
diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala b/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala
index a4efafcb27..cba4aaffe2 100644
--- a/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/worker/CommandUtils.scala
@@ -29,7 +29,7 @@ import org.apache.spark.launcher.WorkerCommandBuilder
 import org.apache.spark.util.Utils
 
 /**
- ** Utilities for running commands with the spark classpath.
+ * Utilities for running commands with the spark classpath.
  */
 private[deploy]
 object CommandUtils extends Logging {
diff --git a/core/src/main/scala/org/apache/spark/rdd/CoGroupedRDD.scala b/core/src/main/scala/org/apache/spark/rdd/CoGroupedRDD.scala
index e5ebc63082..7bc1eb0436 100644
--- a/core/src/main/scala/org/apache/spark/rdd/CoGroupedRDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/CoGroupedRDD.scala
@@ -29,10 +29,12 @@ import org.apache.spark.serializer.Serializer
 import org.apache.spark.util.collection.{CompactBuffer, ExternalAppendOnlyMap}
 import org.apache.spark.util.Utils
 
-/** The references to rdd and splitIndex are transient because redundant information is stored
-  * in the CoGroupedRDD object.  Because CoGroupedRDD is serialized separately from
-  * CoGroupPartition, if rdd and splitIndex aren't transient, they'll be included twice in the
-  * task closure. */
+/**
+ * The references to rdd and splitIndex are transient because redundant information is stored
+ * in the CoGroupedRDD object.  Because CoGroupedRDD is serialized separately from
+ * CoGroupPartition, if rdd and splitIndex aren't transient, they'll be included twice in the
+ * task closure.
+ */
 private[spark] case class NarrowCoGroupSplitDep(
     @transient rdd: RDD[_],
     @transient splitIndex: Int,
diff --git a/core/src/main/scala/org/apache/spark/rdd/RDD.scala b/core/src/main/scala/org/apache/spark/rdd/RDD.scala
index f96551c793..4a0a2199ef 100644
--- a/core/src/main/scala/org/apache/spark/rdd/RDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/RDD.scala
@@ -255,8 +255,8 @@ abstract class RDD[T: ClassTag](
   }
 
   /**
-    * Returns the number of partitions of this RDD.
-    */
+   * Returns the number of partitions of this RDD.
+   */
   @Since("1.6.0")
   final def getNumPartitions: Int = partitions.length
 
diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala
index 90b1813750..50b452c72f 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala
@@ -295,12 +295,12 @@ private[spark] class CoarseMesosSchedulerBackend(
   }
 
   /**
-    * Launches executors on accepted offers, and declines unused offers. Executors are launched
-    * round-robin on offers.
-    *
-    * @param d SchedulerDriver
-    * @param offers Mesos offers that match attribute constraints
-    */
+   * Launches executors on accepted offers, and declines unused offers. Executors are launched
+   * round-robin on offers.
+   *
+   * @param d SchedulerDriver
+   * @param offers Mesos offers that match attribute constraints
+   */
   private def handleMatchedOffers(d: SchedulerDriver, offers: Buffer[Offer]): Unit = {
     val tasks = buildMesosTasks(offers)
     for (offer <- offers) {
@@ -336,12 +336,12 @@ private[spark] class CoarseMesosSchedulerBackend(
   }
 
   /**
-    * Returns a map from OfferIDs to the tasks to launch on those offers.  In order to maximize
-    * per-task memory and IO, tasks are round-robin assigned to offers.
-    *
-    * @param offers Mesos offers that match attribute constraints
-    * @return A map from OfferID to a list of Mesos tasks to launch on that offer
-    */
+   * Returns a map from OfferIDs to the tasks to launch on those offers.  In order to maximize
+   * per-task memory and IO, tasks are round-robin assigned to offers.
+   *
+   * @param offers Mesos offers that match attribute constraints
+   * @return A map from OfferID to a list of Mesos tasks to launch on that offer
+   */
   private def buildMesosTasks(offers: Buffer[Offer]): Map[OfferID, List[MesosTaskInfo]] = {
     // offerID -> tasks
     val tasks = new HashMap[OfferID, List[MesosTaskInfo]].withDefaultValue(Nil)
diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterScheduler.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterScheduler.scala
index c41fa58607..73bd4c58e1 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterScheduler.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterScheduler.scala
@@ -453,12 +453,12 @@ private[spark] class MesosClusterScheduler(
   }
 
   /**
-    * Escape args for Unix-like shells, unless already quoted by the user.
-    * Based on: http://www.gnu.org/software/bash/manual/html_node/Double-Quotes.html
-    * and http://www.grymoire.com/Unix/Quote.html
-    * @param value argument
-    * @return escaped argument
-    */
+   * Escape args for Unix-like shells, unless already quoted by the user.
+   * Based on: http://www.gnu.org/software/bash/manual/html_node/Double-Quotes.html
+   * and http://www.grymoire.com/Unix/Quote.html
+   * @param value argument
+   * @return escaped argument
+   */
   private[scheduler] def shellEscape(value: String): String = {
     val WrappedInQuotes = """^(".+"|'.+')$""".r
     val ShellSpecialChars = (""".*([ '<>&|\?\*;!#\\(\)"$`]).*""").r
diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala
index 9a12a61f2f..35f914355d 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala
@@ -148,8 +148,8 @@ private[mesos] trait MesosSchedulerUtils extends Logging {
   }
 
   /**
-    * Signal that the scheduler has registered with Mesos.
-    */
+   * Signal that the scheduler has registered with Mesos.
+   */
   protected def markRegistered(): Unit = {
     registerLatch.countDown()
   }
diff --git a/core/src/main/scala/org/apache/spark/shuffle/ShuffleManager.scala b/core/src/main/scala/org/apache/spark/shuffle/ShuffleManager.scala
index 76fd249fbd..364fad664e 100644
--- a/core/src/main/scala/org/apache/spark/shuffle/ShuffleManager.scala
+++ b/core/src/main/scala/org/apache/spark/shuffle/ShuffleManager.scala
@@ -54,9 +54,9 @@ private[spark] trait ShuffleManager {
       context: TaskContext): ShuffleReader[K, C]
 
   /**
-    * Remove a shuffle's metadata from the ShuffleManager.
-    * @return true if the metadata removed successfully, otherwise false.
-    */
+   * Remove a shuffle's metadata from the ShuffleManager.
+   * @return true if the metadata removed successfully, otherwise false.
+   */
   def unregisterShuffle(shuffleId: Int): Boolean
 
   /**
diff --git a/core/src/main/scala/org/apache/spark/storage/memory/MemoryStore.scala b/core/src/main/scala/org/apache/spark/storage/memory/MemoryStore.scala
index df38d11e43..99be4de065 100644
--- a/core/src/main/scala/org/apache/spark/storage/memory/MemoryStore.scala
+++ b/core/src/main/scala/org/apache/spark/storage/memory/MemoryStore.scala
@@ -455,16 +455,16 @@ private[spark] class MemoryStore(
   }
 
   /**
-    * Try to evict blocks to free up a given amount of space to store a particular block.
-    * Can fail if either the block is bigger than our memory or it would require replacing
-    * another block from the same RDD (which leads to a wasteful cyclic replacement pattern for
-    * RDDs that don't fit into memory that we want to avoid).
-    *
-    * @param blockId the ID of the block we are freeing space for, if any
-    * @param space the size of this block
-    * @param memoryMode the type of memory to free (on- or off-heap)
-    * @return the amount of memory (in bytes) freed by eviction
-    */
+   * Try to evict blocks to free up a given amount of space to store a particular block.
+   * Can fail if either the block is bigger than our memory or it would require replacing
+   * another block from the same RDD (which leads to a wasteful cyclic replacement pattern for
+   * RDDs that don't fit into memory that we want to avoid).
+   *
+   * @param blockId the ID of the block we are freeing space for, if any
+   * @param space the size of this block
+   * @param memoryMode the type of memory to free (on- or off-heap)
+   * @return the amount of memory (in bytes) freed by eviction
+   */
   private[spark] def evictBlocksToFreeSpace(
       blockId: Option[BlockId],
       space: Long,
diff --git a/core/src/main/scala/org/apache/spark/util/Utils.scala b/core/src/main/scala/org/apache/spark/util/Utils.scala
index 73768ff4c8..50bcf85805 100644
--- a/core/src/main/scala/org/apache/spark/util/Utils.scala
+++ b/core/src/main/scala/org/apache/spark/util/Utils.scala
@@ -256,10 +256,11 @@ private[spark] object Utils extends Logging {
     dir
   }
 
-  /** Copy all data from an InputStream to an OutputStream. NIO way of file stream to file stream
-    * copying is disabled by default unless explicitly set transferToEnabled as true,
-    * the parameter transferToEnabled should be configured by spark.file.transferTo = [true|false].
-    */
+  /**
+   * Copy all data from an InputStream to an OutputStream. NIO way of file stream to file stream
+   * copying is disabled by default unless explicitly set transferToEnabled as true,
+   * the parameter transferToEnabled should be configured by spark.file.transferTo = [true|false].
+   */
   def copyStream(in: InputStream,
                  out: OutputStream,
                  closeStreams: Boolean = false,
@@ -1564,9 +1565,11 @@ private[spark] object Utils extends Logging {
     else -1
   }
 
-  /** Returns the system properties map that is thread-safe to iterator over. It gets the
-    * properties which have been set explicitly, as well as those for which only a default value
-    * has been defined. */
+  /**
+   * Returns the system properties map that is thread-safe to iterator over. It gets the
+   * properties which have been set explicitly, as well as those for which only a default value
+   * has been defined.
+   */
   def getSystemProperties: Map[String, String] = {
     System.getProperties.stringPropertyNames().asScala
       .map(key => (key, System.getProperty(key))).toMap
diff --git a/core/src/test/scala/org/apache/spark/Smuggle.scala b/core/src/test/scala/org/apache/spark/Smuggle.scala
index 9f0a1b4c25..9d9217ea1b 100644
--- a/core/src/test/scala/org/apache/spark/Smuggle.scala
+++ b/core/src/test/scala/org/apache/spark/Smuggle.scala
@@ -24,16 +24,16 @@ import scala.collection.mutable
 import scala.language.implicitConversions
 
 /**
-  * Utility wrapper to "smuggle" objects into tasks while bypassing serialization.
-  * This is intended for testing purposes, primarily to make locks, semaphores, and
-  * other constructs that would not survive serialization available from within tasks.
-  * A Smuggle reference is itself serializable, but after being serialized and
-  * deserialized, it still refers to the same underlying "smuggled" object, as long
-  * as it was deserialized within the same JVM. This can be useful for tests that
-  * depend on the timing of task completion to be deterministic, since one can "smuggle"
-  * a lock or semaphore into the task, and then the task can block until the test gives
-  * the go-ahead to proceed via the lock.
-  */
+ * Utility wrapper to "smuggle" objects into tasks while bypassing serialization.
+ * This is intended for testing purposes, primarily to make locks, semaphores, and
+ * other constructs that would not survive serialization available from within tasks.
+ * A Smuggle reference is itself serializable, but after being serialized and
+ * deserialized, it still refers to the same underlying "smuggled" object, as long
+ * as it was deserialized within the same JVM. This can be useful for tests that
+ * depend on the timing of task completion to be deterministic, since one can "smuggle"
+ * a lock or semaphore into the task, and then the task can block until the test gives
+ * the go-ahead to proceed via the lock.
+ */
 class Smuggle[T] private(val key: Symbol) extends Serializable {
   def smuggledObject: T = Smuggle.get(key)
 }
@@ -41,13 +41,13 @@ class Smuggle[T] private(val key: Symbol) extends Serializable {
 
 object Smuggle {
   /**
-    * Wraps the specified object to be smuggled into a serialized task without
-    * being serialized itself.
-    *
-    * @param smuggledObject
-    * @tparam T
-    * @return Smuggle wrapper around smuggledObject.
-    */
+   * Wraps the specified object to be smuggled into a serialized task without
+   * being serialized itself.
+   *
+   * @param smuggledObject
+   * @tparam T
+   * @return Smuggle wrapper around smuggledObject.
+   */
   def apply[T](smuggledObject: T): Smuggle[T] = {
     val key = Symbol(UUID.randomUUID().toString)
     lock.writeLock().lock()
@@ -72,12 +72,12 @@ object Smuggle {
   }
 
   /**
-    * Implicit conversion of a Smuggle wrapper to the object being smuggled.
-    *
-    * @param smuggle the wrapper to unpack.
-    * @tparam T
-    * @return the smuggled object represented by the wrapper.
-    */
+   * Implicit conversion of a Smuggle wrapper to the object being smuggled.
+   *
+   * @param smuggle the wrapper to unpack.
+   * @tparam T
+   * @return the smuggled object represented by the wrapper.
+   */
   implicit def unpackSmuggledObject[T](smuggle : Smuggle[T]): T = smuggle.smuggledObject
 
 }
diff --git a/core/src/test/scala/org/apache/spark/memory/MemoryManagerSuite.scala b/core/src/test/scala/org/apache/spark/memory/MemoryManagerSuite.scala
index 3d1a0e9795..99d5b496bc 100644
--- a/core/src/test/scala/org/apache/spark/memory/MemoryManagerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/memory/MemoryManagerSuite.scala
@@ -78,18 +78,18 @@ private[memory] trait MemoryManagerSuite extends SparkFunSuite with BeforeAndAft
   }
 
   /**
-    * Simulate the part of [[MemoryStore.evictBlocksToFreeSpace]] that releases storage memory.
-    *
-    * This is a significant simplification of the real method, which actually drops existing
-    * blocks based on the size of each block. Instead, here we simply release as many bytes
-    * as needed to ensure the requested amount of free space. This allows us to set up the
-    * test without relying on the [[org.apache.spark.storage.BlockManager]], which brings in
-    * many other dependencies.
-    *
-    * Every call to this method will set a global variable, [[evictBlocksToFreeSpaceCalled]], that
-    * records the number of bytes this is called with. This variable is expected to be cleared
-    * by the test code later through [[assertEvictBlocksToFreeSpaceCalled]].
-    */
+   * Simulate the part of [[MemoryStore.evictBlocksToFreeSpace]] that releases storage memory.
+   *
+   * This is a significant simplification of the real method, which actually drops existing
+   * blocks based on the size of each block. Instead, here we simply release as many bytes
+   * as needed to ensure the requested amount of free space. This allows us to set up the
+   * test without relying on the [[org.apache.spark.storage.BlockManager]], which brings in
+   * many other dependencies.
+   *
+   * Every call to this method will set a global variable, [[evictBlocksToFreeSpaceCalled]], that
+   * records the number of bytes this is called with. This variable is expected to be cleared
+   * by the test code later through [[assertEvictBlocksToFreeSpaceCalled]].
+   */
   private def evictBlocksToFreeSpaceAnswer(mm: MemoryManager): Answer[Long] = {
     new Answer[Long] {
       override def answer(invocation: InvocationOnMock): Long = {
diff --git a/examples/src/main/scala/org/apache/spark/examples/BroadcastTest.scala b/examples/src/main/scala/org/apache/spark/examples/BroadcastTest.scala
index 3da5236745..af5a815f6e 100644
--- a/examples/src/main/scala/org/apache/spark/examples/BroadcastTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/BroadcastTest.scala
@@ -21,8 +21,8 @@ package org.apache.spark.examples
 import org.apache.spark.{SparkConf, SparkContext}
 
 /**
-  * Usage: BroadcastTest [slices] [numElem] [blockSize]
-  */
+ * Usage: BroadcastTest [slices] [numElem] [blockSize]
+ */
 object BroadcastTest {
   def main(args: Array[String]) {
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/DFSReadWriteTest.scala b/examples/src/main/scala/org/apache/spark/examples/DFSReadWriteTest.scala
index 743fc13db7..7bf023667d 100644
--- a/examples/src/main/scala/org/apache/spark/examples/DFSReadWriteTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/DFSReadWriteTest.scala
@@ -25,16 +25,16 @@ import scala.io.Source._
 import org.apache.spark.{SparkConf, SparkContext}
 
 /**
-  * Simple test for reading and writing to a distributed
-  * file system.  This example does the following:
-  *
-  *   1. Reads local file
-  *   2. Computes word count on local file
-  *   3. Writes local file to a DFS
-  *   4. Reads the file back from the DFS
-  *   5. Computes word count on the file using Spark
-  *   6. Compares the word count results
-  */
+ * Simple test for reading and writing to a distributed
+ * file system.  This example does the following:
+ *
+ *   1. Reads local file
+ *   2. Computes word count on local file
+ *   3. Writes local file to a DFS
+ *   4. Reads the file back from the DFS
+ *   5. Computes word count on the file using Spark
+ *   6. Compares the word count results
+ */
 object DFSReadWriteTest {
 
   private var localFilePath: File = new File(".")
diff --git a/examples/src/main/scala/org/apache/spark/examples/GroupByTest.scala b/examples/src/main/scala/org/apache/spark/examples/GroupByTest.scala
index 08b6c717d4..4db229b5de 100644
--- a/examples/src/main/scala/org/apache/spark/examples/GroupByTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/GroupByTest.scala
@@ -23,8 +23,8 @@ import java.util.Random
 import org.apache.spark.{SparkConf, SparkContext}
 
 /**
-  * Usage: GroupByTest [numMappers] [numKVPairs] [KeySize] [numReducers]
-  */
+ * Usage: GroupByTest [numMappers] [numKVPairs] [KeySize] [numReducers]
+ */
 object GroupByTest {
   def main(args: Array[String]) {
     val sparkConf = new SparkConf().setAppName("GroupBy Test")
diff --git a/examples/src/main/scala/org/apache/spark/examples/MultiBroadcastTest.scala b/examples/src/main/scala/org/apache/spark/examples/MultiBroadcastTest.scala
index 134c3d1d63..3eb0c27723 100644
--- a/examples/src/main/scala/org/apache/spark/examples/MultiBroadcastTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/MultiBroadcastTest.scala
@@ -22,8 +22,8 @@ import org.apache.spark.{SparkConf, SparkContext}
 import org.apache.spark.rdd.RDD
 
 /**
-  * Usage: MultiBroadcastTest [slices] [numElem]
-  */
+ * Usage: MultiBroadcastTest [slices] [numElem]
+ */
 object MultiBroadcastTest {
   def main(args: Array[String]) {
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/SimpleSkewedGroupByTest.scala b/examples/src/main/scala/org/apache/spark/examples/SimpleSkewedGroupByTest.scala
index 7c09664c2f..ec07e6323e 100644
--- a/examples/src/main/scala/org/apache/spark/examples/SimpleSkewedGroupByTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/SimpleSkewedGroupByTest.scala
@@ -23,8 +23,8 @@ import java.util.Random
 import org.apache.spark.{SparkConf, SparkContext}
 
 /**
-  * Usage: SimpleSkewedGroupByTest [numMappers] [numKVPairs] [valSize] [numReducers] [ratio]
-  */
+ * Usage: SimpleSkewedGroupByTest [numMappers] [numKVPairs] [valSize] [numReducers] [ratio]
+ */
 object SimpleSkewedGroupByTest {
   def main(args: Array[String]) {
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/SkewedGroupByTest.scala b/examples/src/main/scala/org/apache/spark/examples/SkewedGroupByTest.scala
index d498af9c39..8e4c2b6229 100644
--- a/examples/src/main/scala/org/apache/spark/examples/SkewedGroupByTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/SkewedGroupByTest.scala
@@ -23,8 +23,8 @@ import java.util.Random
 import org.apache.spark.{SparkConf, SparkContext}
 
 /**
-  * Usage: GroupByTest [numMappers] [numKVPairs] [KeySize] [numReducers]
-  */
+ * Usage: GroupByTest [numMappers] [numKVPairs] [KeySize] [numReducers]
+ */
 object SkewedGroupByTest {
   def main(args: Array[String]) {
     val sparkConf = new SparkConf().setAppName("GroupBy Test")
diff --git a/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewGenerator.scala b/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewGenerator.scala
index 50216b9bd4..0ddd065f0d 100644
--- a/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewGenerator.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewGenerator.scala
@@ -38,17 +38,18 @@ object PageView extends Serializable {
 }
 
 // scalastyle:off
-/** Generates streaming events to simulate page views on a website.
-  *
-  * This should be used in tandem with PageViewStream.scala. Example:
-  *
-  * To run the generator
-  * `$ bin/run-example org.apache.spark.examples.streaming.clickstream.PageViewGenerator 44444 10`
-  * To process the generated stream
-  * `$ bin/run-example \
-  *    org.apache.spark.examples.streaming.clickstream.PageViewStream errorRatePerZipCode localhost 44444`
-  *
-  */
+/**
+ * Generates streaming events to simulate page views on a website.
+ *
+ * This should be used in tandem with PageViewStream.scala. Example:
+ *
+ * To run the generator
+ * `$ bin/run-example org.apache.spark.examples.streaming.clickstream.PageViewGenerator 44444 10`
+ * To process the generated stream
+ * `$ bin/run-example \
+ *    org.apache.spark.examples.streaming.clickstream.PageViewStream errorRatePerZipCode localhost 44444`
+ *
+ */
 // scalastyle:on
 object PageViewGenerator {
   val pages = Map("http://foo.com/" -> .7,
diff --git a/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewStream.scala b/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewStream.scala
index 773a2e5fc2..1ba093f57b 100644
--- a/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewStream.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/streaming/clickstream/PageViewStream.scala
@@ -22,16 +22,17 @@ import org.apache.spark.examples.streaming.StreamingExamples
 import org.apache.spark.streaming.{Seconds, StreamingContext}
 
 // scalastyle:off
-/** Analyses a streaming dataset of web page views. This class demonstrates several types of
-  * operators available in Spark streaming.
-  *
-  * This should be used in tandem with PageViewStream.scala. Example:
-  * To run the generator
-  * `$ bin/run-example org.apache.spark.examples.streaming.clickstream.PageViewGenerator 44444 10`
-  * To process the generated stream
-  * `$ bin/run-example \
-  *    org.apache.spark.examples.streaming.clickstream.PageViewStream errorRatePerZipCode localhost 44444`
-  */
+/**
+ * Analyses a streaming dataset of web page views. This class demonstrates several types of
+ * operators available in Spark streaming.
+ *
+ * This should be used in tandem with PageViewStream.scala. Example:
+ * To run the generator
+ * `$ bin/run-example org.apache.spark.examples.streaming.clickstream.PageViewGenerator 44444 10`
+ * To process the generated stream
+ * `$ bin/run-example \
+ *    org.apache.spark.examples.streaming.clickstream.PageViewStream errorRatePerZipCode localhost 44444`
+ */
 // scalastyle:on
 object PageViewStream {
   def main(args: Array[String]) {
diff --git a/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala b/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala
index 7dc9606913..6e7c3f358e 100644
--- a/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala
+++ b/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala
@@ -185,13 +185,14 @@ class FlumeReceiver(
 
   override def preferredLocation: Option[String] = Option(host)
 
-  /** A Netty Pipeline factory that will decompress incoming data from
-    * and the Netty client and compress data going back to the client.
-    *
-    * The compression on the return is required because Flume requires
-    * a successful response to indicate it can remove the event/batch
-    * from the configured channel
-    */
+  /**
+   * A Netty Pipeline factory that will decompress incoming data from
+   * and the Netty client and compress data going back to the client.
+   *
+   * The compression on the return is required because Flume requires
+   * a successful response to indicate it can remove the event/batch
+   * from the configured channel
+   */
   private[streaming]
   class CompressionChannelPipelineFactory extends ChannelPipelineFactory {
     def getPipeline(): ChannelPipeline = {
diff --git a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala
index a660d2a00c..02917becf0 100644
--- a/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala
+++ b/external/kafka/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDDPartition.scala
@@ -19,13 +19,14 @@ package org.apache.spark.streaming.kafka
 
 import org.apache.spark.Partition
 
-/** @param topic kafka topic name
-  * @param partition kafka partition id
-  * @param fromOffset inclusive starting offset
-  * @param untilOffset exclusive ending offset
-  * @param host preferred kafka host, i.e. the leader at the time the rdd was created
-  * @param port preferred kafka host's port
-  */
+/**
+ * @param topic kafka topic name
+ * @param partition kafka partition id
+ * @param fromOffset inclusive starting offset
+ * @param untilOffset exclusive ending offset
+ * @param host preferred kafka host, i.e. the leader at the time the rdd was created
+ * @param port preferred kafka host's port
+ */
 private[kafka]
 class KafkaRDDPartition(
   val index: Int,
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/GraphOps.scala b/graphx/src/main/scala/org/apache/spark/graphx/GraphOps.scala
index a783fe305f..868658dfe5 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/GraphOps.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/GraphOps.scala
@@ -415,11 +415,11 @@ class GraphOps[VD: ClassTag, ED: ClassTag](graph: Graph[VD, ED]) extends Seriali
   }
 
   /**
-    * Compute the connected component membership of each vertex and return a graph with the vertex
-    * value containing the lowest vertex id in the connected component containing that vertex.
-    *
-    * @see [[org.apache.spark.graphx.lib.ConnectedComponents$#run]]
-    */
+   * Compute the connected component membership of each vertex and return a graph with the vertex
+   * value containing the lowest vertex id in the connected component containing that vertex.
+   *
+   * @see [[org.apache.spark.graphx.lib.ConnectedComponents$#run]]
+   */
   def connectedComponents(): Graph[VertexId, ED] = {
     ConnectedComponents.run(graph)
   }
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/lib/ConnectedComponents.scala b/graphx/src/main/scala/org/apache/spark/graphx/lib/ConnectedComponents.scala
index 137c512c99..4e9b13162e 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/lib/ConnectedComponents.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/lib/ConnectedComponents.scala
@@ -60,15 +60,15 @@ object ConnectedComponents {
   } // end of connectedComponents
 
   /**
-    * Compute the connected component membership of each vertex and return a graph with the vertex
-    * value containing the lowest vertex id in the connected component containing that vertex.
-    *
-    * @tparam VD the vertex attribute type (discarded in the computation)
-    * @tparam ED the edge attribute type (preserved in the computation)
-    * @param graph the graph for which to compute the connected components
-    * @return a graph with vertex attributes containing the smallest vertex in each
-    *         connected component
-    */
+   * Compute the connected component membership of each vertex and return a graph with the vertex
+   * value containing the lowest vertex id in the connected component containing that vertex.
+   *
+   * @tparam VD the vertex attribute type (discarded in the computation)
+   * @tparam ED the edge attribute type (preserved in the computation)
+   * @param graph the graph for which to compute the connected components
+   * @return a graph with vertex attributes containing the smallest vertex in each
+   *         connected component
+   */
   def run[VD: ClassTag, ED: ClassTag](graph: Graph[VD, ED]): Graph[VertexId, ED] = {
     run(graph, Int.MaxValue)
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/ElementwiseProduct.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/ElementwiseProduct.scala
index 2c7ffdb7ba..1b0a9a12e8 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/ElementwiseProduct.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/ElementwiseProduct.scala
@@ -38,9 +38,9 @@ class ElementwiseProduct(override val uid: String)
   def this() = this(Identifiable.randomUID("elemProd"))
 
   /**
-    * the vector to multiply with input vectors
-    * @group param
-    */
+   * the vector to multiply with input vectors
+   * @group param
+   */
   val scalingVec: Param[Vector] = new Param(this, "scalingVec", "vector for hadamard product")
 
   /** @group setParam */
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/api/python/GaussianMixtureModelWrapper.scala b/mllib/src/main/scala/org/apache/spark/mllib/api/python/GaussianMixtureModelWrapper.scala
index a689b09341..364d5eea08 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/api/python/GaussianMixtureModelWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/api/python/GaussianMixtureModelWrapper.scala
@@ -24,15 +24,15 @@ import org.apache.spark.mllib.clustering.GaussianMixtureModel
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 
 /**
-  * Wrapper around GaussianMixtureModel to provide helper methods in Python
-  */
+ * Wrapper around GaussianMixtureModel to provide helper methods in Python
+ */
 private[python] class GaussianMixtureModelWrapper(model: GaussianMixtureModel) {
   val weights: Vector = Vectors.dense(model.weights)
   val k: Int = weights.size
 
   /**
-    * Returns gaussians as a List of Vectors and Matrices corresponding each MultivariateGaussian
-    */
+   * Returns gaussians as a List of Vectors and Matrices corresponding each MultivariateGaussian
+   */
   val gaussians: Array[Byte] = {
     val modelGaussians = model.gaussians.map { gaussian =>
       Array[Any](gaussian.mu, gaussian.sigma)
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/api/python/Word2VecModelWrapper.scala b/mllib/src/main/scala/org/apache/spark/mllib/api/python/Word2VecModelWrapper.scala
index 073f03e16f..05273c3434 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/api/python/Word2VecModelWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/api/python/Word2VecModelWrapper.scala
@@ -27,8 +27,8 @@ import org.apache.spark.mllib.feature.Word2VecModel
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 
 /**
-  * Wrapper around Word2VecModel to provide helper methods in Python
-  */
+ * Wrapper around Word2VecModel to provide helper methods in Python
+ */
 private[python] class Word2VecModelWrapper(model: Word2VecModel) {
   def transform(word: String): Vector = {
     model.transform(word)
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
index 6e571fe35a..8c09b69b3c 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
@@ -123,14 +123,18 @@ sealed trait Matrix extends Serializable {
   @Since("1.4.0")
   def toString(maxLines: Int, maxLineWidth: Int): String = toBreeze.toString(maxLines, maxLineWidth)
 
-  /** Map the values of this matrix using a function. Generates a new matrix. Performs the
-    * function on only the backing array. For example, an operation such as addition or
-    * subtraction will only be performed on the non-zero values in a `SparseMatrix`. */
+  /**
+   * Map the values of this matrix using a function. Generates a new matrix. Performs the
+   * function on only the backing array. For example, an operation such as addition or
+   * subtraction will only be performed on the non-zero values in a `SparseMatrix`.
+   */
   private[spark] def map(f: Double => Double): Matrix
 
-  /** Update all the values of this matrix using the function f. Performed in-place on the
-    * backing array. For example, an operation such as addition or subtraction will only be
-    * performed on the non-zero values in a `SparseMatrix`. */
+  /**
+   * Update all the values of this matrix using the function f. Performed in-place on the
+   * backing array. For example, an operation such as addition or subtraction will only be
+   * performed on the non-zero values in a `SparseMatrix`.
+   */
   private[mllib] def update(f: Double => Double): Matrix
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearRegressionWithSGD.scala b/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearRegressionWithSGD.scala
index e8f4422fd4..84764963b5 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearRegressionWithSGD.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/regression/StreamingLinearRegressionWithSGD.scala
@@ -81,8 +81,8 @@ class StreamingLinearRegressionWithSGD private[mllib] (
   }
 
   /**
-    * Set the number of iterations of gradient descent to run per update. Default: 50.
-    */
+   * Set the number of iterations of gradient descent to run per update. Default: 50.
+   */
   @Since("1.1.0")
   def setNumIterations(numIterations: Int): this.type = {
     this.algorithm.optimizer.setNumIterations(numIterations)
diff --git a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala
index 67a616dc15..c5dc6ba221 100644
--- a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala
+++ b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkILoop.scala
@@ -797,9 +797,11 @@ class SparkILoop(
     // echo("Switched " + (if (old) "off" else "on") + " result printing.")
   }
 
-  /** Run one command submitted by the user.  Two values are returned:
-    * (1) whether to keep running, (2) the line to record for replay,
-    * if any. */
+  /**
+   * Run one command submitted by the user.  Two values are returned:
+   * (1) whether to keep running, (2) the line to record for replay,
+   * if any.
+   */
   private[repl] def command(line: String): Result = {
     if (line startsWith ":") {
       val cmd = line.tail takeWhile (x => !x.isWhitespace)
@@ -841,12 +843,13 @@ class SparkILoop(
   }
   import paste.{ ContinueString, PromptString }
 
-  /** Interpret expressions starting with the first line.
-    * Read lines until a complete compilation unit is available
-    * or until a syntax error has been seen.  If a full unit is
-    * read, go ahead and interpret it.  Return the full string
-    * to be recorded for replay, if any.
-    */
+  /**
+   * Interpret expressions starting with the first line.
+   * Read lines until a complete compilation unit is available
+   * or until a syntax error has been seen.  If a full unit is
+   * read, go ahead and interpret it.  Return the full string
+   * to be recorded for replay, if any.
+   */
   private def interpretStartingWith(code: String): Option[String] = {
     // signal completion non-completion input has been received
     in.completion.resetVerbosity()
diff --git a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkImports.scala b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkImports.scala
index 1d0fe10d3d..f22776592c 100644
--- a/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkImports.scala
+++ b/repl/scala-2.10/src/main/scala/org/apache/spark/repl/SparkImports.scala
@@ -118,8 +118,9 @@ private[repl] trait SparkImports {
     case class ReqAndHandler(req: Request, handler: MemberHandler) { }
 
     def reqsToUse: List[ReqAndHandler] = {
-      /** Loop through a list of MemberHandlers and select which ones to keep.
-        * 'wanted' is the set of names that need to be imported.
+      /**
+       * Loop through a list of MemberHandlers and select which ones to keep.
+       * 'wanted' is the set of names that need to be imported.
        */
       def select(reqs: List[ReqAndHandler], wanted: Set[Name]): List[ReqAndHandler] = {
         // Single symbol imports might be implicits! See bug #1752.  Rather than
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/Encoder.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/Encoder.scala
index 1f20e26354..e0bfe3c32f 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/Encoder.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/Encoder.scala
@@ -140,27 +140,27 @@ object Encoders {
   def STRING: Encoder[java.lang.String] = ExpressionEncoder()
 
   /**
-    * An encoder for nullable decimal type.
-    * @since 1.6.0
-    */
+   * An encoder for nullable decimal type.
+   * @since 1.6.0
+   */
   def DECIMAL: Encoder[java.math.BigDecimal] = ExpressionEncoder()
 
   /**
-    * An encoder for nullable date type.
-    * @since 1.6.0
-    */
+   * An encoder for nullable date type.
+   * @since 1.6.0
+   */
   def DATE: Encoder[java.sql.Date] = ExpressionEncoder()
 
   /**
-    * An encoder for nullable timestamp type.
-    * @since 1.6.0
-    */
+   * An encoder for nullable timestamp type.
+   * @since 1.6.0
+   */
   def TIMESTAMP: Encoder[java.sql.Timestamp] = ExpressionEncoder()
 
   /**
-    * An encoder for arrays of bytes.
-    * @since 1.6.1
-    */
+   * An encoder for arrays of bytes.
+   * @since 1.6.1
+   */
   def BINARY: Encoder[Array[Byte]] = ExpressionEncoder()
 
   /**
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
index 05e2b9a447..a6e317ebf0 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
@@ -733,9 +733,9 @@ class Analyzer(
     }
 
     /**
-      * Add the missing attributes into projectList of Project/Window or aggregateExpressions of
-      * Aggregate.
-      */
+     * Add the missing attributes into projectList of Project/Window or aggregateExpressions of
+     * Aggregate.
+     */
     private def addMissingAttr(plan: LogicalPlan, missingAttrs: AttributeSet): LogicalPlan = {
       if (missingAttrs.isEmpty) {
         return plan
@@ -767,9 +767,9 @@ class Analyzer(
     }
 
     /**
-      * Resolve the expression on a specified logical plan and it's child (recursively), until
-      * the expression is resolved or meet a non-unary node or Subquery.
-      */
+     * Resolve the expression on a specified logical plan and it's child (recursively), until
+     * the expression is resolved or meet a non-unary node or Subquery.
+     */
     @tailrec
     private def resolveExpressionRecursively(expr: Expression, plan: LogicalPlan): Expression = {
       val resolved = resolveExpression(expr, plan)
@@ -1398,8 +1398,8 @@ class Analyzer(
   }
 
   /**
-    * Check and add order to [[AggregateWindowFunction]]s.
-    */
+   * Check and add order to [[AggregateWindowFunction]]s.
+   */
   object ResolveWindowOrder extends Rule[LogicalPlan] {
     def apply(plan: LogicalPlan): LogicalPlan = plan transform {
       case logical: LogicalPlan => logical transformExpressions {
@@ -1489,8 +1489,8 @@ object EliminateSubqueryAliases extends Rule[LogicalPlan] {
 }
 
 /**
-  * Removes [[Union]] operators from the plan if it just has one child.
-  */
+ * Removes [[Union]] operators from the plan if it just has one child.
+ */
 object EliminateUnions extends Rule[LogicalPlan] {
   def apply(plan: LogicalPlan): LogicalPlan = plan transform {
     case Union(children) if children.size == 1 => children.head
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Projection.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Projection.scala
index 053e612f3e..354311c5e7 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Projection.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Projection.scala
@@ -136,9 +136,9 @@ object UnsafeProjection {
   }
 
   /**
-    * Same as other create()'s but allowing enabling/disabling subexpression elimination.
-    * TODO: refactor the plumbing and clean this up.
-    */
+   * Same as other create()'s but allowing enabling/disabling subexpression elimination.
+   * TODO: refactor the plumbing and clean this up.
+   */
   def create(
       exprs: Seq[Expression],
       inputSchema: Seq[Attribute],
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
index cd490dd676..b64d3eea49 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
@@ -58,10 +58,10 @@ class CodegenContext {
   val references: mutable.ArrayBuffer[Any] = new mutable.ArrayBuffer[Any]()
 
   /**
-    * Add an object to `references`, create a class member to access it.
-    *
-    * Returns the name of class member.
-    */
+   * Add an object to `references`, create a class member to access it.
+   *
+   * Returns the name of class member.
+   */
   def addReferenceObj(name: String, obj: Any, className: String = null): String = {
     val term = freshName(name)
     val idx = references.length
@@ -72,9 +72,9 @@ class CodegenContext {
   }
 
   /**
-    * Holding a list of generated columns as input of current operator, will be used by
-    * BoundReference to generate code.
-    */
+   * Holding a list of generated columns as input of current operator, will be used by
+   * BoundReference to generate code.
+   */
   var currentVars: Seq[ExprCode] = null
 
   /**
@@ -169,14 +169,14 @@ class CodegenContext {
   final var INPUT_ROW = "i"
 
   /**
-    * The map from a variable name to it's next ID.
-    */
+   * The map from a variable name to it's next ID.
+   */
   private val freshNameIds = new mutable.HashMap[String, Int]
   freshNameIds += INPUT_ROW -> 1
 
   /**
-    * A prefix used to generate fresh name.
-    */
+   * A prefix used to generate fresh name.
+   */
   var freshNamePrefix = ""
 
   /**
@@ -234,8 +234,8 @@ class CodegenContext {
   }
 
   /**
-    * Update a column in MutableRow from ExprCode.
-    */
+   * Update a column in MutableRow from ExprCode.
+   */
   def updateColumn(
       row: String,
       dataType: DataType,
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/grouping.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/grouping.scala
index 437e417266..3be761c867 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/grouping.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/grouping.scala
@@ -22,8 +22,8 @@ import org.apache.spark.sql.catalyst.expressions.codegen.CodegenFallback
 import org.apache.spark.sql.types._
 
 /**
-  * A placeholder expression for cube/rollup, which will be replaced by analyzer
-  */
+ * A placeholder expression for cube/rollup, which will be replaced by analyzer
+ */
 trait GroupingSet extends Expression with CodegenFallback {
 
   def groupByExprs: Seq[Expression]
@@ -43,9 +43,9 @@ case class Cube(groupByExprs: Seq[Expression]) extends GroupingSet {}
 case class Rollup(groupByExprs: Seq[Expression]) extends GroupingSet {}
 
 /**
-  * Indicates whether a specified column expression in a GROUP BY list is aggregated or not.
-  * GROUPING returns 1 for aggregated or 0 for not aggregated in the result set.
-  */
+ * Indicates whether a specified column expression in a GROUP BY list is aggregated or not.
+ * GROUPING returns 1 for aggregated or 0 for not aggregated in the result set.
+ */
 case class Grouping(child: Expression) extends Expression with Unevaluable {
   override def references: AttributeSet = AttributeSet(VirtualColumn.groupingIdAttribute :: Nil)
   override def children: Seq[Expression] = child :: Nil
@@ -54,10 +54,10 @@ case class Grouping(child: Expression) extends Expression with Unevaluable {
 }
 
 /**
-  * GroupingID is a function that computes the level of grouping.
-  *
-  * If groupByExprs is empty, it means all grouping expressions in GroupingSets.
-  */
+ * GroupingID is a function that computes the level of grouping.
+ *
+ * If groupByExprs is empty, it means all grouping expressions in GroupingSets.
+ */
 case class GroupingID(groupByExprs: Seq[Expression]) extends Expression with Unevaluable {
   override def references: AttributeSet = AttributeSet(VirtualColumn.groupingIdAttribute :: Nil)
   override def children: Seq[Expression] = groupByExprs
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/misc.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/misc.scala
index e8a3e129b4..eb8dc1423a 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/misc.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/misc.scala
@@ -467,8 +467,8 @@ object Murmur3HashFunction extends InterpretedHashFunction {
 }
 
 /**
-  * Print the result of an expression to stderr (used for debugging codegen).
-  */
+ * Print the result of an expression to stderr (used for debugging codegen).
+ */
 case class PrintToStderr(child: Expression) extends UnaryExpression {
 
   override def dataType: DataType = child.dataType
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
index a5ab390c76..69b09bcb35 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
@@ -31,9 +31,9 @@ import org.apache.spark.sql.catalyst.rules._
 import org.apache.spark.sql.types._
 
 /**
-  * Abstract class all optimizers should inherit of, contains the standard batches (extending
-  * Optimizers can override this.
-  */
+ * Abstract class all optimizers should inherit of, contains the standard batches (extending
+ * Optimizers can override this.
+ */
 abstract class Optimizer extends RuleExecutor[LogicalPlan] {
   def batches: Seq[Batch] = {
     // Technically some of the rules in Finish Analysis are not optimizer rules and belong more
@@ -111,11 +111,11 @@ abstract class Optimizer extends RuleExecutor[LogicalPlan] {
 }
 
 /**
-  * Non-abstract representation of the standard Spark optimizing strategies
-  *
-  * To ensure extendability, we leave the standard rules in the abstract optimizer rules, while
-  * specific rules go to the subclasses
-  */
+ * Non-abstract representation of the standard Spark optimizing strategies
+ *
+ * To ensure extendability, we leave the standard rules in the abstract optimizer rules, while
+ * specific rules go to the subclasses
+ */
 object DefaultOptimizer extends Optimizer
 
 /**
@@ -962,21 +962,21 @@ object PushPredicateThroughAggregate extends Rule[LogicalPlan] with PredicateHel
 }
 
 /**
-  * Reorder the joins and push all the conditions into join, so that the bottom ones have at least
-  * one condition.
-  *
-  * The order of joins will not be changed if all of them already have at least one condition.
-  */
+ * Reorder the joins and push all the conditions into join, so that the bottom ones have at least
+ * one condition.
+ *
+ * The order of joins will not be changed if all of them already have at least one condition.
+ */
 object ReorderJoin extends Rule[LogicalPlan] with PredicateHelper {
 
   /**
-    * Join a list of plans together and push down the conditions into them.
-    *
-    * The joined plan are picked from left to right, prefer those has at least one join condition.
-    *
-    * @param input a list of LogicalPlans to join.
-    * @param conditions a list of condition for join.
-    */
+   * Join a list of plans together and push down the conditions into them.
+   *
+   * The joined plan are picked from left to right, prefer those has at least one join condition.
+   *
+   * @param input a list of LogicalPlans to join.
+   * @param conditions a list of condition for join.
+   */
   @tailrec
   def createOrderedJoin(input: Seq[LogicalPlan], conditions: Seq[Expression]): LogicalPlan = {
     assert(input.size >= 2)
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
index c350f3049f..8541b1f7c6 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
@@ -1430,8 +1430,8 @@ class AstBuilder extends SqlBaseBaseVisitor[AnyRef] with Logging {
   }
 
   /**
-    * Create a [[StructType]] from a sequence of [[StructField]]s.
-    */
+   * Create a [[StructType]] from a sequence of [[StructField]]s.
+   */
   protected def createStructType(ctx: ColTypeListContext): StructType = {
     StructType(Option(ctx).toSeq.flatMap(visitColTypeList))
   }
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
index 28d2c445b1..6f35d87ebb 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
@@ -140,20 +140,20 @@ object ExtractEquiJoinKeys extends Logging with PredicateHelper {
 }
 
 /**
-  * A pattern that collects the filter and inner joins.
-  *
-  *          Filter
-  *            |
-  *        inner Join
-  *          /    \            ---->      (Seq(plan0, plan1, plan2), conditions)
-  *      Filter   plan2
-  *        |
-  *  inner join
-  *      /    \
-  *   plan0    plan1
-  *
-  * Note: This pattern currently only works for left-deep trees.
-  */
+ * A pattern that collects the filter and inner joins.
+ *
+ *          Filter
+ *            |
+ *        inner Join
+ *          /    \            ---->      (Seq(plan0, plan1, plan2), conditions)
+ *      Filter   plan2
+ *        |
+ *  inner join
+ *      /    \
+ *   plan0    plan1
+ *
+ * Note: This pattern currently only works for left-deep trees.
+ */
 object ExtractFiltersAndInnerJoins extends PredicateHelper {
 
   // flatten all inner joins, which are next to each other
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala
index 22a4461e66..609a33e2f1 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala
@@ -122,8 +122,8 @@ abstract class QueryPlan[PlanType <: QueryPlan[PlanType]] extends TreeNode[PlanT
     AttributeSet(children.flatMap(_.asInstanceOf[QueryPlan[PlanType]].output))
 
   /**
-    * The set of all attributes that are produced by this node.
-    */
+   * The set of all attributes that are produced by this node.
+   */
   def producedAttributes: AttributeSet = AttributeSet.empty
 
   /**
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala
index be9f1ffa22..d449088498 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/physical/partitioning.scala
@@ -76,9 +76,9 @@ case class OrderedDistribution(ordering: Seq[SortOrder]) extends Distribution {
 }
 
 /**
-  * Represents data where tuples are broadcasted to every node. It is quite common that the
-  * entire set of tuples is transformed into different data structure.
-  */
+ * Represents data where tuples are broadcasted to every node. It is quite common that the
+ * entire set of tuples is transformed into different data structure.
+ */
 case class BroadcastDistribution(mode: BroadcastMode) extends Distribution
 
 /**
diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/OptimizerExtendableSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/OptimizerExtendableSuite.scala
index 7e3da6bea7..6e5672ddc3 100644
--- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/OptimizerExtendableSuite.scala
+++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/OptimizerExtendableSuite.scala
@@ -23,21 +23,21 @@ import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
 import org.apache.spark.sql.catalyst.rules.Rule
 
 /**
-  * This is a test for SPARK-7727 if the Optimizer is kept being extendable
-  */
+ * This is a test for SPARK-7727 if the Optimizer is kept being extendable
+ */
 class OptimizerExtendableSuite extends SparkFunSuite {
 
   /**
-    * Dummy rule for test batches
-    */
+   * Dummy rule for test batches
+   */
   object DummyRule extends Rule[LogicalPlan] {
     def apply(p: LogicalPlan): LogicalPlan = p
   }
 
   /**
-    * This class represents a dummy extended optimizer that takes the batches of the
-    * Optimizer and adds custom ones.
-    */
+   * This class represents a dummy extended optimizer that takes the batches of the
+   * Optimizer and adds custom ones.
+   */
   class ExtendedOptimizer extends Optimizer {
 
     // rules set to DummyRule, would not be executed anyways
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala b/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala
index 221782ee8f..d4290fee0a 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala
@@ -712,13 +712,13 @@ class SQLContext private[sql](
   }
 
   /**
-    * :: Experimental ::
-    * Creates a [[Dataset]] with a single [[LongType]] column named `id`, containing elements
-    * in an range from `start` to `end` (exclusive) with an step value.
-    *
-    * @since 2.0.0
-    * @group dataset
-    */
+   * :: Experimental ::
+   * Creates a [[Dataset]] with a single [[LongType]] column named `id`, containing elements
+   * in an range from `start` to `end` (exclusive) with an step value.
+   *
+   * @since 2.0.0
+   * @group dataset
+   */
   @Experimental
   def range(start: Long, end: Long, step: Long): Dataset[java.lang.Long] = {
     range(start, end, step, numPartitions = sparkContext.defaultParallelism)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/CacheManager.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/CacheManager.scala
index f3478a873a..124ec09efd 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/CacheManager.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/CacheManager.scala
@@ -109,9 +109,10 @@ private[sql] class CacheManager extends Logging {
     cachedData.remove(dataIndex)
   }
 
-  /** Tries to remove the data for the given [[Dataset]] from the cache
-    * if it's cached
-    */
+  /**
+   * Tries to remove the data for the given [[Dataset]] from the cache
+   * if it's cached
+   */
   private[sql] def tryUncacheQuery(
       query: Dataset[_],
       blocking: Boolean = true): Boolean = writeLock {
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala
index b1b3d4ac81..ff19d1be1c 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala
@@ -84,8 +84,8 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
   private[sql] def metrics: Map[String, SQLMetric[_, _]] = Map.empty
 
   /**
-    * Reset all the metrics.
-    */
+   * Reset all the metrics.
+   */
   private[sql] def resetMetrics(): Unit = {
     metrics.valuesIterator.foreach(_.reset())
   }
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala
index 9bdf611f6e..9f539c4929 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala
@@ -31,8 +31,8 @@ import org.apache.spark.sql.execution.metric.{LongSQLMetricValue, SQLMetrics}
 import org.apache.spark.sql.internal.SQLConf
 
 /**
-  * An interface for those physical operators that support codegen.
-  */
+ * An interface for those physical operators that support codegen.
+ */
 trait CodegenSupport extends SparkPlan {
 
   /** Prefix used in the current operator's variable names. */
@@ -46,10 +46,10 @@ trait CodegenSupport extends SparkPlan {
   }
 
   /**
-    * Creates a metric using the specified name.
-    *
-    * @return name of the variable representing the metric
-    */
+   * Creates a metric using the specified name.
+   *
+   * @return name of the variable representing the metric
+   */
   def metricTerm(ctx: CodegenContext, name: String): String = {
     val metric = ctx.addReferenceObj(name, longMetric(name))
     val value = ctx.freshName("metricValue")
@@ -59,25 +59,25 @@ trait CodegenSupport extends SparkPlan {
   }
 
   /**
-    * Whether this SparkPlan support whole stage codegen or not.
-    */
+   * Whether this SparkPlan support whole stage codegen or not.
+   */
   def supportCodegen: Boolean = true
 
   /**
-    * Which SparkPlan is calling produce() of this one. It's itself for the first SparkPlan.
-    */
+   * Which SparkPlan is calling produce() of this one. It's itself for the first SparkPlan.
+   */
   protected var parent: CodegenSupport = null
 
   /**
-    * Returns all the RDDs of InternalRow which generates the input rows.
-    *
-    * Note: right now we support up to two RDDs.
-    */
+   * Returns all the RDDs of InternalRow which generates the input rows.
+   *
+   * Note: right now we support up to two RDDs.
+   */
   def upstreams(): Seq[RDD[InternalRow]]
 
   /**
-    * Returns Java source code to process the rows from upstream.
-    */
+   * Returns Java source code to process the rows from upstream.
+   */
   final def produce(ctx: CodegenContext, parent: CodegenSupport): String = {
     this.parent = parent
     ctx.freshNamePrefix = variablePrefix
@@ -89,28 +89,28 @@ trait CodegenSupport extends SparkPlan {
   }
 
   /**
-    * Generate the Java source code to process, should be overridden by subclass to support codegen.
-    *
-    * doProduce() usually generate the framework, for example, aggregation could generate this:
-    *
-    *   if (!initialized) {
-    *     # create a hash map, then build the aggregation hash map
-    *     # call child.produce()
-    *     initialized = true;
-    *   }
-    *   while (hashmap.hasNext()) {
-    *     row = hashmap.next();
-    *     # build the aggregation results
-    *     # create variables for results
-    *     # call consume(), which will call parent.doConsume()
+   * Generate the Java source code to process, should be overridden by subclass to support codegen.
+   *
+   * doProduce() usually generate the framework, for example, aggregation could generate this:
+   *
+   *   if (!initialized) {
+   *     # create a hash map, then build the aggregation hash map
+   *     # call child.produce()
+   *     initialized = true;
+   *   }
+   *   while (hashmap.hasNext()) {
+   *     row = hashmap.next();
+   *     # build the aggregation results
+   *     # create variables for results
+   *     # call consume(), which will call parent.doConsume()
    *      if (shouldStop()) return;
-    *   }
-    */
+   *   }
+   */
   protected def doProduce(ctx: CodegenContext): String
 
   /**
-    * Consume the generated columns or row from current SparkPlan, call it's parent's doConsume().
-    */
+   * Consume the generated columns or row from current SparkPlan, call it's parent's doConsume().
+   */
   final def consume(ctx: CodegenContext, outputVars: Seq[ExprCode], row: String = null): String = {
     val inputVars =
       if (row != null) {
@@ -158,9 +158,9 @@ trait CodegenSupport extends SparkPlan {
   }
 
   /**
-    * Returns source code to evaluate all the variables, and clear the code of them, to prevent
-    * them to be evaluated twice.
-    */
+   * Returns source code to evaluate all the variables, and clear the code of them, to prevent
+   * them to be evaluated twice.
+   */
   protected def evaluateVariables(variables: Seq[ExprCode]): String = {
     val evaluate = variables.filter(_.code != "").map(_.code.trim).mkString("\n")
     variables.foreach(_.code = "")
@@ -168,9 +168,9 @@ trait CodegenSupport extends SparkPlan {
   }
 
   /**
-    * Returns source code to evaluate the variables for required attributes, and clear the code
-    * of evaluated variables, to prevent them to be evaluated twice..
-    */
+   * Returns source code to evaluate the variables for required attributes, and clear the code
+   * of evaluated variables, to prevent them to be evaluated twice..
+   */
   protected def evaluateRequiredVariables(
       attributes: Seq[Attribute],
       variables: Seq[ExprCode],
@@ -194,18 +194,18 @@ trait CodegenSupport extends SparkPlan {
   def usedInputs: AttributeSet = references
 
   /**
-    * Generate the Java source code to process the rows from child SparkPlan.
-    *
-    * This should be override by subclass to support codegen.
-    *
-    * For example, Filter will generate the code like this:
-    *
-    *   # code to evaluate the predicate expression, result is isNull1 and value2
-    *   if (isNull1 || !value2) continue;
-    *   # call consume(), which will call parent.doConsume()
-    *
-    * Note: A plan can either consume the rows as UnsafeRow (row), or a list of variables (input).
-    */
+   * Generate the Java source code to process the rows from child SparkPlan.
+   *
+   * This should be override by subclass to support codegen.
+   *
+   * For example, Filter will generate the code like this:
+   *
+   *   # code to evaluate the predicate expression, result is isNull1 and value2
+   *   if (isNull1 || !value2) continue;
+   *   # call consume(), which will call parent.doConsume()
+   *
+   * Note: A plan can either consume the rows as UnsafeRow (row), or a list of variables (input).
+   */
   def doConsume(ctx: CodegenContext, input: Seq[ExprCode], row: ExprCode): String = {
     throw new UnsupportedOperationException
   }
@@ -213,11 +213,11 @@ trait CodegenSupport extends SparkPlan {
 
 
 /**
-  * InputAdapter is used to hide a SparkPlan from a subtree that support codegen.
-  *
-  * This is the leaf node of a tree with WholeStageCodegen, is used to generate code that consumes
-  * an RDD iterator of InternalRow.
-  */
+ * InputAdapter is used to hide a SparkPlan from a subtree that support codegen.
+ *
+ * This is the leaf node of a tree with WholeStageCodegen, is used to generate code that consumes
+ * an RDD iterator of InternalRow.
+ */
 case class InputAdapter(child: SparkPlan) extends UnaryNode with CodegenSupport {
 
   override def output: Seq[Attribute] = child.output
@@ -260,33 +260,33 @@ object WholeStageCodegen {
 }
 
 /**
-  * WholeStageCodegen compile a subtree of plans that support codegen together into single Java
-  * function.
-  *
-  * Here is the call graph of to generate Java source (plan A support codegen, but plan B does not):
-  *
-  *   WholeStageCodegen       Plan A               FakeInput        Plan B
-  * =========================================================================
-  *
-  * -> execute()
-  *     |
-  *  doExecute() --------->   upstreams() -------> upstreams() ------> execute()
-  *     |
-  *     +----------------->   produce()
-  *                             |
-  *                          doProduce()  -------> produce()
-  *                                                   |
-  *                                                doProduce()
-  *                                                   |
-  *                         doConsume() <--------- consume()
-  *                             |
-  *  doConsume()  <--------  consume()
-  *
-  * SparkPlan A should override doProduce() and doConsume().
-  *
-  * doCodeGen() will create a CodeGenContext, which will hold a list of variables for input,
-  * used to generated code for BoundReference.
-  */
+ * WholeStageCodegen compile a subtree of plans that support codegen together into single Java
+ * function.
+ *
+ * Here is the call graph of to generate Java source (plan A support codegen, but plan B does not):
+ *
+ *   WholeStageCodegen       Plan A               FakeInput        Plan B
+ * =========================================================================
+ *
+ * -> execute()
+ *     |
+ *  doExecute() --------->   upstreams() -------> upstreams() ------> execute()
+ *     |
+ *     +----------------->   produce()
+ *                             |
+ *                          doProduce()  -------> produce()
+ *                                                   |
+ *                                                doProduce()
+ *                                                   |
+ *                         doConsume() <--------- consume()
+ *                             |
+ *  doConsume()  <--------  consume()
+ *
+ * SparkPlan A should override doProduce() and doConsume().
+ *
+ * doCodeGen() will create a CodeGenContext, which will hold a list of variables for input,
+ * used to generated code for BoundReference.
+ */
 case class WholeStageCodegen(child: SparkPlan) extends UnaryNode with CodegenSupport {
 
   override def output: Seq[Attribute] = child.output
@@ -422,8 +422,8 @@ case class WholeStageCodegen(child: SparkPlan) extends UnaryNode with CodegenSup
 
 
 /**
-  * Find the chained plans that support codegen, collapse them together as WholeStageCodegen.
-  */
+ * Find the chained plans that support codegen, collapse them together as WholeStageCodegen.
+ */
 case class CollapseCodegenStages(conf: SQLConf) extends Rule[SparkPlan] {
 
   private def supportCodegen(e: Expression): Boolean = e match {
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala
index 7d0567842c..806089196c 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala
@@ -444,8 +444,8 @@ private[execution] final case class RangeBoundOrdering(
 }
 
 /**
-  * The interface of row buffer for a partition
-  */
+ * The interface of row buffer for a partition
+ */
 private[execution] abstract class RowBuffer {
 
   /** Number of rows. */
@@ -462,8 +462,8 @@ private[execution] abstract class RowBuffer {
 }
 
 /**
-  * A row buffer based on ArrayBuffer (the number of rows is limited)
-  */
+ * A row buffer based on ArrayBuffer (the number of rows is limited)
+ */
 private[execution] class ArrayRowBuffer(buffer: ArrayBuffer[UnsafeRow]) extends RowBuffer {
 
   private[this] var cursor: Int = -1
@@ -493,8 +493,8 @@ private[execution] class ArrayRowBuffer(buffer: ArrayBuffer[UnsafeRow]) extends
 }
 
 /**
-  * An external buffer of rows based on UnsafeExternalSorter
-  */
+ * An external buffer of rows based on UnsafeExternalSorter
+ */
 private[execution] class ExternalRowBuffer(sorter: UnsafeExternalSorter, numFields: Int)
   extends RowBuffer {
 
@@ -654,12 +654,16 @@ private[execution] final class SlidingWindowFunctionFrame(
   /** The rows within current sliding window. */
   private[this] val buffer = new util.ArrayDeque[InternalRow]()
 
-  /** Index of the first input row with a value greater than the upper bound of the current
-    * output row. */
+  /**
+   * Index of the first input row with a value greater than the upper bound of the current
+   * output row.
+   */
   private[this] var inputHighIndex = 0
 
-  /** Index of the first input row with a value equal to or greater than the lower bound of the
-    * current output row. */
+  /**
+   * Index of the first input row with a value equal to or greater than the lower bound of the
+   * current output row.
+   */
   private[this] var inputLowIndex = 0
 
   /** Prepare the frame for calculating a new partition. Reset all variables. */
@@ -763,8 +767,10 @@ private[execution] final class UnboundedPrecedingWindowFunctionFrame(
   /** The next row from `input`. */
   private[this] var nextRow: InternalRow = null
 
-  /** Index of the first input row with a value greater than the upper bound of the current
-   * output row. */
+  /**
+   * Index of the first input row with a value greater than the upper bound of the current
+   * output row.
+   */
   private[this] var inputIndex = 0
 
   /** Prepare the frame for calculating a new partition. */
@@ -819,8 +825,10 @@ private[execution] final class UnboundedFollowingWindowFunctionFrame(
   /** Rows of the partition currently being processed. */
   private[this] var input: RowBuffer = null
 
-  /** Index of the first input row with a value equal to or greater than the lower bound of the
-   * current output row. */
+  /**
+   * Index of the first input row with a value equal to or greater than the lower bound of the
+   * current output row.
+   */
   private[this] var inputIndex = 0
 
   /** Prepare the frame for calculating a new partition. */
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/AggregationIterator.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/AggregationIterator.scala
index 15627a7004..042c731901 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/AggregationIterator.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/AggregationIterator.scala
@@ -47,17 +47,17 @@ abstract class AggregationIterator(
   ///////////////////////////////////////////////////////////////////////////
 
   /**
-    * The following combinations of AggregationMode are supported:
-    * - Partial
-    * - PartialMerge (for single distinct)
-    * - Partial and PartialMerge (for single distinct)
-    * - Final
-    * - Complete (for SortBasedAggregate with functions that does not support Partial)
-    * - Final and Complete (currently not used)
-    *
-    * TODO: AggregateMode should have only two modes: Update and Merge, AggregateExpression
-    * could have a flag to tell it's final or not.
-    */
+   * The following combinations of AggregationMode are supported:
+   * - Partial
+   * - PartialMerge (for single distinct)
+   * - Partial and PartialMerge (for single distinct)
+   * - Final
+   * - Complete (for SortBasedAggregate with functions that does not support Partial)
+   * - Final and Complete (currently not used)
+   *
+   * TODO: AggregateMode should have only two modes: Update and Merge, AggregateExpression
+   * could have a flag to tell it's final or not.
+   */
   {
     val modes = aggregateExpressions.map(_.mode).distinct.toSet
     require(modes.size <= 2,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/SortBasedAggregationIterator.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/SortBasedAggregationIterator.scala
index 8f974980bb..de1491d357 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/SortBasedAggregationIterator.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/SortBasedAggregationIterator.scala
@@ -46,9 +46,9 @@ class SortBasedAggregationIterator(
     newMutableProjection) {
 
   /**
-    * Creates a new aggregation buffer and initializes buffer values
-    * for all aggregate functions.
-    */
+   * Creates a new aggregation buffer and initializes buffer values
+   * for all aggregate functions.
+   */
   private def newBuffer: MutableRow = {
     val bufferSchema = aggregateFunctions.flatMap(_.aggBufferAttributes)
     val bufferRowSize: Int = bufferSchema.length
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/TungstenAggregate.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/TungstenAggregate.scala
index 7c215d1b96..60027edc7c 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/TungstenAggregate.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/TungstenAggregate.scala
@@ -266,8 +266,8 @@ case class TungstenAggregate(
   private var sorterTerm: String = _
 
   /**
-    * This is called by generated Java class, should be public.
-    */
+   * This is called by generated Java class, should be public.
+   */
   def createHashMap(): UnsafeFixedWidthAggregationMap = {
     // create initialized aggregate buffer
     val initExpr = declFunctions.flatMap(f => f.initialValues)
@@ -286,15 +286,15 @@ case class TungstenAggregate(
   }
 
   /**
-    * This is called by generated Java class, should be public.
-    */
+   * This is called by generated Java class, should be public.
+   */
   def createUnsafeJoiner(): UnsafeRowJoiner = {
     GenerateUnsafeRowJoiner.create(groupingKeySchema, bufferSchema)
   }
 
   /**
-    * Called by generated Java class to finish the aggregate and return a KVIterator.
-    */
+   * Called by generated Java class to finish the aggregate and return a KVIterator.
+   */
   def finishAggregate(
       hashMap: UnsafeFixedWidthAggregationMap,
       sorter: UnsafeKVExternalSorter): KVIterator[UnsafeRow, UnsafeRow] = {
@@ -372,8 +372,8 @@ case class TungstenAggregate(
   }
 
   /**
-    * Generate the code for output.
-    */
+   * Generate the code for output.
+   */
   private def generateResultCode(
       ctx: CodegenContext,
       keyTerm: String,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala
index f3514cd14c..159fdc99dd 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala
@@ -168,10 +168,10 @@ private[spark] class SqlNewHadoopRDD[V: ClassTag](
       private[this] var reader: RecordReader[Void, V] = null
 
       /**
-        * If the format is ParquetInputFormat, try to create the optimized RecordReader. If this
-        * fails (for example, unsupported schema), try with the normal reader.
-        * TODO: plumb this through a different way?
-        */
+       * If the format is ParquetInputFormat, try to create the optimized RecordReader. If this
+       * fails (for example, unsupported schema), try with the normal reader.
+       * TODO: plumb this through a different way?
+       */
       if (enableVectorizedParquetReader &&
         format.getClass.getName == "org.apache.parquet.hadoop.ParquetInputFormat") {
         val parquetReader: VectorizedParquetRecordReader = new VectorizedParquetRecordReader()
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchema.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchema.scala
index 797f740dc5..ea843a1013 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchema.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchema.scala
@@ -33,11 +33,11 @@ import org.apache.spark.unsafe.types.UTF8String
 private[csv] object CSVInferSchema {
 
   /**
-    * Similar to the JSON schema inference
-    *     1. Infer type of each row
-    *     2. Merge row types to find common type
-    *     3. Replace any null types with string type
-    */
+   * Similar to the JSON schema inference
+   *     1. Infer type of each row
+   *     2. Merge row types to find common type
+   *     3. Replace any null types with string type
+   */
   def infer(
       tokenRdd: RDD[Array[String]],
       header: Array[String],
@@ -75,9 +75,9 @@ private[csv] object CSVInferSchema {
   }
 
   /**
-    * Infer type of string field. Given known type Double, and a string "1", there is no
-    * point checking if it is an Int, as the final type must be Double or higher.
-    */
+   * Infer type of string field. Given known type Double, and a string "1", there is no
+   * point checking if it is an Int, as the final type must be Double or higher.
+   */
   def inferField(typeSoFar: DataType, field: String, nullValue: String = ""): DataType = {
     if (field == null || field.isEmpty || field == nullValue) {
       typeSoFar
@@ -142,9 +142,9 @@ private[csv] object CSVInferSchema {
   private val numericPrecedence: IndexedSeq[DataType] = HiveTypeCoercion.numericPrecedence
 
   /**
-    * Copied from internal Spark api
-    * [[org.apache.spark.sql.catalyst.analysis.HiveTypeCoercion]]
-    */
+   * Copied from internal Spark api
+   * [[org.apache.spark.sql.catalyst.analysis.HiveTypeCoercion]]
+   */
   val findTightestCommonType: (DataType, DataType) => Option[DataType] = {
     case (t1, t2) if t1 == t2 => Some(t1)
     case (NullType, t1) => Some(t1)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala
index c0d6f6fbf7..34fcbdf871 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala
@@ -38,8 +38,8 @@ import org.apache.spark.util.SerializableConfiguration
 import org.apache.spark.util.collection.BitSet
 
 /**
-  * Provides access to CSV data from pure SQL statements.
-  */
+ * Provides access to CSV data from pure SQL statements.
+ */
 class DefaultSource extends FileFormat with DataSourceRegister {
 
   override def shortName(): String = "csv"
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/ddl.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/ddl.scala
index 877e159fbd..2e88d588be 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/ddl.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/ddl.scala
@@ -51,11 +51,11 @@ case class DescribeCommand(
 }
 
 /**
-  * Used to represent the operation of create table using a data source.
+ * Used to represent the operation of create table using a data source.
  *
-  * @param allowExisting If it is true, we will do nothing when the table already exists.
-  *                      If it is false, an exception will be thrown
-  */
+ * @param allowExisting If it is true, we will do nothing when the table already exists.
+ *                      If it is false, an exception will be thrown
+ */
 case class CreateTableUsing(
     tableIdent: TableIdentifier,
     userSpecifiedSchema: Option[StructType],
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
index 0ed1ed41b0..41e566c27b 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
@@ -122,8 +122,8 @@ case class BroadcastHashJoin(
   }
 
   /**
-    * Returns a tuple of Broadcast of HashedRelation and the variable name for it.
-    */
+   * Returns a tuple of Broadcast of HashedRelation and the variable name for it.
+   */
   private def prepareBroadcast(ctx: CodegenContext): (Broadcast[HashedRelation], String) = {
     // create a name for HashedRelation
     val broadcastRelation = buildPlan.executeBroadcast[HashedRelation]()
@@ -139,9 +139,9 @@ case class BroadcastHashJoin(
   }
 
   /**
-    * Returns the code for generating join key for stream side, and expression of whether the key
-    * has any null in it or not.
-    */
+   * Returns the code for generating join key for stream side, and expression of whether the key
+   * has any null in it or not.
+   */
   private def genStreamSideJoinKey(
       ctx: CodegenContext,
       input: Seq[ExprCode]): (ExprCode, String) = {
@@ -160,8 +160,8 @@ case class BroadcastHashJoin(
   }
 
   /**
-    * Generates the code for variable of build side.
-    */
+   * Generates the code for variable of build side.
+   */
   private def genBuildSideVars(ctx: CodegenContext, matched: String): Seq[ExprCode] = {
     ctx.currentVars = null
     ctx.INPUT_ROW = matched
@@ -188,8 +188,8 @@ case class BroadcastHashJoin(
   }
 
   /**
-    * Generates the code for Inner join.
-    */
+   * Generates the code for Inner join.
+   */
   private def codegenInner(ctx: CodegenContext, input: Seq[ExprCode]): String = {
     val (broadcastRelation, relationTerm) = prepareBroadcast(ctx)
     val (keyEv, anyNull) = genStreamSideJoinKey(ctx, input)
@@ -254,8 +254,8 @@ case class BroadcastHashJoin(
 
 
   /**
-    * Generates the code for left or right outer join.
-    */
+   * Generates the code for left or right outer join.
+   */
   private def codegenOuter(ctx: CodegenContext, input: Seq[ExprCode]): String = {
     val (broadcastRelation, relationTerm) = prepareBroadcast(ctx)
     val (keyEv, anyNull) = genStreamSideJoinKey(ctx, input)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/CartesianProduct.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/CartesianProduct.scala
index fb65b50da8..edb4c5a16f 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/CartesianProduct.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/CartesianProduct.scala
@@ -28,10 +28,10 @@ import org.apache.spark.util.CompletionIterator
 import org.apache.spark.util.collection.unsafe.sort.UnsafeExternalSorter
 
 /**
-  * An optimized CartesianRDD for UnsafeRow, which will cache the rows from second child RDD,
-  * will be much faster than building the right partition for every row in left RDD, it also
-  * materialize the right RDD (in case of the right RDD is nondeterministic).
-  */
+ * An optimized CartesianRDD for UnsafeRow, which will cache the rows from second child RDD,
+ * will be much faster than building the right partition for every row in left RDD, it also
+ * materialize the right RDD (in case of the right RDD is nondeterministic).
+ */
 private[spark]
 class UnsafeCartesianRDD(left : RDD[UnsafeRow], right : RDD[UnsafeRow], numFieldsOfRight: Int)
   extends CartesianRDD[UnsafeRow, UnsafeRow](left.sparkContext, left, right) {
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashJoin.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashJoin.scala
index 5f42d07273..c298b7dee0 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashJoin.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashJoin.scala
@@ -64,10 +64,10 @@ trait HashJoin {
   }
 
   /**
-    * Try to rewrite the key as LongType so we can use getLong(), if they key can fit with a long.
-    *
-    * If not, returns the original expressions.
-    */
+   * Try to rewrite the key as LongType so we can use getLong(), if they key can fit with a long.
+   *
+   * If not, returns the original expressions.
+   */
   def rewriteKeyExpr(keys: Seq[Expression]): Seq[Expression] = {
     var keyExpr: Expression = null
     var width = 0
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashedRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashedRelation.scala
index dc4793e85a..91c470d187 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashedRelation.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashedRelation.scala
@@ -38,20 +38,20 @@ import org.apache.spark.util.collection.CompactBuffer
  */
 private[execution] sealed trait HashedRelation {
   /**
-    * Returns matched rows.
-    */
+   * Returns matched rows.
+   */
   def get(key: InternalRow): Seq[InternalRow]
 
   /**
-    * Returns matched rows for a key that has only one column with LongType.
-    */
+   * Returns matched rows for a key that has only one column with LongType.
+   */
   def get(key: Long): Seq[InternalRow] = {
     throw new UnsupportedOperationException
   }
 
   /**
-    * Returns the size of used memory.
-    */
+   * Returns the size of used memory.
+   */
   def getMemorySize: Long = 1L  // to make the test happy
 
   /**
@@ -77,20 +77,20 @@ private[execution] sealed trait HashedRelation {
 }
 
 /**
-  * Interface for a hashed relation that have only one row per key.
-  *
-  * We should call getValue() for better performance.
-  */
+ * Interface for a hashed relation that have only one row per key.
+ *
+ * We should call getValue() for better performance.
+ */
 private[execution] trait UniqueHashedRelation extends HashedRelation {
 
   /**
-    * Returns the matched single row.
-    */
+   * Returns the matched single row.
+   */
   def getValue(key: InternalRow): InternalRow
 
   /**
-    * Returns the matched single row with key that have only one column of LongType.
-    */
+   * Returns the matched single row with key that have only one column of LongType.
+   */
   def getValue(key: Long): InternalRow = {
     throw new UnsupportedOperationException
   }
@@ -345,8 +345,8 @@ private[joins] object UnsafeHashedRelation {
 }
 
 /**
-  * An interface for a hashed relation that the key is a Long.
-  */
+ * An interface for a hashed relation that the key is a Long.
+ */
 private[joins] trait LongHashedRelation extends HashedRelation {
   override def get(key: InternalRow): Seq[InternalRow] = {
     get(key.getLong(0))
@@ -396,26 +396,26 @@ private[joins] final class UniqueLongHashedRelation(
 }
 
 /**
-  * A relation that pack all the rows into a byte array, together with offsets and sizes.
-  *
-  * All the bytes of UnsafeRow are packed together as `bytes`:
-  *
-  *  [  Row0  ][  Row1  ][] ... [  RowN  ]
-  *
-  * With keys:
-  *
-  *   start    start+1   ...       start+N
-  *
-  * `offsets` are offsets of UnsafeRows in the `bytes`
-  * `sizes` are the numbers of bytes of UnsafeRows, 0 means no row for this key.
-  *
-  *  For example, two UnsafeRows (24 bytes and 32 bytes), with keys as 3 and 5 will stored as:
-  *
-  *  start   = 3
-  *  offsets = [0, 0, 24]
-  *  sizes   = [24, 0, 32]
-  *  bytes   = [0 - 24][][24 - 56]
-  */
+ * A relation that pack all the rows into a byte array, together with offsets and sizes.
+ *
+ * All the bytes of UnsafeRow are packed together as `bytes`:
+ *
+ *  [  Row0  ][  Row1  ][] ... [  RowN  ]
+ *
+ * With keys:
+ *
+ *   start    start+1   ...       start+N
+ *
+ * `offsets` are offsets of UnsafeRows in the `bytes`
+ * `sizes` are the numbers of bytes of UnsafeRows, 0 means no row for this key.
+ *
+ *  For example, two UnsafeRows (24 bytes and 32 bytes), with keys as 3 and 5 will stored as:
+ *
+ *  start   = 3
+ *  offsets = [0, 0, 24]
+ *  sizes   = [24, 0, 32]
+ *  bytes   = [0 - 24][][24 - 56]
+ */
 private[joins] final class LongArrayRelation(
     private var numFields: Int,
     private var start: Long,
@@ -483,8 +483,8 @@ private[joins] final class LongArrayRelation(
 }
 
 /**
-  * Create hashed relation with key that is long.
-  */
+ * Create hashed relation with key that is long.
+ */
 private[joins] object LongHashedRelation {
 
   val DENSE_FACTOR = 0.2
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/SortMergeJoin.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/SortMergeJoin.scala
index 60bd8ea39a..0e7b2f2f31 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/SortMergeJoin.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/SortMergeJoin.scala
@@ -256,9 +256,9 @@ case class SortMergeJoin(
   }
 
   /**
-    * Generate a function to scan both left and right to find a match, returns the term for
-    * matched one row from left side and buffered rows from right side.
-    */
+   * Generate a function to scan both left and right to find a match, returns the term for
+   * matched one row from left side and buffered rows from right side.
+   */
   private def genScanner(ctx: CodegenContext): (String, String) = {
     // Create class member for next row from both sides.
     val leftRow = ctx.freshName("leftRow")
@@ -341,12 +341,12 @@ case class SortMergeJoin(
   }
 
   /**
-    * Creates variables for left part of result row.
-    *
-    * In order to defer the access after condition and also only access once in the loop,
-    * the variables should be declared separately from accessing the columns, we can't use the
-    * codegen of BoundReference here.
-    */
+   * Creates variables for left part of result row.
+   *
+   * In order to defer the access after condition and also only access once in the loop,
+   * the variables should be declared separately from accessing the columns, we can't use the
+   * codegen of BoundReference here.
+   */
   private def createLeftVars(ctx: CodegenContext, leftRow: String): Seq[ExprCode] = {
     ctx.INPUT_ROW = leftRow
     left.output.zipWithIndex.map { case (a, i) =>
@@ -370,9 +370,9 @@ case class SortMergeJoin(
   }
 
   /**
-    * Creates the variables for right part of result row, using BoundReference, since the right
-    * part are accessed inside the loop.
-    */
+   * Creates the variables for right part of result row, using BoundReference, since the right
+   * part are accessed inside the loop.
+   */
   private def createRightVar(ctx: CodegenContext, rightRow: String): Seq[ExprCode] = {
     ctx.INPUT_ROW = rightRow
     right.output.zipWithIndex.map { case (a, i) =>
@@ -381,12 +381,12 @@ case class SortMergeJoin(
   }
 
   /**
-    * Splits variables based on whether it's used by condition or not, returns the code to create
-    * these variables before the condition and after the condition.
-    *
-    * Only a few columns are used by condition, then we can skip the accessing of those columns
-    * that are not used by condition also filtered out by condition.
-    */
+   * Splits variables based on whether it's used by condition or not, returns the code to create
+   * these variables before the condition and after the condition.
+   *
+   * Only a few columns are used by condition, then we can skip the accessing of those columns
+   * that are not used by condition also filtered out by condition.
+   */
   private def splitVarsByCondition(
       attributes: Seq[Attribute],
       variables: Seq[ExprCode]): (String, String) = {
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala
index 998eb82de1..8ece3c971a 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala
@@ -468,10 +468,10 @@ private[state] class HDFSBackedStateStoreProvider(
   }
 
   /**
-    * Clean up old snapshots and delta files that are not needed any more. It ensures that last
-    * few versions of the store can be recovered from the files, so re-executed RDD operations
-    * can re-apply updates on the past versions of the store.
-    */
+   * Clean up old snapshots and delta files that are not needed any more. It ensures that last
+   * few versions of the store can be recovered from the files, so re-executed RDD operations
+   * can re-apply updates on the past versions of the store.
+   */
   private[state] def cleanup(): Unit = {
     try {
       val files = fetchFiles()
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala
index 24a01f5be1..012b125d6b 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala
@@ -45,8 +45,8 @@ private[ui] case class SparkPlanGraph(
   }
 
   /**
-    * All the SparkPlanGraphNodes, including those inside of WholeStageCodegen.
-    */
+   * All the SparkPlanGraphNodes, including those inside of WholeStageCodegen.
+   */
   val allNodes: Seq[SparkPlanGraphNode] = {
     nodes.flatMap {
       case cluster: SparkPlanGraphCluster => cluster.nodes :+ cluster
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
index baf947d037..da58ba2add 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
@@ -332,95 +332,94 @@ object functions {
   }
 
   /**
-    * Aggregate function: returns the first value in a group.
-    *
-    * The function by default returns the first values it sees. It will return the first non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: returns the first value in a group.
+   *
+   * The function by default returns the first values it sees. It will return the first non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def first(e: Column, ignoreNulls: Boolean): Column = withAggregateFunction {
     new First(e.expr, Literal(ignoreNulls))
   }
 
   /**
-    * Aggregate function: returns the first value of a column in a group.
-    *
-    * The function by default returns the first values it sees. It will return the first non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: returns the first value of a column in a group.
+   *
+   * The function by default returns the first values it sees. It will return the first non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def first(columnName: String, ignoreNulls: Boolean): Column = {
     first(Column(columnName), ignoreNulls)
   }
 
   /**
-    * Aggregate function: returns the first value in a group.
-    *
-    * The function by default returns the first values it sees. It will return the first non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 1.3.0
-    */
+   * Aggregate function: returns the first value in a group.
+   *
+   * The function by default returns the first values it sees. It will return the first non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 1.3.0
+   */
   def first(e: Column): Column = first(e, ignoreNulls = false)
 
   /**
-    * Aggregate function: returns the first value of a column in a group.
-    *
-    * The function by default returns the first values it sees. It will return the first non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 1.3.0
-    */
+   * Aggregate function: returns the first value of a column in a group.
+   *
+   * The function by default returns the first values it sees. It will return the first non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 1.3.0
+   */
   def first(columnName: String): Column = first(Column(columnName))
 
-
   /**
-    * Aggregate function: indicates whether a specified column in a GROUP BY list is aggregated
-    * or not, returns 1 for aggregated or 0 for not aggregated in the result set.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: indicates whether a specified column in a GROUP BY list is aggregated
+   * or not, returns 1 for aggregated or 0 for not aggregated in the result set.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def grouping(e: Column): Column = Column(Grouping(e.expr))
 
   /**
-    * Aggregate function: indicates whether a specified column in a GROUP BY list is aggregated
-    * or not, returns 1 for aggregated or 0 for not aggregated in the result set.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: indicates whether a specified column in a GROUP BY list is aggregated
+   * or not, returns 1 for aggregated or 0 for not aggregated in the result set.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def grouping(columnName: String): Column = grouping(Column(columnName))
 
   /**
-    * Aggregate function: returns the level of grouping, equals to
-    *
-    *   (grouping(c1) << (n-1)) + (grouping(c2) << (n-2)) + ... + grouping(cn)
-    *
-    * Note: the list of columns should match with grouping columns exactly, or empty (means all the
-    * grouping columns).
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: returns the level of grouping, equals to
+   *
+   *   (grouping(c1) << (n-1)) + (grouping(c2) << (n-2)) + ... + grouping(cn)
+   *
+   * Note: the list of columns should match with grouping columns exactly, or empty (means all the
+   * grouping columns).
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def grouping_id(cols: Column*): Column = Column(GroupingID(cols.map(_.expr)))
 
   /**
-    * Aggregate function: returns the level of grouping, equals to
-    *
-    *   (grouping(c1) << (n-1)) + (grouping(c2) << (n-2)) + ... + grouping(cn)
-    *
-    * Note: the list of columns should match with grouping columns exactly.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: returns the level of grouping, equals to
+   *
+   *   (grouping(c1) << (n-1)) + (grouping(c2) << (n-2)) + ... + grouping(cn)
+   *
+   * Note: the list of columns should match with grouping columns exactly.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def grouping_id(colName: String, colNames: String*): Column = {
     grouping_id((Seq(colName) ++ colNames).map(n => Column(n)) : _*)
   }
@@ -442,51 +441,51 @@ object functions {
   def kurtosis(columnName: String): Column = kurtosis(Column(columnName))
 
   /**
-    * Aggregate function: returns the last value in a group.
-    *
-    * The function by default returns the last values it sees. It will return the last non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: returns the last value in a group.
+   *
+   * The function by default returns the last values it sees. It will return the last non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def last(e: Column, ignoreNulls: Boolean): Column = withAggregateFunction {
     new Last(e.expr, Literal(ignoreNulls))
   }
 
   /**
-    * Aggregate function: returns the last value of the column in a group.
-    *
-    * The function by default returns the last values it sees. It will return the last non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 2.0.0
-    */
+   * Aggregate function: returns the last value of the column in a group.
+   *
+   * The function by default returns the last values it sees. It will return the last non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 2.0.0
+   */
   def last(columnName: String, ignoreNulls: Boolean): Column = {
     last(Column(columnName), ignoreNulls)
   }
 
   /**
-    * Aggregate function: returns the last value in a group.
-    *
-    * The function by default returns the last values it sees. It will return the last non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 1.3.0
-    */
+   * Aggregate function: returns the last value in a group.
+   *
+   * The function by default returns the last values it sees. It will return the last non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 1.3.0
+   */
   def last(e: Column): Column = last(e, ignoreNulls = false)
 
   /**
-    * Aggregate function: returns the last value of the column in a group.
-    *
-    * The function by default returns the last values it sees. It will return the last non-null
-    * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
-    *
-    * @group agg_funcs
-    * @since 1.3.0
-    */
+   * Aggregate function: returns the last value of the column in a group.
+   *
+   * The function by default returns the last values it sees. It will return the last non-null
+   * value it sees when ignoreNulls is set to true. If all values are null, then null is returned.
+   *
+   * @group agg_funcs
+   * @since 1.3.0
+   */
   def last(columnName: String): Column = last(Column(columnName), ignoreNulls = false)
 
   /**
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
index e8834d052c..14e14710f6 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
@@ -152,19 +152,19 @@ trait StreamSinkProvider {
 @DeveloperApi
 trait CreatableRelationProvider {
   /**
-    * Creates a relation with the given parameters based on the contents of the given
-    * DataFrame. The mode specifies the expected behavior of createRelation when
-    * data already exists.
-    * Right now, there are three modes, Append, Overwrite, and ErrorIfExists.
-    * Append mode means that when saving a DataFrame to a data source, if data already exists,
-    * contents of the DataFrame are expected to be appended to existing data.
-    * Overwrite mode means that when saving a DataFrame to a data source, if data already exists,
-    * existing data is expected to be overwritten by the contents of the DataFrame.
-    * ErrorIfExists mode means that when saving a DataFrame to a data source,
-    * if data already exists, an exception is expected to be thrown.
-     *
-     * @since 1.3.0
-    */
+   * Creates a relation with the given parameters based on the contents of the given
+   * DataFrame. The mode specifies the expected behavior of createRelation when
+   * data already exists.
+   * Right now, there are three modes, Append, Overwrite, and ErrorIfExists.
+   * Append mode means that when saving a DataFrame to a data source, if data already exists,
+   * contents of the DataFrame are expected to be appended to existing data.
+   * Overwrite mode means that when saving a DataFrame to a data source, if data already exists,
+   * existing data is expected to be overwritten by the contents of the DataFrame.
+   * ErrorIfExists mode means that when saving a DataFrame to a data source,
+   * if data already exists, an exception is expected to be thrown.
+   *
+   * @since 1.3.0
+   */
   def createRelation(
       sqlContext: SQLContext,
       mode: SaveMode,
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala b/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala
index 854a662cc4..d160f8ab8c 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala
@@ -286,8 +286,8 @@ abstract class QueryTest extends PlanTest {
   }
 
   /**
-    * Asserts that a given [[Dataset]] does not have missing inputs in all the analyzed plans.
-    */
+   * Asserts that a given [[Dataset]] does not have missing inputs in all the analyzed plans.
+   */
   def assertEmptyMissingInput(query: Dataset[_]): Unit = {
     assert(query.queryExecution.analyzed.missingInput.isEmpty,
       s"The analyzed logical plan has missing inputs: ${query.queryExecution.analyzed}")
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/BenchmarkWholeStageCodegen.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/BenchmarkWholeStageCodegen.scala
index 55906793c0..289e1b6db9 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/BenchmarkWholeStageCodegen.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/BenchmarkWholeStageCodegen.scala
@@ -32,10 +32,10 @@ import org.apache.spark.unsafe.map.BytesToBytesMap
 import org.apache.spark.util.Benchmark
 
 /**
-  * Benchmark to measure whole stage codegen performance.
-  * To run this:
-  *  build/sbt "sql/test-only *BenchmarkWholeStageCodegen"
-  */
+ * Benchmark to measure whole stage codegen performance.
+ * To run this:
+ *  build/sbt "sql/test-only *BenchmarkWholeStageCodegen"
+ */
 class BenchmarkWholeStageCodegen extends SparkFunSuite {
   lazy val conf = new SparkConf().setMaster("local[1]").setAppName("benchmark")
     .set("spark.sql.shuffle.partitions", "1")
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVParserSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVParserSuite.scala
index dc54883277..aaeecef5f3 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVParserSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVParserSuite.scala
@@ -20,8 +20,8 @@ package org.apache.spark.sql.execution.datasources.csv
 import org.apache.spark.SparkFunSuite
 
 /**
-  * test cases for StringIteratorReader
-  */
+ * test cases for StringIteratorReader
+ */
 class CSVParserSuite extends SparkFunSuite {
 
   private def readAll(iter: Iterator[String]) = {
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala b/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala
index c1e151d08b..ac37e8e022 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/StreamingContext.scala
@@ -497,9 +497,10 @@ class StreamingContext private[streaming] (
     new TransformedDStream[T](dstreams, sparkContext.clean(transformFunc))
   }
 
-  /** Add a [[org.apache.spark.streaming.scheduler.StreamingListener]] object for
-    * receiving system events related to streaming.
-    */
+  /**
+   * Add a [[org.apache.spark.streaming.scheduler.StreamingListener]] object for
+   * receiving system events related to streaming.
+   */
   def addStreamingListener(streamingListener: StreamingListener) {
     scheduler.listenerBus.addListener(streamingListener)
   }
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/api/java/JavaStreamingContext.scala b/streaming/src/main/scala/org/apache/spark/streaming/api/java/JavaStreamingContext.scala
index 05f4da6fac..922e4a5e4d 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/api/java/JavaStreamingContext.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/api/java/JavaStreamingContext.scala
@@ -517,9 +517,10 @@ class JavaStreamingContext(val ssc: StreamingContext) extends Closeable {
     ssc.remember(duration)
   }
 
-  /** Add a [[org.apache.spark.streaming.scheduler.StreamingListener]] object for
-    * receiving system events related to streaming.
-    */
+  /**
+   * Add a [[org.apache.spark.streaming.scheduler.StreamingListener]] object for
+   * receiving system events related to streaming.
+   */
   def addStreamingListener(streamingListener: StreamingListener) {
     ssc.addStreamingListener(streamingListener)
   }
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/receiver/RateLimiter.scala b/streaming/src/main/scala/org/apache/spark/streaming/receiver/RateLimiter.scala
index 0a861f22b1..fbac4880bd 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/receiver/RateLimiter.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/receiver/RateLimiter.scala
@@ -22,17 +22,18 @@ import com.google.common.util.concurrent.{RateLimiter => GuavaRateLimiter}
 import org.apache.spark.SparkConf
 import org.apache.spark.internal.Logging
 
-/** Provides waitToPush() method to limit the rate at which receivers consume data.
-  *
-  * waitToPush method will block the thread if too many messages have been pushed too quickly,
-  * and only return when a new message has been pushed. It assumes that only one message is
-  * pushed at a time.
-  *
-  * The spark configuration spark.streaming.receiver.maxRate gives the maximum number of messages
-  * per second that each receiver will accept.
-  *
-  * @param conf spark configuration
-  */
+/**
+ * Provides waitToPush() method to limit the rate at which receivers consume data.
+ *
+ * waitToPush method will block the thread if too many messages have been pushed too quickly,
+ * and only return when a new message has been pushed. It assumes that only one message is
+ * pushed at a time.
+ *
+ * The spark configuration spark.streaming.receiver.maxRate gives the maximum number of messages
+ * per second that each receiver will accept.
+ *
+ * @param conf spark configuration
+ */
 private[receiver] abstract class RateLimiter(conf: SparkConf) extends Logging {
 
   // treated as an upper limit
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobSet.scala b/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobSet.scala
index 66d5ffb797..0baedaf275 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobSet.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobSet.scala
@@ -21,9 +21,10 @@ import scala.collection.mutable.HashSet
 
 import org.apache.spark.streaming.Time
 
-/** Class representing a set of Jobs
-  * belong to the same batch.
-  */
+/**
+ * Class representing a set of Jobs
+ * belong to the same batch.
+ */
 private[streaming]
 case class JobSet(
     time: Time,
diff --git a/tools/src/main/scala/org/apache/spark/tools/GenerateMIMAIgnore.scala b/tools/src/main/scala/org/apache/spark/tools/GenerateMIMAIgnore.scala
index 0df3c501de..c9058ff409 100644
--- a/tools/src/main/scala/org/apache/spark/tools/GenerateMIMAIgnore.scala
+++ b/tools/src/main/scala/org/apache/spark/tools/GenerateMIMAIgnore.scala
@@ -91,10 +91,11 @@ object GenerateMIMAIgnore {
     (ignoredClasses.flatMap(c => Seq(c, c.replace("$", "#"))).toSet, ignoredMembers.toSet)
   }
 
-  /** Scala reflection does not let us see inner function even if they are upgraded
-    * to public for some reason. So had to resort to java reflection to get all inner
-    * functions with $$ in there name.
-    */
+  /**
+   * Scala reflection does not let us see inner function even if they are upgraded
+   * to public for some reason. So had to resort to java reflection to get all inner
+   * functions with $$ in there name.
+   */
   def getInnerFunctions(classSymbol: unv.ClassSymbol): Seq[String] = {
     try {
       Class.forName(classSymbol.fullName, false, classLoader).getMethods.map(_.getName)
diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtil.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtil.scala
index 5af2c29808..4b36da309d 100644
--- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtil.scala
+++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtil.scala
@@ -135,8 +135,8 @@ class YarnSparkHadoopUtil extends SparkHadoopUtil {
   }
 
   /**
-    * Obtains token for the Hive metastore and adds them to the credentials.
-    */
+   * Obtains token for the Hive metastore and adds them to the credentials.
+   */
   def obtainTokenForHiveMetastore(
       sparkConf: SparkConf,
       conf: Configuration,
@@ -149,8 +149,8 @@ class YarnSparkHadoopUtil extends SparkHadoopUtil {
   }
 
   /**
-    * Obtain a security token for HBase.
-    */
+   * Obtain a security token for HBase.
+   */
   def obtainTokenForHBase(
       sparkConf: SparkConf,
       conf: Configuration,
@@ -164,10 +164,10 @@ class YarnSparkHadoopUtil extends SparkHadoopUtil {
   }
 
   /**
-    * Return whether delegation tokens should be retrieved for the given service when security is
-    * enabled. By default, tokens are retrieved, but that behavior can be changed by setting
-    * a service-specific configuration.
-    */
+   * Return whether delegation tokens should be retrieved for the given service when security is
+   * enabled. By default, tokens are retrieved, but that behavior can be changed by setting
+   * a service-specific configuration.
+   */
   private def shouldGetTokens(conf: SparkConf, service: String): Boolean = {
     conf.getBoolean(s"spark.yarn.security.tokens.${service}.enabled", true)
   }
-- 
cgit v1.2.3


From 3f749f7ed443899d667c9e2b2a11bc595d6fc7f6 Mon Sep 17 00:00:00 2001
From: Dongjoon Hyun <dongjoon@apache.org>
Date: Sun, 3 Apr 2016 18:14:16 -0700
Subject: [SPARK-14355][BUILD] Fix typos in Exception/Testcase/Comments and
 static analysis results

## What changes were proposed in this pull request?

This PR contains the following 5 types of maintenance fix over 59 files (+94 lines, -93 lines).
- Fix typos(exception/log strings, testcase name, comments) in 44 lines.
- Fix lint-java errors (MaxLineLength) in 6 lines. (New codes after SPARK-14011)
- Use diamond operators in 40 lines. (New codes after SPARK-13702)
- Fix redundant semicolon in 5 lines.
- Rename class `InferSchemaSuite` to `CSVInferSchemaSuite` in CSVInferSchemaSuite.scala.

## How was this patch tested?

Manual and pass the Jenkins tests.

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #12139 from dongjoon-hyun/SPARK-14355.
---
 .../spark/network/client/TransportClientFactory.java       |  2 +-
 .../spark/network/client/TransportResponseHandler.java     |  6 +++---
 .../spark/network/server/OneForOneStreamManager.java       |  2 +-
 .../apache/spark/network/sasl/ShuffleSecretManager.java    |  2 +-
 .../collection/unsafe/sort/UnsafeSorterSpillMerger.java    |  2 +-
 core/src/main/scala/org/apache/spark/api/r/RRunner.scala   |  2 +-
 .../scala/org/apache/spark/util/random/RandomSampler.scala |  2 +-
 .../spark/shuffle/sort/UnsafeShuffleWriterSuite.java       |  4 ++--
 .../main/java/org/apache/spark/examples/JavaLogQuery.java  |  4 ++--
 .../mllib/JavaMultiLabelClassificationMetricsExample.java  | 14 +++++++-------
 .../mllib/JavaPowerIterationClusteringExample.java         | 10 +++++-----
 .../examples/mllib/JavaStratifiedSamplingExample.java      |  2 +-
 .../spark/examples/streaming/JavaFlumeEventCount.java      |  4 ++--
 .../apache/spark/streaming/flume/JavaFlumeStreamSuite.java | 11 ++++++-----
 .../org/apache/spark/launcher/CommandBuilderUtils.java     |  2 +-
 .../main/java/org/apache/spark/launcher/SparkLauncher.java |  2 +-
 .../org/apache/spark/launcher/LauncherServerSuite.java     |  2 +-
 .../spark/launcher/SparkSubmitCommandBuilderSuite.java     |  2 +-
 .../spark/ml/classification/DecisionTreeClassifier.scala   |  2 +-
 .../apache/spark/ml/param/shared/SharedParamsCodeGen.scala |  2 +-
 .../org/apache/spark/ml/param/shared/sharedParams.scala    |  4 ++--
 .../apache/spark/ml/regression/DecisionTreeRegressor.scala |  2 +-
 .../main/scala/org/apache/spark/ml/tree/treeModels.scala   |  2 +-
 .../spark/mllib/classification/LogisticRegression.scala    |  2 +-
 .../java/org/apache/spark/ml/param/JavaTestParams.java     |  2 +-
 .../JavaStreamingLogisticRegressionSuite.java              |  4 ++--
 .../spark/mllib/clustering/JavaStreamingKMeansSuite.java   |  4 ++--
 .../org/apache/spark/mllib/linalg/JavaVectorsSuite.java    |  4 ++--
 .../regression/JavaStreamingLinearRegressionSuite.java     |  4 ++--
 .../apache/spark/ml/regression/LinearRegressionSuite.scala |  2 +-
 .../spark/sql/execution/UnsafeExternalRowSorter.java       |  2 +-
 .../scala/org/apache/spark/sql/catalyst/CatalystConf.scala |  2 +-
 .../apache/spark/sql/catalyst/expressions/Expression.scala |  4 ++--
 .../sql/catalyst/expressions/codegen/CodeGenerator.scala   |  2 +-
 .../sql/catalyst/expressions/conditionalExpressions.scala  |  2 +-
 .../org/apache/spark/sql/catalyst/parser/AstBuilder.scala  |  4 ++--
 .../src/test/scala/org/apache/spark/sql/RowTest.scala      |  2 +-
 .../apache/spark/sql/execution/UnsafeKVExternalSorter.java |  2 +-
 .../spark/sql/execution/vectorized/ColumnarBatch.java      |  4 ++--
 .../spark/sql/execution/vectorized/OnHeapColumnVector.java |  2 +-
 .../main/scala/org/apache/spark/sql/ContinuousQuery.scala  |  2 +-
 sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala |  2 +-
 .../org/apache/spark/sql/execution/SparkStrategies.scala   |  2 +-
 .../main/scala/org/apache/spark/sql/execution/Window.scala |  2 +-
 .../org/apache/spark/sql/execution/basicOperators.scala    |  2 +-
 .../spark/sql/execution/columnar/ColumnBuilder.scala       |  2 +-
 .../apache/spark/sql/execution/stat/StatFunctions.scala    |  2 +-
 .../spark/sql/execution/streaming/FileStreamSink.scala     |  2 +-
 .../streaming/state/HDFSBackedStateStoreProvider.scala     |  2 +-
 .../spark/sql/execution/streaming/state/StateStore.scala   |  2 +-
 .../scala/org/apache/spark/sql/jdbc/JdbcDialects.scala     |  2 +-
 .../java/test/org/apache/spark/sql/JavaDatasetSuite.java   | 10 +++++-----
 .../src/test/scala/org/apache/spark/sql/QueryTest.scala    |  4 ++--
 .../execution/datasources/csv/CSVInferSchemaSuite.scala    |  2 +-
 .../sql/execution/datasources/parquet/ParquetIOSuite.scala |  2 +-
 .../org/apache/spark/sql/streaming/FileStressSuite.scala   |  2 +-
 .../scala/org/apache/spark/sql/hive/test/TestHive.scala    |  4 ++--
 .../spark/sql/hive/execution/HiveComparisonTest.scala      |  2 +-
 .../scala/org/apache/spark/sql/hive/parquetSuites.scala    |  4 ++--
 59 files changed, 94 insertions(+), 93 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/common/network-common/src/main/java/org/apache/spark/network/client/TransportClientFactory.java b/common/network-common/src/main/java/org/apache/spark/network/client/TransportClientFactory.java
index 5a36e18b09..b5a9d6671f 100644
--- a/common/network-common/src/main/java/org/apache/spark/network/client/TransportClientFactory.java
+++ b/common/network-common/src/main/java/org/apache/spark/network/client/TransportClientFactory.java
@@ -94,7 +94,7 @@ public class TransportClientFactory implements Closeable {
     this.context = Preconditions.checkNotNull(context);
     this.conf = context.getConf();
     this.clientBootstraps = Lists.newArrayList(Preconditions.checkNotNull(clientBootstraps));
-    this.connectionPool = new ConcurrentHashMap<SocketAddress, ClientPool>();
+    this.connectionPool = new ConcurrentHashMap<>();
     this.numConnectionsPerPeer = conf.numConnectionsPerPeer();
     this.rand = new Random();
 
diff --git a/common/network-common/src/main/java/org/apache/spark/network/client/TransportResponseHandler.java b/common/network-common/src/main/java/org/apache/spark/network/client/TransportResponseHandler.java
index f0e2004d2d..8a69223c88 100644
--- a/common/network-common/src/main/java/org/apache/spark/network/client/TransportResponseHandler.java
+++ b/common/network-common/src/main/java/org/apache/spark/network/client/TransportResponseHandler.java
@@ -64,9 +64,9 @@ public class TransportResponseHandler extends MessageHandler<ResponseMessage> {
 
   public TransportResponseHandler(Channel channel) {
     this.channel = channel;
-    this.outstandingFetches = new ConcurrentHashMap<StreamChunkId, ChunkReceivedCallback>();
-    this.outstandingRpcs = new ConcurrentHashMap<Long, RpcResponseCallback>();
-    this.streamCallbacks = new ConcurrentLinkedQueue<StreamCallback>();
+    this.outstandingFetches = new ConcurrentHashMap<>();
+    this.outstandingRpcs = new ConcurrentHashMap<>();
+    this.streamCallbacks = new ConcurrentLinkedQueue<>();
     this.timeOfLastRequestNs = new AtomicLong(0);
   }
 
diff --git a/common/network-common/src/main/java/org/apache/spark/network/server/OneForOneStreamManager.java b/common/network-common/src/main/java/org/apache/spark/network/server/OneForOneStreamManager.java
index e2222ae085..ae7e520b2f 100644
--- a/common/network-common/src/main/java/org/apache/spark/network/server/OneForOneStreamManager.java
+++ b/common/network-common/src/main/java/org/apache/spark/network/server/OneForOneStreamManager.java
@@ -63,7 +63,7 @@ public class OneForOneStreamManager extends StreamManager {
     // For debugging purposes, start with a random stream id to help identifying different streams.
     // This does not need to be globally unique, only unique to this class.
     nextStreamId = new AtomicLong((long) new Random().nextInt(Integer.MAX_VALUE) * 1000);
-    streams = new ConcurrentHashMap<Long, StreamState>();
+    streams = new ConcurrentHashMap<>();
   }
 
   @Override
diff --git a/common/network-shuffle/src/main/java/org/apache/spark/network/sasl/ShuffleSecretManager.java b/common/network-shuffle/src/main/java/org/apache/spark/network/sasl/ShuffleSecretManager.java
index 268cb40121..56a025c4d9 100644
--- a/common/network-shuffle/src/main/java/org/apache/spark/network/sasl/ShuffleSecretManager.java
+++ b/common/network-shuffle/src/main/java/org/apache/spark/network/sasl/ShuffleSecretManager.java
@@ -37,7 +37,7 @@ public class ShuffleSecretManager implements SecretKeyHolder {
   private static final String SPARK_SASL_USER = "sparkSaslUser";
 
   public ShuffleSecretManager() {
-    shuffleSecretMap = new ConcurrentHashMap<String, String>();
+    shuffleSecretMap = new ConcurrentHashMap<>();
   }
 
   /**
diff --git a/core/src/main/java/org/apache/spark/util/collection/unsafe/sort/UnsafeSorterSpillMerger.java b/core/src/main/java/org/apache/spark/util/collection/unsafe/sort/UnsafeSorterSpillMerger.java
index 2b1c860e55..01aed95878 100644
--- a/core/src/main/java/org/apache/spark/util/collection/unsafe/sort/UnsafeSorterSpillMerger.java
+++ b/core/src/main/java/org/apache/spark/util/collection/unsafe/sort/UnsafeSorterSpillMerger.java
@@ -45,7 +45,7 @@ final class UnsafeSorterSpillMerger {
         }
       }
     };
-    priorityQueue = new PriorityQueue<UnsafeSorterIterator>(numSpills, comparator);
+    priorityQueue = new PriorityQueue<>(numSpills, comparator);
   }
 
   /**
diff --git a/core/src/main/scala/org/apache/spark/api/r/RRunner.scala b/core/src/main/scala/org/apache/spark/api/r/RRunner.scala
index ff279ec270..07d1fa2c4a 100644
--- a/core/src/main/scala/org/apache/spark/api/r/RRunner.scala
+++ b/core/src/main/scala/org/apache/spark/api/r/RRunner.scala
@@ -182,7 +182,7 @@ private[spark] class RRunner[U](
           }
           stream.flush()
         } catch {
-          // TODO: We should propogate this error to the task thread
+          // TODO: We should propagate this error to the task thread
           case e: Exception =>
             logError("R Writer thread got an exception", e)
         } finally {
diff --git a/core/src/main/scala/org/apache/spark/util/random/RandomSampler.scala b/core/src/main/scala/org/apache/spark/util/random/RandomSampler.scala
index d397cca4b4..8c67364ef1 100644
--- a/core/src/main/scala/org/apache/spark/util/random/RandomSampler.scala
+++ b/core/src/main/scala/org/apache/spark/util/random/RandomSampler.scala
@@ -326,7 +326,7 @@ class GapSamplingReplacement(
   /**
    * Skip elements with replication factor zero (i.e. elements that won't be sampled).
    * Samples 'k' from geometric distribution  P(k) = (1-q)(q)^k, where q = e^(-f), that is
-   * q is the probabililty of Poisson(0; f)
+   * q is the probability of Poisson(0; f)
    */
   private def advance(): Unit = {
     val u = math.max(rng.nextDouble(), epsilon)
diff --git a/core/src/test/java/org/apache/spark/shuffle/sort/UnsafeShuffleWriterSuite.java b/core/src/test/java/org/apache/spark/shuffle/sort/UnsafeShuffleWriterSuite.java
index 44733dcdaf..30750b1bf1 100644
--- a/core/src/test/java/org/apache/spark/shuffle/sort/UnsafeShuffleWriterSuite.java
+++ b/core/src/test/java/org/apache/spark/shuffle/sort/UnsafeShuffleWriterSuite.java
@@ -170,11 +170,11 @@ public class UnsafeShuffleWriterSuite {
   private UnsafeShuffleWriter<Object, Object> createWriter(
       boolean transferToEnabled) throws IOException {
     conf.set("spark.file.transferTo", String.valueOf(transferToEnabled));
-    return new UnsafeShuffleWriter<Object, Object>(
+    return new UnsafeShuffleWriter<>(
       blockManager,
       shuffleBlockResolver,
       taskMemoryManager,
-      new SerializedShuffleHandle<Object, Object>(0, 1, shuffleDep),
+      new SerializedShuffleHandle<>(0, 1, shuffleDep),
       0, // map id
       taskContext,
       conf
diff --git a/examples/src/main/java/org/apache/spark/examples/JavaLogQuery.java b/examples/src/main/java/org/apache/spark/examples/JavaLogQuery.java
index 8abc03e73d..ebb0687b14 100644
--- a/examples/src/main/java/org/apache/spark/examples/JavaLogQuery.java
+++ b/examples/src/main/java/org/apache/spark/examples/JavaLogQuery.java
@@ -82,10 +82,10 @@ public final class JavaLogQuery {
       String user = m.group(3);
       String query = m.group(5);
       if (!user.equalsIgnoreCase("-")) {
-        return new Tuple3<String, String, String>(ip, user, query);
+        return new Tuple3<>(ip, user, query);
       }
     }
-    return new Tuple3<String, String, String>(null, null, null);
+    return new Tuple3<>(null, null, null);
   }
 
   public static Stats extractStats(String line) {
diff --git a/examples/src/main/java/org/apache/spark/examples/mllib/JavaMultiLabelClassificationMetricsExample.java b/examples/src/main/java/org/apache/spark/examples/mllib/JavaMultiLabelClassificationMetricsExample.java
index 5904260e2d..bc99dc023f 100644
--- a/examples/src/main/java/org/apache/spark/examples/mllib/JavaMultiLabelClassificationMetricsExample.java
+++ b/examples/src/main/java/org/apache/spark/examples/mllib/JavaMultiLabelClassificationMetricsExample.java
@@ -34,13 +34,13 @@ public class JavaMultiLabelClassificationMetricsExample {
     JavaSparkContext sc = new JavaSparkContext(conf);
     // $example on$
     List<Tuple2<double[], double[]>> data = Arrays.asList(
-      new Tuple2<double[], double[]>(new double[]{0.0, 1.0}, new double[]{0.0, 2.0}),
-      new Tuple2<double[], double[]>(new double[]{0.0, 2.0}, new double[]{0.0, 1.0}),
-      new Tuple2<double[], double[]>(new double[]{}, new double[]{0.0}),
-      new Tuple2<double[], double[]>(new double[]{2.0}, new double[]{2.0}),
-      new Tuple2<double[], double[]>(new double[]{2.0, 0.0}, new double[]{2.0, 0.0}),
-      new Tuple2<double[], double[]>(new double[]{0.0, 1.0, 2.0}, new double[]{0.0, 1.0}),
-      new Tuple2<double[], double[]>(new double[]{1.0}, new double[]{1.0, 2.0})
+      new Tuple2<>(new double[]{0.0, 1.0}, new double[]{0.0, 2.0}),
+      new Tuple2<>(new double[]{0.0, 2.0}, new double[]{0.0, 1.0}),
+      new Tuple2<>(new double[]{}, new double[]{0.0}),
+      new Tuple2<>(new double[]{2.0}, new double[]{2.0}),
+      new Tuple2<>(new double[]{2.0, 0.0}, new double[]{2.0, 0.0}),
+      new Tuple2<>(new double[]{0.0, 1.0, 2.0}, new double[]{0.0, 1.0}),
+      new Tuple2<>(new double[]{1.0}, new double[]{1.0, 2.0})
     );
     JavaRDD<Tuple2<double[], double[]>> scoreAndLabels = sc.parallelize(data);
 
diff --git a/examples/src/main/java/org/apache/spark/examples/mllib/JavaPowerIterationClusteringExample.java b/examples/src/main/java/org/apache/spark/examples/mllib/JavaPowerIterationClusteringExample.java
index b62fa90c34..91c3bd72da 100644
--- a/examples/src/main/java/org/apache/spark/examples/mllib/JavaPowerIterationClusteringExample.java
+++ b/examples/src/main/java/org/apache/spark/examples/mllib/JavaPowerIterationClusteringExample.java
@@ -40,11 +40,11 @@ public class JavaPowerIterationClusteringExample {
     @SuppressWarnings("unchecked")
     // $example on$
     JavaRDD<Tuple3<Long, Long, Double>> similarities = sc.parallelize(Lists.newArrayList(
-      new Tuple3<Long, Long, Double>(0L, 1L, 0.9),
-      new Tuple3<Long, Long, Double>(1L, 2L, 0.9),
-      new Tuple3<Long, Long, Double>(2L, 3L, 0.9),
-      new Tuple3<Long, Long, Double>(3L, 4L, 0.1),
-      new Tuple3<Long, Long, Double>(4L, 5L, 0.9)));
+      new Tuple3<>(0L, 1L, 0.9),
+      new Tuple3<>(1L, 2L, 0.9),
+      new Tuple3<>(2L, 3L, 0.9),
+      new Tuple3<>(3L, 4L, 0.1),
+      new Tuple3<>(4L, 5L, 0.9)));
 
     PowerIterationClustering pic = new PowerIterationClustering()
       .setK(2)
diff --git a/examples/src/main/java/org/apache/spark/examples/mllib/JavaStratifiedSamplingExample.java b/examples/src/main/java/org/apache/spark/examples/mllib/JavaStratifiedSamplingExample.java
index c27fba2783..86c389e11c 100644
--- a/examples/src/main/java/org/apache/spark/examples/mllib/JavaStratifiedSamplingExample.java
+++ b/examples/src/main/java/org/apache/spark/examples/mllib/JavaStratifiedSamplingExample.java
@@ -36,7 +36,7 @@ public class JavaStratifiedSamplingExample {
     JavaSparkContext jsc = new JavaSparkContext(conf);
 
     // $example on$
-    List<Tuple2<Integer, Character>> list = new ArrayList<Tuple2<Integer, Character>>(
+    List<Tuple2<Integer, Character>> list = new ArrayList<>(
       Arrays.<Tuple2<Integer, Character>>asList(
         new Tuple2(1, 'a'),
         new Tuple2(1, 'b'),
diff --git a/examples/src/main/java/org/apache/spark/examples/streaming/JavaFlumeEventCount.java b/examples/src/main/java/org/apache/spark/examples/streaming/JavaFlumeEventCount.java
index da56637fe8..bae4b78ac2 100644
--- a/examples/src/main/java/org/apache/spark/examples/streaming/JavaFlumeEventCount.java
+++ b/examples/src/main/java/org/apache/spark/examples/streaming/JavaFlumeEventCount.java
@@ -19,7 +19,6 @@ package org.apache.spark.examples.streaming;
 
 import org.apache.spark.SparkConf;
 import org.apache.spark.api.java.function.Function;
-import org.apache.spark.examples.streaming.StreamingExamples;
 import org.apache.spark.streaming.*;
 import org.apache.spark.streaming.api.java.*;
 import org.apache.spark.streaming.flume.FlumeUtils;
@@ -58,7 +57,8 @@ public final class JavaFlumeEventCount {
     Duration batchInterval = new Duration(2000);
     SparkConf sparkConf = new SparkConf().setAppName("JavaFlumeEventCount");
     JavaStreamingContext ssc = new JavaStreamingContext(sparkConf, batchInterval);
-    JavaReceiverInputDStream<SparkFlumeEvent> flumeStream = FlumeUtils.createStream(ssc, host, port);
+    JavaReceiverInputDStream<SparkFlumeEvent> flumeStream =
+      FlumeUtils.createStream(ssc, host, port);
 
     flumeStream.count();
 
diff --git a/external/flume/src/test/java/org/apache/spark/streaming/flume/JavaFlumeStreamSuite.java b/external/flume/src/test/java/org/apache/spark/streaming/flume/JavaFlumeStreamSuite.java
index 3b5e0c7746..ada05f203b 100644
--- a/external/flume/src/test/java/org/apache/spark/streaming/flume/JavaFlumeStreamSuite.java
+++ b/external/flume/src/test/java/org/apache/spark/streaming/flume/JavaFlumeStreamSuite.java
@@ -27,10 +27,11 @@ public class JavaFlumeStreamSuite extends LocalJavaStreamingContext {
   @Test
   public void testFlumeStream() {
     // tests the API, does not actually test data receiving
-    JavaReceiverInputDStream<SparkFlumeEvent> test1 = FlumeUtils.createStream(ssc, "localhost", 12345);
-    JavaReceiverInputDStream<SparkFlumeEvent> test2 = FlumeUtils.createStream(ssc, "localhost", 12345,
-      StorageLevel.MEMORY_AND_DISK_SER_2());
-    JavaReceiverInputDStream<SparkFlumeEvent> test3 = FlumeUtils.createStream(ssc, "localhost", 12345,
-      StorageLevel.MEMORY_AND_DISK_SER_2(), false);
+    JavaReceiverInputDStream<SparkFlumeEvent> test1 = FlumeUtils.createStream(ssc, "localhost",
+      12345);
+    JavaReceiverInputDStream<SparkFlumeEvent> test2 = FlumeUtils.createStream(ssc, "localhost",
+      12345, StorageLevel.MEMORY_AND_DISK_SER_2());
+    JavaReceiverInputDStream<SparkFlumeEvent> test3 = FlumeUtils.createStream(ssc, "localhost",
+      12345, StorageLevel.MEMORY_AND_DISK_SER_2(), false);
   }
 }
diff --git a/launcher/src/main/java/org/apache/spark/launcher/CommandBuilderUtils.java b/launcher/src/main/java/org/apache/spark/launcher/CommandBuilderUtils.java
index 1e55aad5c9..a08c8dcba4 100644
--- a/launcher/src/main/java/org/apache/spark/launcher/CommandBuilderUtils.java
+++ b/launcher/src/main/java/org/apache/spark/launcher/CommandBuilderUtils.java
@@ -34,7 +34,7 @@ class CommandBuilderUtils {
   /** The set of known JVM vendors. */
   enum JavaVendor {
     Oracle, IBM, OpenJDK, Unknown
-  };
+  }
 
   /** Returns whether the given string is null or empty. */
   static boolean isEmpty(String s) {
diff --git a/launcher/src/main/java/org/apache/spark/launcher/SparkLauncher.java b/launcher/src/main/java/org/apache/spark/launcher/SparkLauncher.java
index a542159901..a083f05a2a 100644
--- a/launcher/src/main/java/org/apache/spark/launcher/SparkLauncher.java
+++ b/launcher/src/main/java/org/apache/spark/launcher/SparkLauncher.java
@@ -477,6 +477,6 @@ public class SparkLauncher {
       // No op.
     }
 
-  };
+  }
 
 }
diff --git a/launcher/src/test/java/org/apache/spark/launcher/LauncherServerSuite.java b/launcher/src/test/java/org/apache/spark/launcher/LauncherServerSuite.java
index 5bf2babdd1..a9039b3ec9 100644
--- a/launcher/src/test/java/org/apache/spark/launcher/LauncherServerSuite.java
+++ b/launcher/src/test/java/org/apache/spark/launcher/LauncherServerSuite.java
@@ -175,7 +175,7 @@ public class LauncherServerSuite extends BaseSuite {
 
     TestClient(Socket s) throws IOException {
       super(s);
-      this.inbound = new LinkedBlockingQueue<Message>();
+      this.inbound = new LinkedBlockingQueue<>();
       this.clientThread = new Thread(this);
       clientThread.setName("TestClient");
       clientThread.setDaemon(true);
diff --git a/launcher/src/test/java/org/apache/spark/launcher/SparkSubmitCommandBuilderSuite.java b/launcher/src/test/java/org/apache/spark/launcher/SparkSubmitCommandBuilderSuite.java
index b7f4f2efc5..29cbbe825b 100644
--- a/launcher/src/test/java/org/apache/spark/launcher/SparkSubmitCommandBuilderSuite.java
+++ b/launcher/src/test/java/org/apache/spark/launcher/SparkSubmitCommandBuilderSuite.java
@@ -160,7 +160,7 @@ public class SparkSubmitCommandBuilderSuite extends BaseSuite {
       "SparkPi",
       "42");
 
-    Map<String, String> env = new HashMap<String, String>();
+    Map<String, String> env = new HashMap<>();
     List<String> cmd = buildCommand(sparkSubmitArgs, env);
     assertEquals("foo", findArgValue(cmd, parser.MASTER));
     assertEquals("bar", findArgValue(cmd, parser.DEPLOY_MODE));
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
index 23c4af17f9..4525bf71f6 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
@@ -205,7 +205,7 @@ final class DecisionTreeClassificationModel private[ml] (
   @Since("2.0.0")
   lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(this, numFeatures)
 
-  /** Convert to spark.mllib DecisionTreeModel (losing some infomation) */
+  /** Convert to spark.mllib DecisionTreeModel (losing some information) */
   override private[spark] def toOld: OldDecisionTreeModel = {
     new OldDecisionTreeModel(rootNode.toOld(1), OldAlgo.Classification)
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/param/shared/SharedParamsCodeGen.scala b/mllib/src/main/scala/org/apache/spark/ml/param/shared/SharedParamsCodeGen.scala
index 3ce129b12c..1d03a5b4f4 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/param/shared/SharedParamsCodeGen.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/param/shared/SharedParamsCodeGen.scala
@@ -62,7 +62,7 @@ private[shared] object SharedParamsCodeGen {
         "every 10 iterations", isValid = "(interval: Int) => interval == -1 || interval >= 1"),
       ParamDesc[Boolean]("fitIntercept", "whether to fit an intercept term", Some("true")),
       ParamDesc[String]("handleInvalid", "how to handle invalid entries. Options are skip (which " +
-        "will filter out rows with bad values), or error (which will throw an errror). More " +
+        "will filter out rows with bad values), or error (which will throw an error). More " +
         "options may be added later",
         isValid = "ParamValidators.inArray(Array(\"skip\", \"error\"))"),
       ParamDesc[Boolean]("standardization", "whether to standardize the training features" +
diff --git a/mllib/src/main/scala/org/apache/spark/ml/param/shared/sharedParams.scala b/mllib/src/main/scala/org/apache/spark/ml/param/shared/sharedParams.scala
index 96263c5baf..64d6af2766 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/param/shared/sharedParams.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/param/shared/sharedParams.scala
@@ -270,10 +270,10 @@ private[ml] trait HasFitIntercept extends Params {
 private[ml] trait HasHandleInvalid extends Params {
 
   /**
-   * Param for how to handle invalid entries. Options are skip (which will filter out rows with bad values), or error (which will throw an errror). More options may be added later.
+   * Param for how to handle invalid entries. Options are skip (which will filter out rows with bad values), or error (which will throw an error). More options may be added later.
    * @group param
    */
-  final val handleInvalid: Param[String] = new Param[String](this, "handleInvalid", "how to handle invalid entries. Options are skip (which will filter out rows with bad values), or error (which will throw an errror). More options may be added later", ParamValidators.inArray(Array("skip", "error")))
+  final val handleInvalid: Param[String] = new Param[String](this, "handleInvalid", "how to handle invalid entries. Options are skip (which will filter out rows with bad values), or error (which will throw an error). More options may be added later", ParamValidators.inArray(Array("skip", "error")))
 
   /** @group getParam */
   final def getHandleInvalid: String = $(handleInvalid)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
index 0a3d00e470..1289a317ee 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
@@ -205,7 +205,7 @@ final class DecisionTreeRegressionModel private[ml] (
   @Since("2.0.0")
   lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(this, numFeatures)
 
-  /** Convert to spark.mllib DecisionTreeModel (losing some infomation) */
+  /** Convert to spark.mllib DecisionTreeModel (losing some information) */
   override private[spark] def toOld: OldDecisionTreeModel = {
     new OldDecisionTreeModel(rootNode.toOld(1), OldAlgo.Regression)
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
index 1fad9d6d8c..8ea767b2b3 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
@@ -71,7 +71,7 @@ private[spark] trait DecisionTreeModel {
    */
   private[ml] def maxSplitFeatureIndex(): Int = rootNode.maxSplitFeatureIndex()
 
-  /** Convert to spark.mllib DecisionTreeModel (losing some infomation) */
+  /** Convert to spark.mllib DecisionTreeModel (losing some information) */
   private[spark] def toOld: OldDecisionTreeModel
 }
 
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala
index c0404be019..f10570e662 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala
@@ -418,7 +418,7 @@ class LogisticRegressionWithLBFGS
 
   private def run(input: RDD[LabeledPoint], initialWeights: Vector, userSuppliedWeights: Boolean):
       LogisticRegressionModel = {
-    // ml's Logisitic regression only supports binary classifcation currently.
+    // ml's Logistic regression only supports binary classification currently.
     if (numOfLinearPredictor == 1) {
       def runWithMlLogisitcRegression(elasticNetParam: Double) = {
         // Prepare the ml LogisticRegression based on our settings
diff --git a/mllib/src/test/java/org/apache/spark/ml/param/JavaTestParams.java b/mllib/src/test/java/org/apache/spark/ml/param/JavaTestParams.java
index 65841182df..06f7fbb86e 100644
--- a/mllib/src/test/java/org/apache/spark/ml/param/JavaTestParams.java
+++ b/mllib/src/test/java/org/apache/spark/ml/param/JavaTestParams.java
@@ -89,7 +89,7 @@ public class JavaTestParams extends JavaParams {
     myDoubleParam_ = new DoubleParam(this, "myDoubleParam", "this is a double param",
       ParamValidators.inRange(0.0, 1.0));
     List<String> validStrings = Arrays.asList("a", "b");
-    myStringParam_ = new Param<String>(this, "myStringParam", "this is a string param",
+    myStringParam_ = new Param<>(this, "myStringParam", "this is a string param",
       ParamValidators.inArray(validStrings));
     myDoubleArrayParam_ =
       new DoubleArrayParam(this, "myDoubleArrayParam", "this is a double param");
diff --git a/mllib/src/test/java/org/apache/spark/mllib/classification/JavaStreamingLogisticRegressionSuite.java b/mllib/src/test/java/org/apache/spark/mllib/classification/JavaStreamingLogisticRegressionSuite.java
index c9e5ee22f3..62c6d9b7e3 100644
--- a/mllib/src/test/java/org/apache/spark/mllib/classification/JavaStreamingLogisticRegressionSuite.java
+++ b/mllib/src/test/java/org/apache/spark/mllib/classification/JavaStreamingLogisticRegressionSuite.java
@@ -66,8 +66,8 @@ public class JavaStreamingLogisticRegressionSuite implements Serializable {
     JavaDStream<LabeledPoint> training =
       attachTestInputStream(ssc, Arrays.asList(trainingBatch, trainingBatch), 2);
     List<Tuple2<Integer, Vector>> testBatch = Arrays.asList(
-      new Tuple2<Integer, Vector>(10, Vectors.dense(1.0)),
-      new Tuple2<Integer, Vector>(11, Vectors.dense(0.0)));
+      new Tuple2<>(10, Vectors.dense(1.0)),
+      new Tuple2<>(11, Vectors.dense(0.0)));
     JavaPairDStream<Integer, Vector> test = JavaPairDStream.fromJavaDStream(
       attachTestInputStream(ssc, Arrays.asList(testBatch, testBatch), 2));
     StreamingLogisticRegressionWithSGD slr = new StreamingLogisticRegressionWithSGD()
diff --git a/mllib/src/test/java/org/apache/spark/mllib/clustering/JavaStreamingKMeansSuite.java b/mllib/src/test/java/org/apache/spark/mllib/clustering/JavaStreamingKMeansSuite.java
index d644766d1e..62edbd3a29 100644
--- a/mllib/src/test/java/org/apache/spark/mllib/clustering/JavaStreamingKMeansSuite.java
+++ b/mllib/src/test/java/org/apache/spark/mllib/clustering/JavaStreamingKMeansSuite.java
@@ -66,8 +66,8 @@ public class JavaStreamingKMeansSuite implements Serializable {
     JavaDStream<Vector> training =
       attachTestInputStream(ssc, Arrays.asList(trainingBatch, trainingBatch), 2);
     List<Tuple2<Integer, Vector>> testBatch = Arrays.asList(
-      new Tuple2<Integer, Vector>(10, Vectors.dense(1.0)),
-      new Tuple2<Integer, Vector>(11, Vectors.dense(0.0)));
+      new Tuple2<>(10, Vectors.dense(1.0)),
+      new Tuple2<>(11, Vectors.dense(0.0)));
     JavaPairDStream<Integer, Vector> test = JavaPairDStream.fromJavaDStream(
       attachTestInputStream(ssc, Arrays.asList(testBatch, testBatch), 2));
     StreamingKMeans skmeans = new StreamingKMeans()
diff --git a/mllib/src/test/java/org/apache/spark/mllib/linalg/JavaVectorsSuite.java b/mllib/src/test/java/org/apache/spark/mllib/linalg/JavaVectorsSuite.java
index 77c8c6274f..4ba8e543a9 100644
--- a/mllib/src/test/java/org/apache/spark/mllib/linalg/JavaVectorsSuite.java
+++ b/mllib/src/test/java/org/apache/spark/mllib/linalg/JavaVectorsSuite.java
@@ -37,8 +37,8 @@ public class JavaVectorsSuite implements Serializable {
   public void sparseArrayConstruction() {
     @SuppressWarnings("unchecked")
     Vector v = Vectors.sparse(3, Arrays.asList(
-        new Tuple2<Integer, Double>(0, 2.0),
-        new Tuple2<Integer, Double>(2, 3.0)));
+        new Tuple2<>(0, 2.0),
+        new Tuple2<>(2, 3.0)));
     assertArrayEquals(new double[]{2.0, 0.0, 3.0}, v.toArray(), 0.0);
   }
 }
diff --git a/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java b/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java
index dbf6488d41..ea0ccd7448 100644
--- a/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java
+++ b/mllib/src/test/java/org/apache/spark/mllib/regression/JavaStreamingLinearRegressionSuite.java
@@ -65,8 +65,8 @@ public class JavaStreamingLinearRegressionSuite implements Serializable {
     JavaDStream<LabeledPoint> training =
       attachTestInputStream(ssc, Arrays.asList(trainingBatch, trainingBatch), 2);
     List<Tuple2<Integer, Vector>> testBatch = Arrays.asList(
-      new Tuple2<Integer, Vector>(10, Vectors.dense(1.0)),
-      new Tuple2<Integer, Vector>(11, Vectors.dense(0.0)));
+      new Tuple2<>(10, Vectors.dense(1.0)),
+      new Tuple2<>(11, Vectors.dense(0.0)));
     JavaPairDStream<Integer, Vector> test = JavaPairDStream.fromJavaDStream(
       attachTestInputStream(ssc, Arrays.asList(testBatch, testBatch), 2));
     StreamingLinearRegressionWithSGD slr = new StreamingLinearRegressionWithSGD()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala
index cccb7f8d1b..eb19d13093 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/LinearRegressionSuite.scala
@@ -759,7 +759,7 @@ class LinearRegressionSuite
             .sliding(2)
             .forall(x => x(0) >= x(1)))
       } else {
-        // To clalify that the normal solver is used here.
+        // To clarify that the normal solver is used here.
         assert(model.summary.objectiveHistory.length == 1)
         assert(model.summary.objectiveHistory(0) == 0.0)
         val devianceResidualsR = Array(-0.47082, 0.34635)
diff --git a/sql/catalyst/src/main/java/org/apache/spark/sql/execution/UnsafeExternalRowSorter.java b/sql/catalyst/src/main/java/org/apache/spark/sql/execution/UnsafeExternalRowSorter.java
index aa7fc2121e..7784345a7a 100644
--- a/sql/catalyst/src/main/java/org/apache/spark/sql/execution/UnsafeExternalRowSorter.java
+++ b/sql/catalyst/src/main/java/org/apache/spark/sql/execution/UnsafeExternalRowSorter.java
@@ -151,7 +151,7 @@ public final class UnsafeExternalRowSorter {
             Platform.throwException(e);
           }
           throw new RuntimeException("Exception should have been re-thrown in next()");
-        };
+        }
       };
     } catch (IOException e) {
       cleanupResources();
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/CatalystConf.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/CatalystConf.scala
index d5ac01500b..2b98aacdd7 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/CatalystConf.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/CatalystConf.scala
@@ -26,7 +26,7 @@ private[spark] trait CatalystConf {
   def groupByOrdinal: Boolean
 
   /**
-   * Returns the [[Resolver]] for the current configuration, which can be used to determin if two
+   * Returns the [[Resolver]] for the current configuration, which can be used to determine if two
    * identifiers are equal.
    */
   def resolver: Resolver = {
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
index 5f8899d599..a24a5db8d4 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
@@ -153,8 +153,8 @@ abstract class Expression extends TreeNode[Expression] {
    * evaluate to the same result.
    */
   lazy val canonicalized: Expression = {
-    val canonicalizedChildred = children.map(_.canonicalized)
-    Canonicalize.execute(withNewChildren(canonicalizedChildred))
+    val canonicalizedChildren = children.map(_.canonicalized)
+    Canonicalize.execute(withNewChildren(canonicalizedChildren))
   }
 
   /**
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
index b64d3eea49..1bebd4e904 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
@@ -509,7 +509,7 @@ class CodegenContext {
 
   /**
    * Checks and sets up the state and codegen for subexpression elimination. This finds the
-   * common subexpresses, generates the functions that evaluate those expressions and populates
+   * common subexpressions, generates the functions that evaluate those expressions and populates
    * the mapping of common subexpressions to the generated functions.
    */
   private def subexpressionElimination(expressions: Seq[Expression]) = {
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/conditionalExpressions.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/conditionalExpressions.scala
index 103ab365e3..35a7b46020 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/conditionalExpressions.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/conditionalExpressions.scala
@@ -222,7 +222,7 @@ object CaseWhen {
   }
 
   /**
-   * A factory method to faciliate the creation of this expression when used in parsers.
+   * A factory method to facilitate the creation of this expression when used in parsers.
    * @param branches Expressions at even position are the branch conditions, and expressions at odd
    *                 position are branch values.
    */
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
index 8541b1f7c6..61ea3e4010 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/parser/AstBuilder.scala
@@ -965,7 +965,7 @@ class AstBuilder extends SqlBaseBaseVisitor[AnyRef] with Logging {
 
   /**
    * Create a binary arithmetic expression. The following arithmetic operators are supported:
-   * - Mulitplication: '*'
+   * - Multiplication: '*'
    * - Division: '/'
    * - Hive Long Division: 'DIV'
    * - Modulo: '%'
@@ -1270,7 +1270,7 @@ class AstBuilder extends SqlBaseBaseVisitor[AnyRef] with Logging {
   }
 
   /**
-   * Create a double literal for a number denoted in scientifc notation.
+   * Create a double literal for a number denoted in scientific notation.
    */
   override def visitScientificDecimalLiteral(
       ctx: ScientificDecimalLiteralContext): Literal = withOrigin(ctx) {
diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/RowTest.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/RowTest.scala
index d9577dea1b..c9c9599e7f 100644
--- a/sql/catalyst/src/test/scala/org/apache/spark/sql/RowTest.scala
+++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/RowTest.scala
@@ -121,7 +121,7 @@ class RowTest extends FunSpec with Matchers {
       externalRow should be theSameInstanceAs externalRow.copy()
     }
 
-    it("copy should return same ref for interal rows") {
+    it("copy should return same ref for internal rows") {
       internalRow should be theSameInstanceAs internalRow.copy()
     }
 
diff --git a/sql/core/src/main/java/org/apache/spark/sql/execution/UnsafeKVExternalSorter.java b/sql/core/src/main/java/org/apache/spark/sql/execution/UnsafeKVExternalSorter.java
index d3bfb00b3f..8132bba04c 100644
--- a/sql/core/src/main/java/org/apache/spark/sql/execution/UnsafeKVExternalSorter.java
+++ b/sql/core/src/main/java/org/apache/spark/sql/execution/UnsafeKVExternalSorter.java
@@ -272,5 +272,5 @@ public final class UnsafeKVExternalSorter {
     public void close() {
       cleanupResources();
     }
-  };
+  }
 }
diff --git a/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/ColumnarBatch.java b/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/ColumnarBatch.java
index 792e17911f..d1cc4e6d03 100644
--- a/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/ColumnarBatch.java
+++ b/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/ColumnarBatch.java
@@ -79,7 +79,7 @@ public final class ColumnarBatch {
 
   /**
    * Called to close all the columns in this batch. It is not valid to access the data after
-   * calling this. This must be called at the end to clean up memory allcoations.
+   * calling this. This must be called at the end to clean up memory allocations.
    */
   public void close() {
     for (ColumnVector c: columns) {
@@ -315,7 +315,7 @@ public final class ColumnarBatch {
   public int numRows() { return numRows; }
 
   /**
-   * Returns the number of valid rowss.
+   * Returns the number of valid rows.
    */
   public int numValidRows() {
     assert(numRowsFiltered <= numRows);
diff --git a/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/OnHeapColumnVector.java b/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/OnHeapColumnVector.java
index b1429fe7cb..708a00953a 100644
--- a/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/OnHeapColumnVector.java
+++ b/sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/OnHeapColumnVector.java
@@ -212,7 +212,7 @@ public final class OnHeapColumnVector extends ColumnVector {
   public void putIntsLittleEndian(int rowId, int count, byte[] src, int srcIndex) {
     int srcOffset = srcIndex + Platform.BYTE_ARRAY_OFFSET;
     for (int i = 0; i < count; ++i) {
-      intData[i + rowId] = Platform.getInt(src, srcOffset);;
+      intData[i + rowId] = Platform.getInt(src, srcOffset);
       srcIndex += 4;
       srcOffset += 4;
     }
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/ContinuousQuery.scala b/sql/core/src/main/scala/org/apache/spark/sql/ContinuousQuery.scala
index 1dc9a6893e..d9973b092d 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/ContinuousQuery.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/ContinuousQuery.scala
@@ -94,7 +94,7 @@ trait ContinuousQuery {
   /**
    * Blocks until all available data in the source has been processed an committed to the sink.
    * This method is intended for testing. Note that in the case of continually arriving data, this
-   * method may block forever.  Additionally, this method is only guranteed to block until data that
+   * method may block forever. Additionally, this method is only guaranteed to block until data that
    * has been synchronously appended data to a [[org.apache.spark.sql.execution.streaming.Source]]
    * prior to invocation. (i.e. `getOffset` must immediately reflect the addition).
    */
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala b/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
index 41cb799b97..a39a2113e5 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
@@ -2077,7 +2077,7 @@ class Dataset[T] private[sql](
 
   /**
    * Returns a new [[Dataset]] partitioned by the given partitioning expressions into
-   * `numPartitions`. The resulting Datasetis hash partitioned.
+   * `numPartitions`. The resulting Dataset is hash partitioned.
    *
    * This is the same operation as "DISTRIBUTE BY" in SQL (Hive QL).
    *
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala
index 5bcc172ca7..e1fabf519a 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala
@@ -108,7 +108,7 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
     /**
      * Matches a plan whose single partition should be small enough to build a hash table.
      *
-     * Note: this assume that the number of partition is fixed, requires addtional work if it's
+     * Note: this assume that the number of partition is fixed, requires additional work if it's
      * dynamic.
      */
     def canBuildHashMap(plan: LogicalPlan): Boolean = {
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala
index 806089196c..8e9214fa25 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/Window.scala
@@ -811,7 +811,7 @@ private[execution] final class UnboundedPrecedingWindowFunctionFrame(
  *
  * This is a very expensive operator to use, O(n * (n - 1) /2), because we need to maintain a
  * buffer and must do full recalculation after each row. Reverse iteration would be possible, if
- * the communitativity of the used window functions can be guaranteed.
+ * the commutativity of the used window functions can be guaranteed.
  *
  * @param target to write results to.
  * @param processor to calculate the row values with.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala
index fb1c6182cf..aba500ad8d 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/basicOperators.scala
@@ -146,7 +146,7 @@ case class Filter(condition: Expression, child: SparkPlan)
     // This has the property of not doing redundant IsNotNull checks and taking better advantage of
     // short-circuiting, not loading attributes until they are needed.
     // This is very perf sensitive.
-    // TODO: revisit this. We can consider reodering predicates as well.
+    // TODO: revisit this. We can consider reordering predicates as well.
     val generatedIsNotNullChecks = new Array[Boolean](notNullPreds.length)
     val generated = otherPreds.map { c =>
       val nullChecks = c.references.map { r =>
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/ColumnBuilder.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/ColumnBuilder.scala
index 7e26f19bb7..9a173367f4 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/ColumnBuilder.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/ColumnBuilder.scala
@@ -185,7 +185,7 @@ private[columnar] object ColumnBuilder {
       case udt: UserDefinedType[_] =>
         return apply(udt.sqlType, initialSize, columnName, useCompression)
       case other =>
-        throw new Exception(s"not suppported type: $other")
+        throw new Exception(s"not supported type: $other")
     }
 
     builder.initialize(initialSize, columnName, useCompression)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/stat/StatFunctions.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/stat/StatFunctions.scala
index e0b6709c51..d603f63a08 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/stat/StatFunctions.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/stat/StatFunctions.scala
@@ -296,7 +296,7 @@ private[sql] object StatFunctions extends Logging {
     val defaultRelativeError: Double = 0.01
 
     /**
-     * Statisttics from the Greenwald-Khanna paper.
+     * Statistics from the Greenwald-Khanna paper.
      * @param value the sampled value
      * @param g the minimum rank jump from the previous value's minimum rank
      * @param delta the maximum span of the rank.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSink.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSink.scala
index e819e95d61..6921ae584d 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSink.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/FileStreamSink.scala
@@ -32,7 +32,7 @@ object FileStreamSink {
 
 /**
  * A sink that writes out results to parquet files.  Each batch is written out to a unique
- * directory. After all of the files in a batch have been succesfully written, the list of
+ * directory. After all of the files in a batch have been successfully written, the list of
  * file paths is appended to the log atomically. In the case of partial failures, some duplicate
  * data may be present in the target directory, but only one copy of each file will be present
  * in the log.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala
index 8ece3c971a..1e0a4a5d4f 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala
@@ -178,7 +178,7 @@ private[state] class HDFSBackedStateStoreProvider(
      * This can be called only after committing all the updates made in the current thread.
      */
     override def iterator(): Iterator[(UnsafeRow, UnsafeRow)] = {
-      verify(state == COMMITTED, "Cannot get iterator of store data before comitting")
+      verify(state == COMMITTED, "Cannot get iterator of store data before committing")
       HDFSBackedStateStoreProvider.this.iterator(newVersion)
     }
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/StateStore.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/StateStore.scala
index d60e6185ac..07f63f928b 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/StateStore.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/StateStore.scala
@@ -220,7 +220,7 @@ private[state] object StateStore extends Logging {
       val executorId = SparkEnv.get.blockManager.blockManagerId.executorId
       val verified =
         coordinatorRef.map(_.verifyIfInstanceActive(storeId, executorId)).getOrElse(false)
-      logDebug(s"Verifyied whether the loaded instance $storeId is active: $verified" )
+      logDebug(s"Verified whether the loaded instance $storeId is active: $verified" )
       verified
     } catch {
       case NonFatal(e) =>
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala
index ca2d909e2c..cfe4911cb7 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala
@@ -126,7 +126,7 @@ object JdbcDialects {
 
   /**
    * Register a dialect for use on all new matching jdbc [[org.apache.spark.sql.DataFrame]].
-   * Readding an existing dialect will cause a move-to-front.
+   * Reading an existing dialect will cause a move-to-front.
    *
    * @param dialect The new dialect.
    */
diff --git a/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java b/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java
index a5ab446e08..873f681bdf 100644
--- a/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java
+++ b/sql/core/src/test/java/test/org/apache/spark/sql/JavaDatasetSuite.java
@@ -318,14 +318,14 @@ public class JavaDatasetSuite implements Serializable {
     Encoder<Tuple3<Integer, Long, String>> encoder3 =
       Encoders.tuple(Encoders.INT(), Encoders.LONG(), Encoders.STRING());
     List<Tuple3<Integer, Long, String>> data3 =
-      Arrays.asList(new Tuple3<Integer, Long, String>(1, 2L, "a"));
+      Arrays.asList(new Tuple3<>(1, 2L, "a"));
     Dataset<Tuple3<Integer, Long, String>> ds3 = context.createDataset(data3, encoder3);
     Assert.assertEquals(data3, ds3.collectAsList());
 
     Encoder<Tuple4<Integer, String, Long, String>> encoder4 =
       Encoders.tuple(Encoders.INT(), Encoders.STRING(), Encoders.LONG(), Encoders.STRING());
     List<Tuple4<Integer, String, Long, String>> data4 =
-      Arrays.asList(new Tuple4<Integer, String, Long, String>(1, "b", 2L, "a"));
+      Arrays.asList(new Tuple4<>(1, "b", 2L, "a"));
     Dataset<Tuple4<Integer, String, Long, String>> ds4 = context.createDataset(data4, encoder4);
     Assert.assertEquals(data4, ds4.collectAsList());
 
@@ -333,7 +333,7 @@ public class JavaDatasetSuite implements Serializable {
       Encoders.tuple(Encoders.INT(), Encoders.STRING(), Encoders.LONG(), Encoders.STRING(),
         Encoders.BOOLEAN());
     List<Tuple5<Integer, String, Long, String, Boolean>> data5 =
-      Arrays.asList(new Tuple5<Integer, String, Long, String, Boolean>(1, "b", 2L, "a", true));
+      Arrays.asList(new Tuple5<>(1, "b", 2L, "a", true));
     Dataset<Tuple5<Integer, String, Long, String, Boolean>> ds5 =
       context.createDataset(data5, encoder5);
     Assert.assertEquals(data5, ds5.collectAsList());
@@ -354,7 +354,7 @@ public class JavaDatasetSuite implements Serializable {
       Encoders.tuple(Encoders.INT(),
         Encoders.tuple(Encoders.STRING(), Encoders.STRING(), Encoders.LONG()));
     List<Tuple2<Integer, Tuple3<String, String, Long>>> data2 =
-      Arrays.asList(tuple2(1, new Tuple3<String, String, Long>("a", "b", 3L)));
+      Arrays.asList(tuple2(1, new Tuple3<>("a", "b", 3L)));
     Dataset<Tuple2<Integer, Tuple3<String, String, Long>>> ds2 =
       context.createDataset(data2, encoder2);
     Assert.assertEquals(data2, ds2.collectAsList());
@@ -376,7 +376,7 @@ public class JavaDatasetSuite implements Serializable {
       Encoders.tuple(Encoders.DOUBLE(), Encoders.DECIMAL(), Encoders.DATE(), Encoders.TIMESTAMP(),
         Encoders.FLOAT());
     List<Tuple5<Double, BigDecimal, Date, Timestamp, Float>> data =
-      Arrays.asList(new Tuple5<Double, BigDecimal, Date, Timestamp, Float>(
+      Arrays.asList(new Tuple5<>(
         1.7976931348623157E308, new BigDecimal("0.922337203685477589"),
           Date.valueOf("1970-01-01"), new Timestamp(System.currentTimeMillis()), Float.MAX_VALUE));
     Dataset<Tuple5<Double, BigDecimal, Date, Timestamp, Float>> ds =
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala b/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala
index d160f8ab8c..f7f3bd78e9 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/QueryTest.scala
@@ -105,10 +105,10 @@ abstract class QueryTest extends PlanTest {
       val expected = expectedAnswer.toSet.toSeq.map((a: Any) => a.toString).sorted
       val actual = decoded.toSet.toSeq.map((a: Any) => a.toString).sorted
 
-      val comparision = sideBySide("expected" +: expected, "spark" +: actual).mkString("\n")
+      val comparison = sideBySide("expected" +: expected, "spark" +: actual).mkString("\n")
       fail(
         s"""Decoded objects do not match expected objects:
-            |$comparision
+            |$comparison
             |${ds.resolvedTEncoder.deserializer.treeString}
          """.stripMargin)
     }
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchemaSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchemaSuite.scala
index 3a7cb25b4f..23d422635b 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchemaSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/csv/CSVInferSchemaSuite.scala
@@ -20,7 +20,7 @@ package org.apache.spark.sql.execution.datasources.csv
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.sql.types._
 
-class InferSchemaSuite extends SparkFunSuite {
+class CSVInferSchemaSuite extends SparkFunSuite {
 
   test("String fields types are inferred correctly from null types") {
     assert(CSVInferSchema.inferField(NullType, "") == NullType)
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetIOSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetIOSuite.scala
index 9746187d22..a3017258d6 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetIOSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetIOSuite.scala
@@ -469,7 +469,7 @@ class ParquetIOSuite extends QueryTest with ParquetTest with SharedSQLContext {
     }
   }
 
-  testQuietly("SPARK-9849 DirectParquetOutputCommitter qualified name backwards compatiblity") {
+  testQuietly("SPARK-9849 DirectParquetOutputCommitter qualified name backwards compatibility") {
     val clonedConf = new Configuration(hadoopConfiguration)
 
     // Write to a parquet file and let it fail.
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/streaming/FileStressSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/streaming/FileStressSuite.scala
index 3916430cdf..5b49a0a86a 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/streaming/FileStressSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/streaming/FileStressSuite.scala
@@ -29,7 +29,7 @@ import org.apache.spark.sql.test.SharedSQLContext
 import org.apache.spark.util.Utils
 
 /**
- * A stress test for streamign queries that read and write files.  This test constists of
+ * A stress test for streaming queries that read and write files.  This test consists of
  * two threads:
  *  - one that writes out `numRecords` distinct integers to files of random sizes (the total
  *    number of records is fixed but each files size / creation time is random).
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/test/TestHive.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/test/TestHive.scala
index 4afc8d18a6..9393302355 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/test/TestHive.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/test/TestHive.scala
@@ -380,8 +380,8 @@ class TestHiveContext private[hive](
        """.stripMargin.cmd,
       s"LOAD DATA LOCAL INPATH '${getHiveFile("data/files/episodes.avro")}' INTO TABLE episodes".cmd
     ),
-    // THIS TABLE IS NOT THE SAME AS THE HIVE TEST TABLE episodes_partitioned AS DYNAMIC PARITIONING
-    // IS NOT YET SUPPORTED
+    // THIS TABLE IS NOT THE SAME AS THE HIVE TEST TABLE episodes_partitioned AS DYNAMIC
+    // PARTITIONING IS NOT YET SUPPORTED
     TestTable("episodes_part",
       s"""CREATE TABLE episodes_part (title STRING, air_date STRING, doctor INT)
          |PARTITIONED BY (doctor_pt INT)
diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveComparisonTest.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveComparisonTest.scala
index 4c1b425b16..e67fcbedc3 100644
--- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveComparisonTest.scala
+++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveComparisonTest.scala
@@ -482,7 +482,7 @@ abstract class HiveComparisonTest
                 val tablesGenerated = queryList.zip(executions).flatMap {
                   // We should take executedPlan instead of sparkPlan, because in following codes we
                   // will run the collected plans. As we will do extra processing for sparkPlan such
-                  // as adding exchage, collapsing codegen stages, etc., collecing sparkPlan here
+                  // as adding exchange, collapsing codegen stages, etc., collecting sparkPlan here
                   // will cause some errors when running these plans later.
                   case (q, e) => e.executedPlan.collect {
                     case i: InsertIntoHiveTable if tablesRead contains i.table.tableName =>
diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/parquetSuites.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/parquetSuites.scala
index b6fc61d453..eac65d5720 100644
--- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/parquetSuites.scala
+++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/parquetSuites.scala
@@ -311,7 +311,7 @@ class ParquetMetastoreSuite extends ParquetPartitioningTest {
         case ExecutedCommand(_: InsertIntoHadoopFsRelation) => // OK
         case o => fail("test_insert_parquet should be converted to a " +
           s"${classOf[HadoopFsRelation ].getCanonicalName} and " +
-          s"${classOf[InsertIntoDataSource].getCanonicalName} is expcted as the SparkPlan. " +
+          s"${classOf[InsertIntoDataSource].getCanonicalName} is expected as the SparkPlan. " +
           s"However, found a ${o.toString} ")
       }
 
@@ -341,7 +341,7 @@ class ParquetMetastoreSuite extends ParquetPartitioningTest {
         case ExecutedCommand(_: InsertIntoHadoopFsRelation) => // OK
         case o => fail("test_insert_parquet should be converted to a " +
           s"${classOf[HadoopFsRelation ].getCanonicalName} and " +
-          s"${classOf[InsertIntoDataSource].getCanonicalName} is expcted as the SparkPlan." +
+          s"${classOf[InsertIntoDataSource].getCanonicalName} is expected as the SparkPlan." +
           s"However, found a ${o.toString} ")
       }
 
-- 
cgit v1.2.3


From 89f3befab6c150f87de2fb91b50ea8b414c69095 Mon Sep 17 00:00:00 2001
From: "Joseph K. Bradley" <joseph@databricks.com>
Date: Mon, 4 Apr 2016 10:24:02 -0700
Subject: [SPARK-13784][ML] Persistence for RandomForestClassifier,
 RandomForestRegressor

## What changes were proposed in this pull request?

**Main change**: Added save/load for RandomForestClassifier, RandomForestRegressor (implementation details below)

Modified numTrees method (*deprecation*)
* Goal: Use default implementations of unit tests which assume Estimators and Models share the same set of Params.
* What this PR does: Moves method numTrees outside of trait TreeEnsembleModel.  Adds it to GBT and RF Models.  Deprecates it in RF Models in favor of new method getNumTrees.  In Spark 2.1, we can have RF Models include Param numTrees.

Minor items
* Fixes bugs in GBTClassificationModel, GBTRegressionModel fromOld methods where they assign the wrong old UID.

**Implementation details**
* Split DecisionTreeModelReadWrite.loadTreeNodes into 2 methods in order to reuse some code for ensembles.
* Added EnsembleModelReadWrite object with save/load implementations usable for RFs and GBTs
  * These store all trees' nodes in a single DataFrame, and all trees' metadata in a second DataFrame.
* Split trait RandomForestParams into parts in order to add more Estimator Params to RF models
* Split DefaultParamsWriter.saveMetadata into two methods to allow ensembles to store sub-models' metadata in a single DataFrame.  Same for DefaultParamsReader.loadMetadata

## How was this patch tested?

Adds standard unit tests for RF save/load

Author: Joseph K. Bradley <joseph@databricks.com>
Author: GayathriMurali <gayathri.m.softie@gmail.com>

Closes #12118 from jkbradley/GayathriMurali-SPARK-13784.
---
 .../spark/ml/classification/GBTClassifier.scala    |   7 +-
 .../ml/classification/RandomForestClassifier.scala | 100 ++++++++++++++--
 .../apache/spark/ml/regression/GBTRegressor.scala  |   7 +-
 .../ml/regression/RandomForestRegressor.scala      |  96 +++++++++++++--
 .../org/apache/spark/ml/tree/treeModels.scala      | 131 +++++++++++++++++++--
 .../org/apache/spark/ml/tree/treeParams.scala      |  66 +++++++----
 .../scala/org/apache/spark/ml/util/ReadWrite.scala |  42 ++++++-
 .../RandomForestClassifierSuite.scala              |  40 +++----
 .../ml/regression/RandomForestRegressorSuite.scala |  38 +++---
 9 files changed, 424 insertions(+), 103 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
index 48ce051d0a..bfefaf1a1a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -192,7 +192,7 @@ final class GBTClassificationModel private[ml](
   extends PredictionModel[Vector, GBTClassificationModel]
   with TreeEnsembleModel with Serializable {
 
-  require(numTrees > 0, "GBTClassificationModel requires at least 1 tree.")
+  require(_trees.nonEmpty, "GBTClassificationModel requires at least 1 tree.")
   require(_trees.length == _treeWeights.length, "GBTClassificationModel given trees, treeWeights" +
     s" of non-matching lengths (${_trees.length}, ${_treeWeights.length}, respectively).")
 
@@ -227,6 +227,9 @@ final class GBTClassificationModel private[ml](
     if (prediction > 0.0) 1.0 else 0.0
   }
 
+  /** Number of trees in ensemble */
+  val numTrees: Int = trees.length
+
   @Since("1.4.0")
   override def copy(extra: ParamMap): GBTClassificationModel = {
     copyValues(new GBTClassificationModel(uid, _trees, _treeWeights, numFeatures),
@@ -272,6 +275,6 @@ private[ml] object GBTClassificationModel {
       DecisionTreeRegressionModel.fromOld(tree, null, categoricalFeatures)
     }
     val uid = if (parent != null) parent.uid else Identifiable.randomUID("gbtc")
-    new GBTClassificationModel(parent.uid, newTrees, oldModel.treeWeights, numFeatures)
+    new GBTClassificationModel(uid, newTrees, oldModel.treeWeights, numFeatures)
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
index 82fa05a604..2ad893f4fa 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
@@ -17,11 +17,15 @@
 
 package org.apache.spark.ml.classification
 
+import org.json4s.{DefaultFormats, JObject}
+import org.json4s.JsonDSL._
+
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml.param.ParamMap
-import org.apache.spark.ml.tree.{DecisionTreeModel, RandomForestParams, TreeClassifierParams, TreeEnsembleModel}
+import org.apache.spark.ml.tree._
 import org.apache.spark.ml.tree.impl.RandomForest
-import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
+import org.apache.spark.ml.util._
+import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.mllib.linalg.{DenseVector, SparseVector, Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
@@ -43,7 +47,7 @@ import org.apache.spark.sql.functions._
 final class RandomForestClassifier @Since("1.4.0") (
     @Since("1.4.0") override val uid: String)
   extends ProbabilisticClassifier[Vector, RandomForestClassifier, RandomForestClassificationModel]
-  with RandomForestParams with TreeClassifierParams {
+  with RandomForestClassifierParams with DefaultParamsWritable {
 
   @Since("1.4.0")
   def this() = this(Identifiable.randomUID("rfc"))
@@ -120,7 +124,7 @@ final class RandomForestClassifier @Since("1.4.0") (
 
 @Since("1.4.0")
 @Experimental
-object RandomForestClassifier {
+object RandomForestClassifier extends DefaultParamsReadable[RandomForestClassifier] {
   /** Accessor for supported impurity settings: entropy, gini */
   @Since("1.4.0")
   final val supportedImpurities: Array[String] = TreeClassifierParams.supportedImpurities
@@ -129,6 +133,9 @@ object RandomForestClassifier {
   @Since("1.4.0")
   final val supportedFeatureSubsetStrategies: Array[String] =
     RandomForestParams.supportedFeatureSubsetStrategies
+
+  @Since("2.0.0")
+  override def load(path: String): RandomForestClassifier = super.load(path)
 }
 
 /**
@@ -136,8 +143,9 @@ object RandomForestClassifier {
  * [[http://en.wikipedia.org/wiki/Random_forest  Random Forest]] model for classification.
  * It supports both binary and multiclass labels, as well as both continuous and categorical
  * features.
+ *
  * @param _trees  Decision trees in the ensemble.
- *               Warning: These have null parents.
+ *                Warning: These have null parents.
  */
 @Since("1.4.0")
 @Experimental
@@ -147,12 +155,14 @@ final class RandomForestClassificationModel private[ml] (
     @Since("1.6.0") override val numFeatures: Int,
     @Since("1.5.0") override val numClasses: Int)
   extends ProbabilisticClassificationModel[Vector, RandomForestClassificationModel]
-  with TreeEnsembleModel with Serializable {
+  with RandomForestClassificationModelParams with TreeEnsembleModel with MLWritable
+  with Serializable {
 
-  require(numTrees > 0, "RandomForestClassificationModel requires at least 1 tree.")
+  require(_trees.nonEmpty, "RandomForestClassificationModel requires at least 1 tree.")
 
   /**
    * Construct a random forest classification model, with all trees weighted equally.
+   *
    * @param trees  Component trees
    */
   private[ml] def this(
@@ -165,7 +175,7 @@ final class RandomForestClassificationModel private[ml] (
   override def trees: Array[DecisionTreeModel] = _trees.asInstanceOf[Array[DecisionTreeModel]]
 
   // Note: We may add support for weights (based on tree performance) later on.
-  private lazy val _treeWeights: Array[Double] = Array.fill[Double](numTrees)(1.0)
+  private lazy val _treeWeights: Array[Double] = Array.fill[Double](_trees.length)(1.0)
 
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
@@ -208,6 +218,15 @@ final class RandomForestClassificationModel private[ml] (
     }
   }
 
+  /**
+   * Number of trees in ensemble
+   *
+   * @deprecated  Use [[getNumTrees]] instead.  This method will be removed in 2.1.0
+   */
+  // TODO: Once this is removed, then this class can inherit from RandomForestClassifierParams
+  @deprecated("Use getNumTrees instead.  This method will be removed in 2.1.0.", "2.0.0")
+  val numTrees: Int = trees.length
+
   @Since("1.4.0")
   override def copy(extra: ParamMap): RandomForestClassificationModel = {
     copyValues(new RandomForestClassificationModel(uid, _trees, numFeatures, numClasses), extra)
@@ -216,7 +235,7 @@ final class RandomForestClassificationModel private[ml] (
 
   @Since("1.4.0")
   override def toString: String = {
-    s"RandomForestClassificationModel (uid=$uid) with $numTrees trees"
+    s"RandomForestClassificationModel (uid=$uid) with $getNumTrees trees"
   }
 
   /**
@@ -236,12 +255,69 @@ final class RandomForestClassificationModel private[ml] (
   private[ml] def toOld: OldRandomForestModel = {
     new OldRandomForestModel(OldAlgo.Classification, _trees.map(_.toOld))
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter =
+    new RandomForestClassificationModel.RandomForestClassificationModelWriter(this)
 }
 
-private[ml] object RandomForestClassificationModel {
+@Since("2.0.0")
+object RandomForestClassificationModel extends MLReadable[RandomForestClassificationModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[RandomForestClassificationModel] =
+    new RandomForestClassificationModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): RandomForestClassificationModel = super.load(path)
+
+  private[RandomForestClassificationModel]
+  class RandomForestClassificationModelWriter(instance: RandomForestClassificationModel)
+    extends MLWriter {
+
+    override protected def saveImpl(path: String): Unit = {
+      // Note: numTrees is not currently used, but could be nice to store for fast querying.
+      val extraMetadata: JObject = Map(
+        "numFeatures" -> instance.numFeatures,
+        "numClasses" -> instance.numClasses,
+        "numTrees" -> instance.getNumTrees)
+      EnsembleModelReadWrite.saveImpl(instance, path, sqlContext, extraMetadata)
+    }
+  }
+
+  private class RandomForestClassificationModelReader
+    extends MLReader[RandomForestClassificationModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[RandomForestClassificationModel].getName
+    private val treeClassName = classOf[DecisionTreeClassificationModel].getName
+
+    override def load(path: String): RandomForestClassificationModel = {
+      implicit val format = DefaultFormats
+      val (metadata: Metadata, treesData: Array[(Metadata, Node)]) =
+        EnsembleModelReadWrite.loadImpl(path, sqlContext, className, treeClassName)
+      val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
+      val numClasses = (metadata.metadata \ "numClasses").extract[Int]
+      val numTrees = (metadata.metadata \ "numTrees").extract[Int]
+
+      val trees: Array[DecisionTreeClassificationModel] = treesData.map {
+        case (treeMetadata, root) =>
+          val tree =
+            new DecisionTreeClassificationModel(treeMetadata.uid, root, numFeatures, numClasses)
+          DefaultParamsReader.getAndSetParams(tree, treeMetadata)
+          tree
+      }
+      require(numTrees == trees.length, s"RandomForestClassificationModel.load expected $numTrees" +
+        s" trees based on metadata but found ${trees.length} trees.")
+
+      val model = new RandomForestClassificationModel(metadata.uid, trees, numFeatures, numClasses)
+      DefaultParamsReader.getAndSetParams(model, metadata)
+      model
+    }
+  }
 
-  /** (private[ml]) Convert a model from the old API */
-  def fromOld(
+  /** Convert a model from the old API */
+  private[ml] def fromOld(
       oldModel: OldRandomForestModel,
       parent: RandomForestClassifier,
       categoricalFeatures: Map[Int, Int],
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
index 8fca35da51..02e124a1c0 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -179,7 +179,7 @@ final class GBTRegressionModel private[ml](
   extends PredictionModel[Vector, GBTRegressionModel]
   with TreeEnsembleModel with Serializable {
 
-  require(numTrees > 0, "GBTRegressionModel requires at least 1 tree.")
+  require(_trees.nonEmpty, "GBTRegressionModel requires at least 1 tree.")
   require(_trees.length == _treeWeights.length, "GBTRegressionModel given trees, treeWeights of" +
     s" non-matching lengths (${_trees.length}, ${_treeWeights.length}, respectively).")
 
@@ -213,6 +213,9 @@ final class GBTRegressionModel private[ml](
     blas.ddot(numTrees, treePredictions, 1, _treeWeights, 1)
   }
 
+  /** Number of trees in ensemble */
+  val numTrees: Int = trees.length
+
   @Since("1.4.0")
   override def copy(extra: ParamMap): GBTRegressionModel = {
     copyValues(new GBTRegressionModel(uid, _trees, _treeWeights, numFeatures),
@@ -258,6 +261,6 @@ private[ml] object GBTRegressionModel {
       DecisionTreeRegressionModel.fromOld(tree, null, categoricalFeatures)
     }
     val uid = if (parent != null) parent.uid else Identifiable.randomUID("gbtr")
-    new GBTRegressionModel(parent.uid, newTrees, oldModel.treeWeights, numFeatures)
+    new GBTRegressionModel(uid, newTrees, oldModel.treeWeights, numFeatures)
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
index 5b3f3a1f5d..ba56b5cd3f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
@@ -17,12 +17,16 @@
 
 package org.apache.spark.ml.regression
 
+import org.json4s.{DefaultFormats, JObject}
+import org.json4s.JsonDSL._
+
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml.{PredictionModel, Predictor}
 import org.apache.spark.ml.param.ParamMap
-import org.apache.spark.ml.tree.{DecisionTreeModel, RandomForestParams, TreeEnsembleModel, TreeRegressorParams}
+import org.apache.spark.ml.tree._
 import org.apache.spark.ml.tree.impl.RandomForest
-import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
+import org.apache.spark.ml.util._
+import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
@@ -41,7 +45,7 @@ import org.apache.spark.sql.functions._
 @Experimental
 final class RandomForestRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: String)
   extends Predictor[Vector, RandomForestRegressor, RandomForestRegressionModel]
-  with RandomForestParams with TreeRegressorParams {
+  with RandomForestRegressorParams with DefaultParamsWritable {
 
   @Since("1.4.0")
   def this() = this(Identifiable.randomUID("rfr"))
@@ -108,7 +112,7 @@ final class RandomForestRegressor @Since("1.4.0") (@Since("1.4.0") override val
 
 @Since("1.4.0")
 @Experimental
-object RandomForestRegressor {
+object RandomForestRegressor extends DefaultParamsReadable[RandomForestRegressor]{
   /** Accessor for supported impurity settings: variance */
   @Since("1.4.0")
   final val supportedImpurities: Array[String] = TreeRegressorParams.supportedImpurities
@@ -117,12 +121,17 @@ object RandomForestRegressor {
   @Since("1.4.0")
   final val supportedFeatureSubsetStrategies: Array[String] =
     RandomForestParams.supportedFeatureSubsetStrategies
+
+  @Since("2.0.0")
+  override def load(path: String): RandomForestRegressor = super.load(path)
+
 }
 
 /**
  * :: Experimental ::
  * [[http://en.wikipedia.org/wiki/Random_forest  Random Forest]] model for regression.
  * It supports both continuous and categorical features.
+ *
  * @param _trees  Decision trees in the ensemble.
  * @param numFeatures  Number of features used by this model
  */
@@ -133,12 +142,13 @@ final class RandomForestRegressionModel private[ml] (
     private val _trees: Array[DecisionTreeRegressionModel],
     override val numFeatures: Int)
   extends PredictionModel[Vector, RandomForestRegressionModel]
-  with TreeEnsembleModel with Serializable {
+  with RandomForestRegressionModelParams with TreeEnsembleModel with MLWritable with Serializable {
 
-  require(numTrees > 0, "RandomForestRegressionModel requires at least 1 tree.")
+  require(_trees.nonEmpty, "RandomForestRegressionModel requires at least 1 tree.")
 
   /**
    * Construct a random forest regression model, with all trees weighted equally.
+   *
    * @param trees  Component trees
    */
   private[ml] def this(trees: Array[DecisionTreeRegressionModel], numFeatures: Int) =
@@ -148,7 +158,7 @@ final class RandomForestRegressionModel private[ml] (
   override def trees: Array[DecisionTreeModel] = _trees.asInstanceOf[Array[DecisionTreeModel]]
 
   // Note: We may add support for weights (based on tree performance) later on.
-  private lazy val _treeWeights: Array[Double] = Array.fill[Double](numTrees)(1.0)
+  private lazy val _treeWeights: Array[Double] = Array.fill[Double](_trees.length)(1.0)
 
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
@@ -165,9 +175,17 @@ final class RandomForestRegressionModel private[ml] (
     // TODO: When we add a generic Bagging class, handle transform there.  SPARK-7128
     // Predict average of tree predictions.
     // Ignore the weights since all are 1.0 for now.
-    _trees.map(_.rootNode.predictImpl(features).prediction).sum / numTrees
+    _trees.map(_.rootNode.predictImpl(features).prediction).sum / getNumTrees
   }
 
+  /**
+   * Number of trees in ensemble
+   * @deprecated  Use [[getNumTrees]] instead.  This method will be removed in 2.1.0
+   */
+  // TODO: Once this is removed, then this class can inherit from RandomForestRegressorParams
+  @deprecated("Use getNumTrees instead.  This method will be removed in 2.1.0.", "2.0.0")
+  val numTrees: Int = trees.length
+
   @Since("1.4.0")
   override def copy(extra: ParamMap): RandomForestRegressionModel = {
     copyValues(new RandomForestRegressionModel(uid, _trees, numFeatures), extra).setParent(parent)
@@ -175,7 +193,7 @@ final class RandomForestRegressionModel private[ml] (
 
   @Since("1.4.0")
   override def toString: String = {
-    s"RandomForestRegressionModel (uid=$uid) with $numTrees trees"
+    s"RandomForestRegressionModel (uid=$uid) with $getNumTrees trees"
   }
 
   /**
@@ -195,12 +213,63 @@ final class RandomForestRegressionModel private[ml] (
   private[ml] def toOld: OldRandomForestModel = {
     new OldRandomForestModel(OldAlgo.Regression, _trees.map(_.toOld))
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter =
+    new RandomForestRegressionModel.RandomForestRegressionModelWriter(this)
 }
 
-private[ml] object RandomForestRegressionModel {
+@Since("2.0.0")
+object RandomForestRegressionModel extends MLReadable[RandomForestRegressionModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[RandomForestRegressionModel] = new RandomForestRegressionModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): RandomForestRegressionModel = super.load(path)
+
+  private[RandomForestRegressionModel]
+  class RandomForestRegressionModelWriter(instance: RandomForestRegressionModel)
+    extends MLWriter {
+
+    override protected def saveImpl(path: String): Unit = {
+      val extraMetadata: JObject = Map(
+        "numFeatures" -> instance.numFeatures,
+        "numTrees" -> instance.getNumTrees)
+      EnsembleModelReadWrite.saveImpl(instance, path, sqlContext, extraMetadata)
+    }
+  }
+
+  private class RandomForestRegressionModelReader extends MLReader[RandomForestRegressionModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[RandomForestRegressionModel].getName
+    private val treeClassName = classOf[DecisionTreeRegressionModel].getName
+
+    override def load(path: String): RandomForestRegressionModel = {
+      implicit val format = DefaultFormats
+      val (metadata: Metadata, treesData: Array[(Metadata, Node)]) =
+        EnsembleModelReadWrite.loadImpl(path, sqlContext, className, treeClassName)
+      val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
+      val numTrees = (metadata.metadata \ "numTrees").extract[Int]
+
+      val trees: Array[DecisionTreeRegressionModel] = treesData.map { case (treeMetadata, root) =>
+        val tree =
+          new DecisionTreeRegressionModel(treeMetadata.uid, root, numFeatures)
+        DefaultParamsReader.getAndSetParams(tree, treeMetadata)
+        tree
+      }
+      require(numTrees == trees.length, s"RandomForestRegressionModel.load expected $numTrees" +
+        s" trees based on metadata but found ${trees.length} trees.")
+
+      val model = new RandomForestRegressionModel(metadata.uid, trees, numFeatures)
+      DefaultParamsReader.getAndSetParams(model, metadata)
+      model
+    }
+  }
 
-  /** (private[ml]) Convert a model from the old API */
-  def fromOld(
+  /** Convert a model from the old API */
+  private[ml] def fromOld(
       oldModel: OldRandomForestModel,
       parent: RandomForestRegressor,
       categoricalFeatures: Map[Int, Int],
@@ -211,6 +280,7 @@ private[ml] object RandomForestRegressionModel {
       // parent for each tree is null since there is no good way to set this.
       DecisionTreeRegressionModel.fromOld(tree, null, categoricalFeatures)
     }
-    new RandomForestRegressionModel(parent.uid, newTrees, numFeatures)
+    val uid = if (parent != null) parent.uid else Identifiable.randomUID("rfr")
+    new RandomForestRegressionModel(uid, newTrees, numFeatures)
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
index 8ea767b2b3..48b8fd19ad 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
@@ -21,12 +21,15 @@ import org.apache.hadoop.fs.Path
 import org.json4s._
 import org.json4s.jackson.JsonMethods._
 
-import org.apache.spark.ml.param.Param
-import org.apache.spark.ml.util.DefaultParamsReader
+import org.apache.spark.ml.param.{Param, Params}
+import org.apache.spark.ml.tree.DecisionTreeModelReadWrite.NodeData
+import org.apache.spark.ml.util.{DefaultParamsReader, DefaultParamsWriter}
+import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.tree.impurity.ImpurityCalculator
 import org.apache.spark.mllib.tree.model.{DecisionTreeModel => OldDecisionTreeModel}
-import org.apache.spark.sql.SQLContext
+import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.{Dataset, SQLContext}
 import org.apache.spark.util.collection.OpenHashMap
 
 /**
@@ -88,6 +91,11 @@ private[ml] trait TreeEnsembleModel {
   /** Trees in this ensemble. Warning: These have null parent Estimators. */
   def trees: Array[DecisionTreeModel]
 
+  /**
+   * Number of trees in ensemble
+   */
+  val getNumTrees: Int = trees.length
+
   /** Weights for each tree, zippable with [[trees]] */
   def treeWeights: Array[Double]
 
@@ -98,7 +106,7 @@ private[ml] trait TreeEnsembleModel {
   /** Summary of the model */
   override def toString: String = {
     // Implementing classes should generally override this method to be more descriptive.
-    s"TreeEnsembleModel with $numTrees trees"
+    s"TreeEnsembleModel with ${trees.length} trees"
   }
 
   /** Full description of model */
@@ -109,9 +117,6 @@ private[ml] trait TreeEnsembleModel {
     }.fold("")(_ + _)
   }
 
-  /** Number of trees in ensemble */
-  val numTrees: Int = trees.length
-
   /** Total number of nodes, summed over all trees in the ensemble. */
   lazy val totalNumNodes: Int = trees.map(_.numNodes).sum
 }
@@ -316,6 +321,10 @@ private[ml] object DecisionTreeModelReadWrite {
     }
   }
 
+  /**
+   * Load a decision tree from a file.
+   * @return  Root node of reconstructed tree
+   */
   def loadTreeNodes(
       path: String,
       metadata: DefaultParamsReader.Metadata,
@@ -331,9 +340,18 @@ private[ml] object DecisionTreeModelReadWrite {
 
     val dataPath = new Path(path, "data").toString
     val data = sqlContext.read.parquet(dataPath).as[NodeData]
+    buildTreeFromNodes(data.collect(), impurityType)
+  }
 
+  /**
+   * Given all data for all nodes in a tree, rebuild the tree.
+   * @param data  Unsorted node data
+   * @param impurityType  Impurity type for this tree
+   * @return Root node of reconstructed tree
+   */
+  def buildTreeFromNodes(data: Array[NodeData], impurityType: String): Node = {
     // Load all nodes, sorted by ID.
-    val nodes: Array[NodeData] = data.collect().sortBy(_.id)
+    val nodes = data.sortBy(_.id)
     // Sanity checks; could remove
     assert(nodes.head.id == 0, s"Decision Tree load failed.  Expected smallest node ID to be 0," +
       s" but found ${nodes.head.id}")
@@ -358,3 +376,100 @@ private[ml] object DecisionTreeModelReadWrite {
     finalNodes.head
   }
 }
+
+private[ml] object EnsembleModelReadWrite {
+
+  /**
+   * Helper method for saving a tree ensemble to disk.
+   *
+   * @param instance  Tree ensemble model
+   * @param path  Path to which to save the ensemble model.
+   * @param extraMetadata  Metadata such as numFeatures, numClasses, numTrees.
+   */
+  def saveImpl[M <: Params with TreeEnsembleModel](
+      instance: M,
+      path: String,
+      sql: SQLContext,
+      extraMetadata: JObject): Unit = {
+    DefaultParamsWriter.saveMetadata(instance, path, sql.sparkContext, Some(extraMetadata))
+    val treesMetadataJson: Array[(Int, String)] = instance.trees.zipWithIndex.map {
+      case (tree, treeID) =>
+        treeID -> DefaultParamsWriter.getMetadataToSave(tree.asInstanceOf[Params], sql.sparkContext)
+    }
+    val treesMetadataPath = new Path(path, "treesMetadata").toString
+    sql.createDataFrame(treesMetadataJson).toDF("treeID", "metadata")
+      .write.parquet(treesMetadataPath)
+    val dataPath = new Path(path, "data").toString
+    val nodeDataRDD = sql.sparkContext.parallelize(instance.trees.zipWithIndex).flatMap {
+      case (tree, treeID) => EnsembleNodeData.build(tree, treeID)
+    }
+    sql.createDataFrame(nodeDataRDD).write.parquet(dataPath)
+  }
+
+  /**
+   * Helper method for loading a tree ensemble from disk.
+   * This reconstructs all trees, returning the root nodes.
+   * @param path  Path given to [[saveImpl()]]
+   * @param className  Class name for ensemble model type
+   * @param treeClassName  Class name for tree model type in the ensemble
+   * @return  (ensemble metadata, array over trees of (tree metadata, root node)),
+   *          where the root node is linked with all descendents
+   * @see [[saveImpl()]] for how the model was saved
+   */
+  def loadImpl(
+      path: String,
+      sql: SQLContext,
+      className: String,
+      treeClassName: String): (Metadata, Array[(Metadata, Node)]) = {
+    import sql.implicits._
+    implicit val format = DefaultFormats
+    val metadata = DefaultParamsReader.loadMetadata(path, sql.sparkContext, className)
+
+    // Get impurity to construct ImpurityCalculator for each node
+    val impurityType: String = {
+      val impurityJson: JValue = metadata.getParamValue("impurity")
+      Param.jsonDecode[String](compact(render(impurityJson)))
+    }
+
+    val treesMetadataPath = new Path(path, "treesMetadata").toString
+    val treesMetadataRDD: RDD[(Int, Metadata)] = sql.read.parquet(treesMetadataPath)
+      .select("treeID", "metadata").as[(Int, String)].rdd.map {
+      case (treeID: Int, json: String) =>
+        treeID -> DefaultParamsReader.parseMetadata(json, treeClassName)
+    }
+    val treesMetadata: Array[Metadata] = treesMetadataRDD.sortByKey().values.collect()
+
+    val dataPath = new Path(path, "data").toString
+    val nodeData: Dataset[EnsembleNodeData] =
+      sql.read.parquet(dataPath).as[EnsembleNodeData]
+    val rootNodesRDD: RDD[(Int, Node)] =
+      nodeData.rdd.map(d => (d.treeID, d.nodeData)).groupByKey().map {
+        case (treeID: Int, nodeData: Iterable[NodeData]) =>
+          treeID -> DecisionTreeModelReadWrite.buildTreeFromNodes(nodeData.toArray, impurityType)
+      }
+    val rootNodes: Array[Node] = rootNodesRDD.sortByKey().values.collect()
+    (metadata, treesMetadata.zip(rootNodes))
+  }
+
+  /**
+   * Info for one [[Node]] in a tree ensemble
+   *
+   * @param treeID  Tree index
+   * @param nodeData  Data for this node
+   */
+  case class EnsembleNodeData(
+      treeID: Int,
+      nodeData: NodeData)
+
+  object EnsembleNodeData {
+    /**
+     * Create [[EnsembleNodeData]] instances for the given tree.
+     *
+     * @return Sequence of nodes for this tree
+     */
+    def build(tree: DecisionTreeModel, treeID: Int): Seq[EnsembleNodeData] = {
+      val (nodeData: Seq[NodeData], _) = NodeData.build(tree.rootNode, 0)
+      nodeData.map(nd => EnsembleNodeData(treeID, nd))
+    }
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
index 4fbd957677..78e6d3bfac 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
@@ -315,22 +315,8 @@ private[ml] trait TreeEnsembleParams extends DecisionTreeParams {
   }
 }
 
-/**
- * Parameters for Random Forest algorithms.
- *
- * Note: Marked as private and DeveloperApi since this may be made public in the future.
- */
-private[ml] trait RandomForestParams extends TreeEnsembleParams {
-
-  /**
-   * Number of trees to train (>= 1).
-   * If 1, then no bootstrapping is used.  If > 1, then bootstrapping is done.
-   * TODO: Change to always do bootstrapping (simpler).  SPARK-7130
-   * (default = 20)
-   * @group param
-   */
-  final val numTrees: IntParam = new IntParam(this, "numTrees", "Number of trees to train (>= 1)",
-    ParamValidators.gtEq(1))
+/** Used for [[RandomForestParams]] */
+private[ml] trait HasFeatureSubsetStrategy extends Params {
 
   /**
    * The number of features to consider for splits at each tree node.
@@ -362,27 +348,65 @@ private[ml] trait RandomForestParams extends TreeEnsembleParams {
     (value: String) =>
       RandomForestParams.supportedFeatureSubsetStrategies.contains(value.toLowerCase))
 
-  setDefault(numTrees -> 20, featureSubsetStrategy -> "auto")
+  setDefault(featureSubsetStrategy -> "auto")
 
   /** @group setParam */
-  def setNumTrees(value: Int): this.type = set(numTrees, value)
+  def setFeatureSubsetStrategy(value: String): this.type = set(featureSubsetStrategy, value)
 
   /** @group getParam */
-  final def getNumTrees: Int = $(numTrees)
+  final def getFeatureSubsetStrategy: String = $(featureSubsetStrategy).toLowerCase
+}
+
+/**
+ * Used for [[RandomForestParams]].
+ * This is separated out from [[RandomForestParams]] because of an issue with the
+ * `numTrees` method conflicting with this Param in the Estimator.
+ */
+private[ml] trait HasNumTrees extends Params {
+
+  /**
+   * Number of trees to train (>= 1).
+   * If 1, then no bootstrapping is used.  If > 1, then bootstrapping is done.
+   * TODO: Change to always do bootstrapping (simpler).  SPARK-7130
+   * (default = 20)
+   * @group param
+   */
+  final val numTrees: IntParam = new IntParam(this, "numTrees", "Number of trees to train (>= 1)",
+    ParamValidators.gtEq(1))
+
+  setDefault(numTrees -> 20)
 
   /** @group setParam */
-  def setFeatureSubsetStrategy(value: String): this.type = set(featureSubsetStrategy, value)
+  def setNumTrees(value: Int): this.type = set(numTrees, value)
 
   /** @group getParam */
-  final def getFeatureSubsetStrategy: String = $(featureSubsetStrategy).toLowerCase
+  final def getNumTrees: Int = $(numTrees)
 }
 
+/**
+ * Parameters for Random Forest algorithms.
+ */
+private[ml] trait RandomForestParams extends TreeEnsembleParams
+  with HasFeatureSubsetStrategy with HasNumTrees
+
 private[spark] object RandomForestParams {
   // These options should be lowercase.
   final val supportedFeatureSubsetStrategies: Array[String] =
     Array("auto", "all", "onethird", "sqrt", "log2").map(_.toLowerCase)
 }
 
+private[ml] trait RandomForestClassifierParams
+  extends RandomForestParams with TreeClassifierParams
+
+private[ml] trait RandomForestClassificationModelParams extends TreeEnsembleParams
+  with HasFeatureSubsetStrategy with TreeClassifierParams
+
+private[ml] trait RandomForestRegressorParams
+  extends RandomForestParams with TreeRegressorParams
+
+private[ml] trait RandomForestRegressionModelParams extends TreeEnsembleParams
+  with HasFeatureSubsetStrategy with TreeRegressorParams
+
 /**
  * Parameters for Gradient-Boosted Tree algorithms.
  *
diff --git a/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala b/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
index 39999ede30..7dec07ea14 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/util/ReadWrite.scala
@@ -144,6 +144,7 @@ private[ml] trait DefaultParamsWritable extends MLWritable { self: Params =>
 
 /**
  * Abstract class for utility classes that can load ML instances.
+ *
  * @tparam T ML instance type
  */
 @Experimental
@@ -162,6 +163,7 @@ abstract class MLReader[T] extends BaseReadWrite {
 
 /**
  * Trait for objects that provide [[MLReader]].
+ *
  * @tparam T ML instance type
  */
 @Experimental
@@ -192,6 +194,7 @@ private[ml] trait DefaultParamsReadable[T] extends MLReadable[T] {
  * Default [[MLWriter]] implementation for transformers and estimators that contain basic
  * (json4s-serializable) params and no data. This will not handle more complex params or types with
  * data (e.g., models with coefficients).
+ *
  * @param instance object to save
  */
 private[ml] class DefaultParamsWriter(instance: Params) extends MLWriter {
@@ -211,6 +214,7 @@ private[ml] object DefaultParamsWriter {
    *  - uid
    *  - paramMap
    *  - (optionally, extra metadata)
+   *
    * @param extraMetadata  Extra metadata to be saved at same level as uid, paramMap, etc.
    * @param paramMap  If given, this is saved in the "paramMap" field.
    *                  Otherwise, all [[org.apache.spark.ml.param.Param]]s are encoded using
@@ -222,6 +226,22 @@ private[ml] object DefaultParamsWriter {
       sc: SparkContext,
       extraMetadata: Option[JObject] = None,
       paramMap: Option[JValue] = None): Unit = {
+    val metadataPath = new Path(path, "metadata").toString
+    val metadataJson = getMetadataToSave(instance, sc, extraMetadata, paramMap)
+    sc.parallelize(Seq(metadataJson), 1).saveAsTextFile(metadataPath)
+  }
+
+  /**
+   * Helper for [[saveMetadata()]] which extracts the JSON to save.
+   * This is useful for ensemble models which need to save metadata for many sub-models.
+   *
+   * @see [[saveMetadata()]] for details on what this includes.
+   */
+  def getMetadataToSave(
+      instance: Params,
+      sc: SparkContext,
+      extraMetadata: Option[JObject] = None,
+      paramMap: Option[JValue] = None): String = {
     val uid = instance.uid
     val cls = instance.getClass.getName
     val params = instance.extractParamMap().toSeq.asInstanceOf[Seq[ParamPair[Any]]]
@@ -239,9 +259,8 @@ private[ml] object DefaultParamsWriter {
       case None =>
         basicMetadata
     }
-    val metadataPath = new Path(path, "metadata").toString
-    val metadataJson = compact(render(metadata))
-    sc.parallelize(Seq(metadataJson), 1).saveAsTextFile(metadataPath)
+    val metadataJson: String = compact(render(metadata))
+    metadataJson
   }
 }
 
@@ -249,6 +268,7 @@ private[ml] object DefaultParamsWriter {
  * Default [[MLReader]] implementation for transformers and estimators that contain basic
  * (json4s-serializable) params and no data. This will not handle more complex params or types with
  * data (e.g., models with coefficients).
+ *
  * @tparam T ML instance type
  * TODO: Consider adding check for correct class name.
  */
@@ -268,6 +288,7 @@ private[ml] object DefaultParamsReader {
 
   /**
    * All info from metadata file.
+   *
    * @param params  paramMap, as a [[JValue]]
    * @param metadata  All metadata, including the other fields
    * @param metadataJson  Full metadata file String (for debugging)
@@ -304,13 +325,26 @@ private[ml] object DefaultParamsReader {
   }
 
   /**
-   * Load metadata from file.
+   * Load metadata saved using [[DefaultParamsWriter.saveMetadata()]]
+   *
    * @param expectedClassName  If non empty, this is checked against the loaded metadata.
    * @throws IllegalArgumentException if expectedClassName is specified and does not match metadata
    */
   def loadMetadata(path: String, sc: SparkContext, expectedClassName: String = ""): Metadata = {
     val metadataPath = new Path(path, "metadata").toString
     val metadataStr = sc.textFile(metadataPath, 1).first()
+    parseMetadata(metadataStr, expectedClassName)
+  }
+
+  /**
+   * Parse metadata JSON string produced by [[DefaultParamsWriter.getMetadataToSave()]].
+   * This is a helper function for [[loadMetadata()]].
+   *
+   * @param metadataStr  JSON string of metadata
+   * @param expectedClassName  If non empty, this is checked against the loaded metadata.
+   * @throws IllegalArgumentException if expectedClassName is specified and does not match metadata
+   */
+  def parseMetadata(metadataStr: String, expectedClassName: String = ""): Metadata = {
     val metadata = parse(metadataStr)
 
     implicit val format = DefaultFormats
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala
index 052bc83c38..aaaa429103 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/RandomForestClassifierSuite.scala
@@ -21,7 +21,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.param.ParamsSuite
 import org.apache.spark.ml.tree.LeafNode
 import org.apache.spark.ml.tree.impl.TreeTests
-import org.apache.spark.ml.util.MLTestingUtils
+import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.{EnsembleTestHelper, RandomForest => OldRandomForest}
@@ -34,7 +34,8 @@ import org.apache.spark.sql.{DataFrame, Row}
 /**
  * Test suite for [[RandomForestClassifier]].
  */
-class RandomForestClassifierSuite extends SparkFunSuite with MLlibTestSparkContext {
+class RandomForestClassifierSuite
+  extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
   import RandomForestClassifierSuite.compareAPIs
 
@@ -190,27 +191,24 @@ class RandomForestClassifierSuite extends SparkFunSuite with MLlibTestSparkConte
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
 
-  // TODO: Reinstate test once save/load are implemented  SPARK-6725
-  /*
-  test("model save/load") {
-    val tempDir = Utils.createTempDir()
-    val path = tempDir.toURI.toString
-
-    val trees =
-      Range(0, 3).map(_ => OldDecisionTreeSuite.createModel(OldAlgo.Classification)).toArray
-    val oldModel = new OldRandomForestModel(OldAlgo.Classification, trees)
-    val newModel = RandomForestClassificationModel.fromOld(oldModel)
-
-    // Save model, load it back, and compare.
-    try {
-      newModel.save(sc, path)
-      val sameNewModel = RandomForestClassificationModel.load(sc, path)
-      TreeTests.checkEqual(newModel, sameNewModel)
-    } finally {
-      Utils.deleteRecursively(tempDir)
+  test("read/write") {
+    def checkModelData(
+        model: RandomForestClassificationModel,
+        model2: RandomForestClassificationModel): Unit = {
+      TreeTests.checkEqual(model, model2)
+      assert(model.numFeatures === model2.numFeatures)
+      assert(model.numClasses === model2.numClasses)
     }
+
+    val rf = new RandomForestClassifier().setNumTrees(2)
+    val rdd = TreeTests.getTreeReadWriteData(sc)
+
+    val allParamSettings = TreeTests.allParamSettings ++ Map("impurity" -> "entropy")
+
+    val continuousData: DataFrame =
+      TreeTests.setMetadata(rdd, Map.empty[Int, Int], numClasses = 2)
+    testEstimatorAndModelReadWrite(rf, continuousData, allParamSettings, checkModelData)
   }
-  */
 }
 
 private object RandomForestClassifierSuite extends SparkFunSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala
index 2ab4f1b146..ca400e1914 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/RandomForestRegressorSuite.scala
@@ -19,7 +19,7 @@ package org.apache.spark.ml.regression
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.tree.impl.TreeTests
-import org.apache.spark.ml.util.MLTestingUtils
+import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.{EnsembleTestHelper, RandomForest => OldRandomForest}
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
@@ -30,7 +30,8 @@ import org.apache.spark.sql.DataFrame
 /**
  * Test suite for [[RandomForestRegressor]].
  */
-class RandomForestRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
+class RandomForestRegressorSuite extends SparkFunSuite with MLlibTestSparkContext
+  with DefaultReadWriteTest{
 
   import RandomForestRegressorSuite.compareAPIs
 
@@ -106,26 +107,23 @@ class RandomForestRegressorSuite extends SparkFunSuite with MLlibTestSparkContex
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
 
-  // TODO: Reinstate test once save/load are implemented  SPARK-6725
-  /*
-  test("model save/load") {
-    val tempDir = Utils.createTempDir()
-    val path = tempDir.toURI.toString
-
-    val trees = Range(0, 3).map(_ => OldDecisionTreeSuite.createModel(OldAlgo.Regression)).toArray
-    val oldModel = new OldRandomForestModel(OldAlgo.Regression, trees)
-    val newModel = RandomForestRegressionModel.fromOld(oldModel)
-
-    // Save model, load it back, and compare.
-    try {
-      newModel.save(sc, path)
-      val sameNewModel = RandomForestRegressionModel.load(sc, path)
-      TreeTests.checkEqual(newModel, sameNewModel)
-    } finally {
-      Utils.deleteRecursively(tempDir)
+  test("read/write") {
+    def checkModelData(
+        model: RandomForestRegressionModel,
+        model2: RandomForestRegressionModel): Unit = {
+      TreeTests.checkEqual(model, model2)
+      assert(model.numFeatures === model2.numFeatures)
     }
+
+    val rf = new RandomForestRegressor().setNumTrees(2)
+    val rdd = TreeTests.getTreeReadWriteData(sc)
+
+    val allParamSettings = TreeTests.allParamSettings ++ Map("impurity" -> "variance")
+
+    val continuousData: DataFrame =
+      TreeTests.setMetadata(rdd, Map.empty[Int, Int], numClasses = 0)
+    testEstimatorAndModelReadWrite(rf, continuousData, allParamSettings, checkModelData)
   }
-  */
 }
 
 private object RandomForestRegressorSuite extends SparkFunSuite {
-- 
cgit v1.2.3


From 8f50574ab4021b9984b0017cd47ba012a894c19a Mon Sep 17 00:00:00 2001
From: "Joseph K. Bradley" <joseph@databricks.com>
Date: Mon, 4 Apr 2016 20:12:09 -0700
Subject: [SPARK-14386][ML] Changed spark.ml ensemble trees methods to return
 concrete types

## What changes were proposed in this pull request?

In spark.ml, GBT and RandomForest expose the trait DecisionTreeModel in the trees method, but they should not since it is a private trait (and not ready to be made public). It will also be more useful to users if we return the concrete types.

This PR: return concrete types

The MIMA checks appear to be OK with this change.

## How was this patch tested?

Existing unit tests

Author: Joseph K. Bradley <joseph@databricks.com>

Closes #12158 from jkbradley/hide-dtm.
---
 .../org/apache/spark/ml/classification/GBTClassifier.scala |  7 +++----
 .../spark/ml/classification/RandomForestClassifier.scala   |  6 +++---
 .../org/apache/spark/ml/regression/GBTRegressor.scala      |  7 +++----
 .../apache/spark/ml/regression/RandomForestRegressor.scala |  5 +++--
 .../main/scala/org/apache/spark/ml/tree/treeModels.scala   | 14 +++++++++-----
 .../scala/org/apache/spark/ml/tree/impl/TreeTests.scala    |  2 +-
 6 files changed, 22 insertions(+), 19 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
index bfefaf1a1a..bee90fb3a5 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -24,8 +24,7 @@ import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{PredictionModel, Predictor}
 import org.apache.spark.ml.param.{Param, ParamMap}
 import org.apache.spark.ml.regression.DecisionTreeRegressionModel
-import org.apache.spark.ml.tree.{DecisionTreeModel, GBTParams, TreeClassifierParams,
-  TreeEnsembleModel}
+import org.apache.spark.ml.tree.{GBTParams, TreeClassifierParams, TreeEnsembleModel}
 import org.apache.spark.ml.tree.impl.GradientBoostedTrees
 import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
 import org.apache.spark.mllib.linalg.Vector
@@ -190,7 +189,7 @@ final class GBTClassificationModel private[ml](
     private val _treeWeights: Array[Double],
     @Since("1.6.0") override val numFeatures: Int)
   extends PredictionModel[Vector, GBTClassificationModel]
-  with TreeEnsembleModel with Serializable {
+  with TreeEnsembleModel[DecisionTreeRegressionModel] with Serializable {
 
   require(_trees.nonEmpty, "GBTClassificationModel requires at least 1 tree.")
   require(_trees.length == _treeWeights.length, "GBTClassificationModel given trees, treeWeights" +
@@ -206,7 +205,7 @@ final class GBTClassificationModel private[ml](
     this(uid, _trees, _treeWeights, -1)
 
   @Since("1.4.0")
-  override def trees: Array[DecisionTreeModel] = _trees.asInstanceOf[Array[DecisionTreeModel]]
+  override def trees: Array[DecisionTreeRegressionModel] = _trees
 
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
index 2ad893f4fa..cb42532271 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
@@ -155,8 +155,8 @@ final class RandomForestClassificationModel private[ml] (
     @Since("1.6.0") override val numFeatures: Int,
     @Since("1.5.0") override val numClasses: Int)
   extends ProbabilisticClassificationModel[Vector, RandomForestClassificationModel]
-  with RandomForestClassificationModelParams with TreeEnsembleModel with MLWritable
-  with Serializable {
+  with RandomForestClassificationModelParams with TreeEnsembleModel[DecisionTreeClassificationModel]
+  with MLWritable with Serializable {
 
   require(_trees.nonEmpty, "RandomForestClassificationModel requires at least 1 tree.")
 
@@ -172,7 +172,7 @@ final class RandomForestClassificationModel private[ml] (
     this(Identifiable.randomUID("rfc"), trees, numFeatures, numClasses)
 
   @Since("1.4.0")
-  override def trees: Array[DecisionTreeModel] = _trees.asInstanceOf[Array[DecisionTreeModel]]
+  override def trees: Array[DecisionTreeClassificationModel] = _trees
 
   // Note: We may add support for weights (based on tree performance) later on.
   private lazy val _treeWeights: Array[Double] = Array.fill[Double](_trees.length)(1.0)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
index 02e124a1c0..cef7c643d7 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -23,8 +23,7 @@ import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{PredictionModel, Predictor}
 import org.apache.spark.ml.param.{Param, ParamMap}
-import org.apache.spark.ml.tree.{DecisionTreeModel, GBTParams, TreeEnsembleModel,
-  TreeRegressorParams}
+import org.apache.spark.ml.tree.{GBTParams, TreeEnsembleModel, TreeRegressorParams}
 import org.apache.spark.ml.tree.impl.GradientBoostedTrees
 import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
 import org.apache.spark.mllib.linalg.Vector
@@ -177,7 +176,7 @@ final class GBTRegressionModel private[ml](
     private val _treeWeights: Array[Double],
     override val numFeatures: Int)
   extends PredictionModel[Vector, GBTRegressionModel]
-  with TreeEnsembleModel with Serializable {
+  with TreeEnsembleModel[DecisionTreeRegressionModel] with Serializable {
 
   require(_trees.nonEmpty, "GBTRegressionModel requires at least 1 tree.")
   require(_trees.length == _treeWeights.length, "GBTRegressionModel given trees, treeWeights of" +
@@ -193,7 +192,7 @@ final class GBTRegressionModel private[ml](
     this(uid, _trees, _treeWeights, -1)
 
   @Since("1.4.0")
-  override def trees: Array[DecisionTreeModel] = _trees.asInstanceOf[Array[DecisionTreeModel]]
+  override def trees: Array[DecisionTreeRegressionModel] = _trees
 
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
index ba56b5cd3f..736cd9f776 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
@@ -142,7 +142,8 @@ final class RandomForestRegressionModel private[ml] (
     private val _trees: Array[DecisionTreeRegressionModel],
     override val numFeatures: Int)
   extends PredictionModel[Vector, RandomForestRegressionModel]
-  with RandomForestRegressionModelParams with TreeEnsembleModel with MLWritable with Serializable {
+  with RandomForestRegressionModelParams with TreeEnsembleModel[DecisionTreeRegressionModel]
+  with MLWritable with Serializable {
 
   require(_trees.nonEmpty, "RandomForestRegressionModel requires at least 1 tree.")
 
@@ -155,7 +156,7 @@ final class RandomForestRegressionModel private[ml] (
     this(Identifiable.randomUID("rfr"), trees, numFeatures)
 
   @Since("1.4.0")
-  override def trees: Array[DecisionTreeModel] = _trees.asInstanceOf[Array[DecisionTreeModel]]
+  override def trees: Array[DecisionTreeRegressionModel] = _trees
 
   // Note: We may add support for weights (based on tree performance) later on.
   private lazy val _treeWeights: Array[Double] = Array.fill[Double](_trees.length)(1.0)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
index 48b8fd19ad..db0ff28d82 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
@@ -17,6 +17,8 @@
 
 package org.apache.spark.ml.tree
 
+import scala.reflect.ClassTag
+
 import org.apache.hadoop.fs.Path
 import org.json4s._
 import org.json4s.jackson.JsonMethods._
@@ -82,14 +84,16 @@ private[spark] trait DecisionTreeModel {
  * Abstraction for models which are ensembles of decision trees
  *
  * TODO: Add support for predicting probabilities and raw predictions  SPARK-3727
+ *
+ * @tparam M  Type of tree model in this ensemble
  */
-private[ml] trait TreeEnsembleModel {
+private[ml] trait TreeEnsembleModel[M <: DecisionTreeModel] {
 
   // Note: We use getTrees since subclasses of TreeEnsembleModel will store subclasses of
   //       DecisionTreeModel.
 
   /** Trees in this ensemble. Warning: These have null parent Estimators. */
-  def trees: Array[DecisionTreeModel]
+  def trees: Array[M]
 
   /**
    * Number of trees in ensemble
@@ -148,7 +152,7 @@ private[ml] object TreeEnsembleModel {
    *                     If -1, then numFeatures is set based on the max feature index in all trees.
    * @return  Feature importance values, of length numFeatures.
    */
-  def featureImportances(trees: Array[DecisionTreeModel], numFeatures: Int): Vector = {
+  def featureImportances[M <: DecisionTreeModel](trees: Array[M], numFeatures: Int): Vector = {
     val totalImportances = new OpenHashMap[Int, Double]()
     trees.foreach { tree =>
       // Aggregate feature importance vector for this tree
@@ -199,7 +203,7 @@ private[ml] object TreeEnsembleModel {
    *                     If -1, then numFeatures is set based on the max feature index in all trees.
    * @return  Feature importance values, of length numFeatures.
    */
-  def featureImportances(tree: DecisionTreeModel, numFeatures: Int): Vector = {
+  def featureImportances[M <: DecisionTreeModel : ClassTag](tree: M, numFeatures: Int): Vector = {
     featureImportances(Array(tree), numFeatures)
   }
 
@@ -386,7 +390,7 @@ private[ml] object EnsembleModelReadWrite {
    * @param path  Path to which to save the ensemble model.
    * @param extraMetadata  Metadata such as numFeatures, numClasses, numTrees.
    */
-  def saveImpl[M <: Params with TreeEnsembleModel](
+  def saveImpl[M <: Params with TreeEnsembleModel[_ <: DecisionTreeModel]](
       instance: M,
       path: String,
       sql: SQLContext,
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala
index bd5bd17147..b650a9f092 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/TreeTests.scala
@@ -131,7 +131,7 @@ private[ml] object TreeTests extends SparkFunSuite {
    * Check if the two models are exactly the same.
    * If the models are not equal, this throws an exception.
    */
-  def checkEqual(a: TreeEnsembleModel, b: TreeEnsembleModel): Unit = {
+  def checkEqual[M <: DecisionTreeModel](a: TreeEnsembleModel[M], b: TreeEnsembleModel[M]): Unit = {
     try {
       a.trees.zip(b.trees).foreach { case (treeA, treeB) =>
         TreeTests.checkEqual(treeA, treeB)
-- 
cgit v1.2.3


From d35690158810465809679ef39548e1400b38d448 Mon Sep 17 00:00:00 2001
From: Shally Sangal <shallysangal@gmail.com>
Date: Tue, 5 Apr 2016 10:41:59 -0700
Subject: [SPARK-14284][ML] KMeansSummary deprecating size; adding clusterSizes

## What changes were proposed in this pull request?

KMeansSummary class : deprecated size and added clusterSizes

Author: Shally Sangal <shallysangal@gmail.com>

Closes #12084 from shallys/master.
---
 mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala | 3 ++-
 mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala   | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
index 38428826a8..a8beef8b12 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
@@ -299,7 +299,8 @@ class KMeansSummary private[clustering] (
    * Size of each cluster.
    */
   @Since("2.0.0")
-  lazy val size: Array[Int] = cluster.rdd.map {
+  lazy val clusterSizes: Array[Int] = cluster.rdd.map {
     case Row(clusterIdx: Int) => (clusterIdx, 1)
   }.reduceByKey(_ + _).collect().sortBy(_._1).map(_._2)
+
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
index d3a0df4063..ed735a4ea3 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
@@ -37,7 +37,7 @@ private[r] class KMeansWrapper private (
 
   lazy val k: Int = kMeansModel.getK
 
-  lazy val size: Array[Int] = kMeansModel.summary.size
+  lazy val size: Array[Int] = kMeansModel.summary.clusterSizes
 
   lazy val cluster: DataFrame = kMeansModel.summary.cluster
 
-- 
cgit v1.2.3


From db0b06c6ea7412266158b1c710bdc8ca30e26430 Mon Sep 17 00:00:00 2001
From: Xusen Yin <yinxusen@gmail.com>
Date: Wed, 6 Apr 2016 11:24:11 -0700
Subject: [SPARK-13786][ML][PYSPARK] Add save/load for pyspark.ml.tuning

## What changes were proposed in this pull request?

https://issues.apache.org/jira/browse/SPARK-13786

Add save/load for Python CrossValidator/Model and TrainValidationSplit/Model.

## How was this patch tested?

Test with Python doctest.

Author: Xusen Yin <yinxusen@gmail.com>

Closes #12020 from yinxusen/SPARK-13786.
---
 .../scala/org/apache/spark/ml/param/params.scala   |  11 +
 .../apache/spark/ml/tuning/CrossValidator.scala    |   9 +
 .../spark/ml/tuning/TrainValidationSplit.scala     |   9 +
 python/pyspark/ml/tests.py                         |  56 ++-
 python/pyspark/ml/tuning.py                        | 407 +++++++++++++++------
 python/pyspark/ml/wrapper.py                       |  23 ++
 6 files changed, 404 insertions(+), 111 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/param/params.scala b/mllib/src/main/scala/org/apache/spark/ml/param/params.scala
index d7837b6730..c368aadd23 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/param/params.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/param/params.scala
@@ -18,6 +18,7 @@
 package org.apache.spark.ml.param
 
 import java.lang.reflect.Modifier
+import java.util.{List => JList}
 import java.util.NoSuchElementException
 
 import scala.annotation.varargs
@@ -833,6 +834,11 @@ final class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any])
     this
   }
 
+  /** Put param pairs with a [[java.util.List]] of values for Python. */
+  private[ml] def put(paramPairs: JList[ParamPair[_]]): this.type = {
+    put(paramPairs.asScala: _*)
+  }
+
   /**
    * Optionally returns the value associated with a param.
    */
@@ -932,6 +938,11 @@ final class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any])
     }
   }
 
+  /** Java-friendly method for Python API */
+  private[ml] def toList: java.util.List[ParamPair[_]] = {
+    this.toSeq.asJava
+  }
+
   /**
    * Number of param pairs in this map.
    */
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
index 040b0093b9..4d9d4d472e 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
@@ -17,6 +17,10 @@
 
 package org.apache.spark.ml.tuning
 
+import java.util.{List => JList}
+
+import scala.collection.JavaConverters._
+
 import com.github.fommil.netlib.F2jBLAS
 import org.apache.hadoop.fs.Path
 import org.json4s.DefaultFormats
@@ -200,6 +204,11 @@ class CrossValidatorModel private[ml] (
     @Since("1.5.0") val avgMetrics: Array[Double])
   extends Model[CrossValidatorModel] with CrossValidatorParams with MLWritable {
 
+  /** A Python-friendly auxiliary constructor. */
+  private[ml] def this(uid: String, bestModel: Model[_], avgMetrics: JList[Double]) = {
+    this(uid, bestModel, avgMetrics.asScala.toArray)
+  }
+
   @Since("1.4.0")
   override def transform(dataset: DataFrame): DataFrame = {
     transformSchema(dataset.schema, logging = true)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
index 07330bb6b0..0f2179c2a1 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
@@ -17,6 +17,10 @@
 
 package org.apache.spark.ml.tuning
 
+import java.util.{List => JList}
+
+import scala.collection.JavaConverters._
+
 import org.apache.hadoop.fs.Path
 import org.json4s.DefaultFormats
 
@@ -198,6 +202,11 @@ class TrainValidationSplitModel private[ml] (
     @Since("1.5.0") val validationMetrics: Array[Double])
   extends Model[TrainValidationSplitModel] with TrainValidationSplitParams with MLWritable {
 
+  /** A Python-friendly auxiliary constructor. */
+  private[ml] def this(uid: String, bestModel: Model[_], validationMetrics: JList[Double]) = {
+    this(uid, bestModel, validationMetrics.asScala.toArray)
+  }
+
   @Since("1.5.0")
   override def transform(dataset: DataFrame): DataFrame = {
     transformSchema(dataset.schema, logging = true)
diff --git a/python/pyspark/ml/tests.py b/python/pyspark/ml/tests.py
index f6159b2c95..e3f873e3a7 100644
--- a/python/pyspark/ml/tests.py
+++ b/python/pyspark/ml/tests.py
@@ -44,7 +44,7 @@ import numpy as np
 from pyspark.ml import Estimator, Model, Pipeline, PipelineModel, Transformer
 from pyspark.ml.classification import LogisticRegression, DecisionTreeClassifier
 from pyspark.ml.clustering import KMeans
-from pyspark.ml.evaluation import RegressionEvaluator
+from pyspark.ml.evaluation import BinaryClassificationEvaluator, RegressionEvaluator
 from pyspark.ml.feature import *
 from pyspark.ml.param import Param, Params, TypeConverters
 from pyspark.ml.param.shared import HasMaxIter, HasInputCol, HasSeed
@@ -53,7 +53,7 @@ from pyspark.ml.tuning import *
 from pyspark.ml.util import keyword_only
 from pyspark.ml.util import MLWritable, MLWriter
 from pyspark.ml.wrapper import JavaWrapper
-from pyspark.mllib.linalg import DenseVector, SparseVector
+from pyspark.mllib.linalg import Vectors, DenseVector, SparseVector
 from pyspark.sql import DataFrame, SQLContext, Row
 from pyspark.sql.functions import rand
 from pyspark.tests import ReusedPySparkTestCase as PySparkTestCase
@@ -479,6 +479,32 @@ class CrossValidatorTests(PySparkTestCase):
                          "Best model should have zero induced error")
         self.assertEqual(1.0, bestModelMetric, "Best model has R-squared of 1")
 
+    def test_save_load(self):
+        temp_path = tempfile.mkdtemp()
+        sqlContext = SQLContext(self.sc)
+        dataset = sqlContext.createDataFrame(
+            [(Vectors.dense([0.0]), 0.0),
+             (Vectors.dense([0.4]), 1.0),
+             (Vectors.dense([0.5]), 0.0),
+             (Vectors.dense([0.6]), 1.0),
+             (Vectors.dense([1.0]), 1.0)] * 10,
+            ["features", "label"])
+        lr = LogisticRegression()
+        grid = ParamGridBuilder().addGrid(lr.maxIter, [0, 1]).build()
+        evaluator = BinaryClassificationEvaluator()
+        cv = CrossValidator(estimator=lr, estimatorParamMaps=grid, evaluator=evaluator)
+        cvModel = cv.fit(dataset)
+        cvPath = temp_path + "/cv"
+        cv.save(cvPath)
+        loadedCV = CrossValidator.load(cvPath)
+        self.assertEqual(loadedCV.getEstimator().uid, cv.getEstimator().uid)
+        self.assertEqual(loadedCV.getEvaluator().uid, cv.getEvaluator().uid)
+        self.assertEqual(loadedCV.getEstimatorParamMaps(), cv.getEstimatorParamMaps())
+        cvModelPath = temp_path + "/cvModel"
+        cvModel.save(cvModelPath)
+        loadedModel = CrossValidatorModel.load(cvModelPath)
+        self.assertEqual(loadedModel.bestModel.uid, cvModel.bestModel.uid)
+
 
 class TrainValidationSplitTests(PySparkTestCase):
 
@@ -530,6 +556,32 @@ class TrainValidationSplitTests(PySparkTestCase):
                          "Best model should have zero induced error")
         self.assertEqual(1.0, bestModelMetric, "Best model has R-squared of 1")
 
+    def test_save_load(self):
+        temp_path = tempfile.mkdtemp()
+        sqlContext = SQLContext(self.sc)
+        dataset = sqlContext.createDataFrame(
+            [(Vectors.dense([0.0]), 0.0),
+             (Vectors.dense([0.4]), 1.0),
+             (Vectors.dense([0.5]), 0.0),
+             (Vectors.dense([0.6]), 1.0),
+             (Vectors.dense([1.0]), 1.0)] * 10,
+            ["features", "label"])
+        lr = LogisticRegression()
+        grid = ParamGridBuilder().addGrid(lr.maxIter, [0, 1]).build()
+        evaluator = BinaryClassificationEvaluator()
+        tvs = TrainValidationSplit(estimator=lr, estimatorParamMaps=grid, evaluator=evaluator)
+        tvsModel = tvs.fit(dataset)
+        tvsPath = temp_path + "/tvs"
+        tvs.save(tvsPath)
+        loadedTvs = TrainValidationSplit.load(tvsPath)
+        self.assertEqual(loadedTvs.getEstimator().uid, tvs.getEstimator().uid)
+        self.assertEqual(loadedTvs.getEvaluator().uid, tvs.getEvaluator().uid)
+        self.assertEqual(loadedTvs.getEstimatorParamMaps(), tvs.getEstimatorParamMaps())
+        tvsModelPath = temp_path + "/tvsModel"
+        tvsModel.save(tvsModelPath)
+        loadedModel = TrainValidationSplitModel.load(tvsModelPath)
+        self.assertEqual(loadedModel.bestModel.uid, tvsModel.bestModel.uid)
+
 
 class PersistenceTest(PySparkTestCase):
 
diff --git a/python/pyspark/ml/tuning.py b/python/pyspark/ml/tuning.py
index a528d22e18..da00f317b3 100644
--- a/python/pyspark/ml/tuning.py
+++ b/python/pyspark/ml/tuning.py
@@ -18,12 +18,15 @@
 import itertools
 import numpy as np
 
+from pyspark import SparkContext
 from pyspark import since
 from pyspark.ml import Estimator, Model
 from pyspark.ml.param import Params, Param, TypeConverters
 from pyspark.ml.param.shared import HasSeed
-from pyspark.ml.util import keyword_only
+from pyspark.ml.util import keyword_only, JavaMLWriter, JavaMLReader, MLReadable, MLWritable
+from pyspark.ml.wrapper import JavaWrapper
 from pyspark.sql.functions import rand
+from pyspark.mllib.common import inherit_doc, _py2java
 
 __all__ = ['ParamGridBuilder', 'CrossValidator', 'CrossValidatorModel', 'TrainValidationSplit',
            'TrainValidationSplitModel']
@@ -91,7 +94,84 @@ class ParamGridBuilder(object):
         return [dict(zip(keys, prod)) for prod in itertools.product(*grid_values)]
 
 
-class CrossValidator(Estimator, HasSeed):
+class ValidatorParams(HasSeed):
+    """
+    Common params for TrainValidationSplit and CrossValidator.
+    """
+
+    estimator = Param(Params._dummy(), "estimator", "estimator to be cross-validated")
+    estimatorParamMaps = Param(Params._dummy(), "estimatorParamMaps", "estimator param maps")
+    evaluator = Param(
+        Params._dummy(), "evaluator",
+        "evaluator used to select hyper-parameters that maximize the validator metric")
+
+    def setEstimator(self, value):
+        """
+        Sets the value of :py:attr:`estimator`.
+        """
+        return self._set(estimator=value)
+
+    def getEstimator(self):
+        """
+        Gets the value of estimator or its default value.
+        """
+        return self.getOrDefault(self.estimator)
+
+    def setEstimatorParamMaps(self, value):
+        """
+        Sets the value of :py:attr:`estimatorParamMaps`.
+        """
+        return self._set(estimatorParamMaps=value)
+
+    def getEstimatorParamMaps(self):
+        """
+        Gets the value of estimatorParamMaps or its default value.
+        """
+        return self.getOrDefault(self.estimatorParamMaps)
+
+    def setEvaluator(self, value):
+        """
+        Sets the value of :py:attr:`evaluator`.
+        """
+        return self._set(evaluator=value)
+
+    def getEvaluator(self):
+        """
+        Gets the value of evaluator or its default value.
+        """
+        return self.getOrDefault(self.evaluator)
+
+    @classmethod
+    def _from_java_impl(cls, java_stage):
+        """
+        Return Python estimator, estimatorParamMaps, and evaluator from a Java ValidatorParams.
+        """
+
+        # Load information from java_stage to the instance.
+        estimator = JavaWrapper._from_java(java_stage.getEstimator())
+        evaluator = JavaWrapper._from_java(java_stage.getEvaluator())
+        epms = [estimator._transfer_param_map_from_java(epm)
+                for epm in java_stage.getEstimatorParamMaps()]
+        return estimator, epms, evaluator
+
+    def _to_java_impl(self):
+        """
+        Return Java estimator, estimatorParamMaps, and evaluator from this Python instance.
+        """
+
+        gateway = SparkContext._gateway
+        cls = SparkContext._jvm.org.apache.spark.ml.param.ParamMap
+
+        java_epms = gateway.new_array(cls, len(self.getEstimatorParamMaps()))
+        for idx, epm in enumerate(self.getEstimatorParamMaps()):
+            java_epms[idx] = self.getEstimator()._transfer_param_map_to_java(epm)
+
+        java_estimator = self.getEstimator()._to_java()
+        java_evaluator = self.getEvaluator()._to_java()
+        return java_estimator, java_epms, java_evaluator
+
+
+class CrossValidator(Estimator, ValidatorParams, MLReadable, MLWritable):
     """
     K-fold cross validation.
 
@@ -116,11 +196,6 @@ class CrossValidator(Estimator, HasSeed):
     .. versionadded:: 1.4.0
     """
 
-    estimator = Param(Params._dummy(), "estimator", "estimator to be cross-validated")
-    estimatorParamMaps = Param(Params._dummy(), "estimatorParamMaps", "estimator param maps")
-    evaluator = Param(
-        Params._dummy(), "evaluator",
-        "evaluator used to select hyper-parameters that maximize the cross-validated metric")
     numFolds = Param(Params._dummy(), "numFolds", "number of folds for cross validation",
                      typeConverter=TypeConverters.toInt)
 
@@ -148,51 +223,6 @@ class CrossValidator(Estimator, HasSeed):
         kwargs = self.setParams._input_kwargs
         return self._set(**kwargs)
 
-    @since("1.4.0")
-    def setEstimator(self, value):
-        """
-        Sets the value of :py:attr:`estimator`.
-        """
-        self._paramMap[self.estimator] = value
-        return self
-
-    @since("1.4.0")
-    def getEstimator(self):
-        """
-        Gets the value of estimator or its default value.
-        """
-        return self.getOrDefault(self.estimator)
-
-    @since("1.4.0")
-    def setEstimatorParamMaps(self, value):
-        """
-        Sets the value of :py:attr:`estimatorParamMaps`.
-        """
-        self._paramMap[self.estimatorParamMaps] = value
-        return self
-
-    @since("1.4.0")
-    def getEstimatorParamMaps(self):
-        """
-        Gets the value of estimatorParamMaps or its default value.
-        """
-        return self.getOrDefault(self.estimatorParamMaps)
-
-    @since("1.4.0")
-    def setEvaluator(self, value):
-        """
-        Sets the value of :py:attr:`evaluator`.
-        """
-        self._paramMap[self.evaluator] = value
-        return self
-
-    @since("1.4.0")
-    def getEvaluator(self):
-        """
-        Gets the value of evaluator or its default value.
-        """
-        return self.getOrDefault(self.evaluator)
-
     @since("1.4.0")
     def setNumFolds(self, value):
         """
@@ -236,7 +266,7 @@ class CrossValidator(Estimator, HasSeed):
         else:
             bestIndex = np.argmin(metrics)
         bestModel = est.fit(dataset, epm[bestIndex])
-        return CrossValidatorModel(bestModel)
+        return self._copyValues(CrossValidatorModel(bestModel))
 
     @since("1.4.0")
     def copy(self, extra=None):
@@ -258,8 +288,58 @@ class CrossValidator(Estimator, HasSeed):
             newCV.setEvaluator(self.getEvaluator().copy(extra))
         return newCV
 
+    @since("2.0.0")
+    def write(self):
+        """Returns an MLWriter instance for this ML instance."""
+        return JavaMLWriter(self)
+
+    @since("2.0.0")
+    def save(self, path):
+        """Save this ML instance to the given path, a shortcut of `write().save(path)`."""
+        self.write().save(path)
+
+    @classmethod
+    @since("2.0.0")
+    def read(cls):
+        """Returns an MLReader instance for this class."""
+        return JavaMLReader(cls)
+
+    @classmethod
+    def _from_java(cls, java_stage):
+        """
+        Given a Java CrossValidator, create and return a Python wrapper of it.
+        Used for ML persistence.
+        """
+
+        estimator, epms, evaluator = super(CrossValidator, cls)._from_java_impl(java_stage)
+        numFolds = java_stage.getNumFolds()
+        seed = java_stage.getSeed()
+        # Create a new instance of this stage.
+        py_stage = cls(estimator=estimator, estimatorParamMaps=epms, evaluator=evaluator,
+                       numFolds=numFolds, seed=seed)
+        py_stage._resetUid(java_stage.uid())
+        return py_stage
 
-class CrossValidatorModel(Model):
+    def _to_java(self):
+        """
+        Transfer this instance to a Java CrossValidator. Used for ML persistence.
+
+        :return: Java object equivalent to this instance.
+        """
+
+        estimator, epms, evaluator = super(CrossValidator, self)._to_java_impl()
+
+        _java_obj = JavaWrapper._new_java_obj("org.apache.spark.ml.tuning.CrossValidator", self.uid)
+        _java_obj.setEstimatorParamMaps(epms)
+        _java_obj.setEvaluator(evaluator)
+        _java_obj.setEstimator(estimator)
+        _java_obj.setSeed(self.getSeed())
+        _java_obj.setNumFolds(self.getNumFolds())
+
+        return _java_obj
+
+
+class CrossValidatorModel(Model, ValidatorParams, MLReadable, MLWritable):
     """
     Model from k-fold cross validation.
 
@@ -289,8 +369,60 @@ class CrossValidatorModel(Model):
             extra = dict()
         return CrossValidatorModel(self.bestModel.copy(extra))
 
+    @since("2.0.0")
+    def write(self):
+        """Returns an MLWriter instance for this ML instance."""
+        return JavaMLWriter(self)
+
+    @since("2.0.0")
+    def save(self, path):
+        """Save this ML instance to the given path, a shortcut of `write().save(path)`."""
+        self.write().save(path)
 
-class TrainValidationSplit(Estimator, HasSeed):
+    @classmethod
+    @since("2.0.0")
+    def read(cls):
+        """Returns an MLReader instance for this class."""
+        return JavaMLReader(cls)
+
+    @classmethod
+    def _from_java(cls, java_stage):
+        """
+        Given a Java CrossValidatorModel, create and return a Python wrapper of it.
+        Used for ML persistence.
+        """
+
+        # Load information from java_stage to the instance.
+        bestModel = JavaWrapper._from_java(java_stage.bestModel())
+        estimator, epms, evaluator = super(CrossValidatorModel, cls)._from_java_impl(java_stage)
+        # Create a new instance of this stage.
+        py_stage = cls(bestModel=bestModel)\
+            .setEstimator(estimator).setEstimatorParamMaps(epms).setEvaluator(evaluator)
+        py_stage._resetUid(java_stage.uid())
+        return py_stage
+
+    def _to_java(self):
+        """
+        Transfer this instance to a Java CrossValidatorModel. Used for ML persistence.
+
+        :return: Java object equivalent to this instance.
+        """
+
+        sc = SparkContext._active_spark_context
+
+        _java_obj = JavaWrapper._new_java_obj("org.apache.spark.ml.tuning.CrossValidatorModel",
+                                              self.uid,
+                                              self.bestModel._to_java(),
+                                              _py2java(sc, []))
+        estimator, epms, evaluator = super(CrossValidatorModel, self)._to_java_impl()
+
+        _java_obj.set("evaluator", evaluator)
+        _java_obj.set("estimator", estimator)
+        _java_obj.set("estimatorParamMaps", epms)
+        return _java_obj
+
+
+class TrainValidationSplit(Estimator, ValidatorParams, MLReadable, MLWritable):
     """
     Train-Validation-Split.
 
@@ -315,11 +447,6 @@ class TrainValidationSplit(Estimator, HasSeed):
     .. versionadded:: 2.0.0
     """
 
-    estimator = Param(Params._dummy(), "estimator", "estimator to be tested")
-    estimatorParamMaps = Param(Params._dummy(), "estimatorParamMaps", "estimator param maps")
-    evaluator = Param(
-        Params._dummy(), "evaluator",
-        "evaluator used to select hyper-parameters that maximize the validated metric")
     trainRatio = Param(Params._dummy(), "trainRatio", "Param for ratio between train and\
      validation data. Must be between 0 and 1.")
 
@@ -347,51 +474,6 @@ class TrainValidationSplit(Estimator, HasSeed):
         kwargs = self.setParams._input_kwargs
         return self._set(**kwargs)
 
-    @since("2.0.0")
-    def setEstimator(self, value):
-        """
-        Sets the value of :py:attr:`estimator`.
-        """
-        self._paramMap[self.estimator] = value
-        return self
-
-    @since("2.0.0")
-    def getEstimator(self):
-        """
-        Gets the value of estimator or its default value.
-        """
-        return self.getOrDefault(self.estimator)
-
-    @since("2.0.0")
-    def setEstimatorParamMaps(self, value):
-        """
-        Sets the value of :py:attr:`estimatorParamMaps`.
-        """
-        self._paramMap[self.estimatorParamMaps] = value
-        return self
-
-    @since("2.0.0")
-    def getEstimatorParamMaps(self):
-        """
-        Gets the value of estimatorParamMaps or its default value.
-        """
-        return self.getOrDefault(self.estimatorParamMaps)
-
-    @since("2.0.0")
-    def setEvaluator(self, value):
-        """
-        Sets the value of :py:attr:`evaluator`.
-        """
-        self._paramMap[self.evaluator] = value
-        return self
-
-    @since("2.0.0")
-    def getEvaluator(self):
-        """
-        Gets the value of evaluator or its default value.
-        """
-        return self.getOrDefault(self.evaluator)
-
     @since("2.0.0")
     def setTrainRatio(self, value):
         """
@@ -429,7 +511,7 @@ class TrainValidationSplit(Estimator, HasSeed):
         else:
             bestIndex = np.argmin(metrics)
         bestModel = est.fit(dataset, epm[bestIndex])
-        return TrainValidationSplitModel(bestModel)
+        return self._copyValues(TrainValidationSplitModel(bestModel))
 
     @since("2.0.0")
     def copy(self, extra=None):
@@ -451,8 +533,59 @@ class TrainValidationSplit(Estimator, HasSeed):
             newTVS.setEvaluator(self.getEvaluator().copy(extra))
         return newTVS
 
+    @since("2.0.0")
+    def write(self):
+        """Returns an MLWriter instance for this ML instance."""
+        return JavaMLWriter(self)
+
+    @since("2.0.0")
+    def save(self, path):
+        """Save this ML instance to the given path, a shortcut of `write().save(path)`."""
+        self.write().save(path)
+
+    @classmethod
+    @since("2.0.0")
+    def read(cls):
+        """Returns an MLReader instance for this class."""
+        return JavaMLReader(cls)
+
+    @classmethod
+    def _from_java(cls, java_stage):
+        """
+        Given a Java TrainValidationSplit, create and return a Python wrapper of it.
+        Used for ML persistence.
+        """
+
+        estimator, epms, evaluator = super(TrainValidationSplit, cls)._from_java_impl(java_stage)
+        trainRatio = java_stage.getTrainRatio()
+        seed = java_stage.getSeed()
+        # Create a new instance of this stage.
+        py_stage = cls(estimator=estimator, estimatorParamMaps=epms, evaluator=evaluator,
+                       trainRatio=trainRatio, seed=seed)
+        py_stage._resetUid(java_stage.uid())
+        return py_stage
+
+    def _to_java(self):
+        """
+        Transfer this instance to a Java TrainValidationSplit. Used for ML persistence.
 
-class TrainValidationSplitModel(Model):
+        :return: Java object equivalent to this instance.
+        """
+
+        estimator, epms, evaluator = super(TrainValidationSplit, self)._to_java_impl()
+
+        _java_obj = JavaWrapper._new_java_obj("org.apache.spark.ml.tuning.TrainValidationSplit",
+                                              self.uid)
+        _java_obj.setEstimatorParamMaps(epms)
+        _java_obj.setEvaluator(evaluator)
+        _java_obj.setEstimator(estimator)
+        _java_obj.setTrainRatio(self.getTrainRatio())
+        _java_obj.setSeed(self.getSeed())
+
+        return _java_obj
+
+
+class TrainValidationSplitModel(Model, ValidatorParams, MLReadable, MLWritable):
     """
     Model from train validation split.
     """
@@ -480,19 +613,75 @@ class TrainValidationSplitModel(Model):
             extra = dict()
         return TrainValidationSplitModel(self.bestModel.copy(extra))
 
+    @since("2.0.0")
+    def write(self):
+        """Returns an MLWriter instance for this ML instance."""
+        return JavaMLWriter(self)
+
+    @since("2.0.0")
+    def save(self, path):
+        """Save this ML instance to the given path, a shortcut of `write().save(path)`."""
+        self.write().save(path)
+
+    @classmethod
+    @since("2.0.0")
+    def read(cls):
+        """Returns an MLReader instance for this class."""
+        return JavaMLReader(cls)
+
+    @classmethod
+    def _from_java(cls, java_stage):
+        """
+        Given a Java TrainValidationSplitModel, create and return a Python wrapper of it.
+        Used for ML persistence.
+        """
+
+        # Load information from java_stage to the instance.
+        bestModel = JavaWrapper._from_java(java_stage.bestModel())
+        estimator, epms, evaluator = \
+            super(TrainValidationSplitModel, cls)._from_java_impl(java_stage)
+        # Create a new instance of this stage.
+        py_stage = cls(bestModel=bestModel)\
+            .setEstimator(estimator).setEstimatorParamMaps(epms).setEvaluator(evaluator)
+        py_stage._resetUid(java_stage.uid())
+        return py_stage
+
+    def _to_java(self):
+        """
+        Transfer this instance to a Java TrainValidationSplitModel. Used for ML persistence.
+
+        :return: Java object equivalent to this instance.
+        """
+
+        sc = SparkContext._active_spark_context
+
+        _java_obj = JavaWrapper._new_java_obj(
+            "org.apache.spark.ml.tuning.TrainValidationSplitModel",
+            self.uid,
+            self.bestModel._to_java(),
+            _py2java(sc, []))
+        estimator, epms, evaluator = super(TrainValidationSplitModel, self)._to_java_impl()
+
+        _java_obj.set("evaluator", evaluator)
+        _java_obj.set("estimator", estimator)
+        _java_obj.set("estimatorParamMaps", epms)
+        return _java_obj
+
+
 if __name__ == "__main__":
     import doctest
+
     from pyspark.context import SparkContext
     from pyspark.sql import SQLContext
     globs = globals().copy()
+
     # The small batch size here ensures that we see multiple batches,
     # even in these small test examples:
     sc = SparkContext("local[2]", "ml.tuning tests")
     sqlContext = SQLContext(sc)
     globs['sc'] = sc
     globs['sqlContext'] = sqlContext
-    (failure_count, test_count) = doctest.testmod(
-        globs=globs, optionflags=doctest.ELLIPSIS)
+    (failure_count, test_count) = doctest.testmod(globs=globs, optionflags=doctest.ELLIPSIS)
     sc.stop()
     if failure_count:
         exit(-1)
diff --git a/python/pyspark/ml/wrapper.py b/python/pyspark/ml/wrapper.py
index 35b0eba926..ca93bf7d7d 100644
--- a/python/pyspark/ml/wrapper.py
+++ b/python/pyspark/ml/wrapper.py
@@ -76,6 +76,17 @@ class JavaWrapper(Params):
                 pair = self._make_java_param_pair(param, paramMap[param])
                 self._java_obj.set(pair)
 
+    def _transfer_param_map_to_java(self, pyParamMap):
+        """
+        Transforms a Python ParamMap into a Java ParamMap.
+        """
+        paramMap = JavaWrapper._new_java_obj("org.apache.spark.ml.param.ParamMap")
+        for param in self.params:
+            if param in pyParamMap:
+                pair = self._make_java_param_pair(param, pyParamMap[param])
+                paramMap.put([pair])
+        return paramMap
+
     def _transfer_params_from_java(self):
         """
         Transforms the embedded params from the companion Java object.
@@ -88,6 +99,18 @@ class JavaWrapper(Params):
                     value = _java2py(sc, self._java_obj.getOrDefault(java_param))
                     self._paramMap[param] = value
 
+    def _transfer_param_map_from_java(self, javaParamMap):
+        """
+        Transforms a Java ParamMap into a Python ParamMap.
+        """
+        sc = SparkContext._active_spark_context
+        paramMap = dict()
+        for pair in javaParamMap.toList():
+            param = pair.param()
+            if self.hasParam(str(param.name())):
+                paramMap[self.getParam(param.name())] = _java2py(sc, pair.value())
+        return paramMap
+
     @staticmethod
     def _empty_java_param_map():
         """
-- 
cgit v1.2.3


From 8cffcb60deb82d04a5c6e144ec9927f6f7addc8b Mon Sep 17 00:00:00 2001
From: Yuhao Yang <hhbyyh@gmail.com>
Date: Wed, 6 Apr 2016 11:36:26 -0700
Subject: [SPARK-14322][MLLIB] Use treeAggregate instead of reduce in
 OnlineLDAOptimizer

## What changes were proposed in this pull request?
jira: https://issues.apache.org/jira/browse/SPARK-14322

OnlineLDAOptimizer uses RDD.reduce in two places where it could use treeAggregate. This can cause scalability issues. This should be an easy fix.
This is also a bug since it modifies the first argument to reduce, so we should use aggregate or treeAggregate.
See this line: https://github.com/apache/spark/blob/f12f11e578169b47e3f8b18b299948c0670ba585/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala#L452
and a few lines below it.

## How was this patch tested?
unit tests

Author: Yuhao Yang <hhbyyh@gmail.com>

Closes #12106 from hhbyyh/ldaTreeReduce.
---
 .../main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala  | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala
index 7491ab0d51..2b404a8651 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala
@@ -451,10 +451,11 @@ final class OnlineLDAOptimizer extends LDAOptimizer {
       }
       Iterator((stat, gammaPart))
     }
-    val statsSum: BDM[Double] = stats.map(_._1).reduce(_ += _)
+    val statsSum: BDM[Double] = stats.map(_._1).treeAggregate(BDM.zeros[Double](k, vocabSize))(
+      _ += _, _ += _)
     expElogbetaBc.unpersist()
     val gammat: BDM[Double] = breeze.linalg.DenseMatrix.vertcat(
-      stats.map(_._2).reduce(_ ++ _).map(_.toDenseMatrix): _*)
+      stats.map(_._2).flatMap(list => list).collect().map(_.toDenseMatrix): _*)
     val batchResult = statsSum :* expElogbeta.t
 
     // Note that this is an optimization to avoid batch.count
-- 
cgit v1.2.3


From af73d9737874f7adaec3cd19ac889ab3badb8e2a Mon Sep 17 00:00:00 2001
From: Zheng RuiFeng <ruifengz@foxmail.com>
Date: Wed, 6 Apr 2016 11:45:16 -0700
Subject: [SPARK-13538][ML] Add GaussianMixture to ML

JIRA: https://issues.apache.org/jira/browse/SPARK-13538

## What changes were proposed in this pull request?

Add GaussianMixture and GaussianMixtureModel to ML package

## How was this patch tested?

unit tests and manual tests were done.
Local Scalastyle checks passed.

Author: Zheng RuiFeng <ruifengz@foxmail.com>
Author: Ruifeng Zheng <ruifengz@foxmail.com>
Author: Joseph K. Bradley <joseph@databricks.com>

Closes #11419 from zhengruifeng/mlgmm.
---
 .../spark/ml/clustering/GaussianMixture.scala      | 311 +++++++++++++++++++++
 .../spark/ml/clustering/GaussianMixtureSuite.scala | 133 +++++++++
 2 files changed, 444 insertions(+)
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala
 create mode 100644 mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala
new file mode 100644
index 0000000000..120bf3cf9d
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala
@@ -0,0 +1,311 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.clustering
+
+import org.apache.hadoop.fs.Path
+
+import org.apache.spark.annotation.{Experimental, Since}
+import org.apache.spark.ml.{Estimator, Model}
+import org.apache.spark.ml.param.{IntParam, ParamMap, Params}
+import org.apache.spark.ml.param.shared._
+import org.apache.spark.ml.util._
+import org.apache.spark.mllib.clustering.{GaussianMixture => MLlibGM, GaussianMixtureModel => MLlibGMModel}
+import org.apache.spark.mllib.linalg._
+import org.apache.spark.mllib.stat.distribution.MultivariateGaussian
+import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.functions.{col, udf}
+import org.apache.spark.sql.types.{IntegerType, StructType}
+
+
+/**
+ * Common params for GaussianMixture and GaussianMixtureModel
+ */
+private[clustering] trait GaussianMixtureParams extends Params with HasMaxIter with HasFeaturesCol
+  with HasSeed with HasPredictionCol with HasProbabilityCol with HasTol {
+
+  /**
+   * Set the number of clusters to create (k). Must be > 1. Default: 2.
+   * @group param
+   */
+  @Since("2.0.0")
+  final val k = new IntParam(this, "k", "number of clusters to create", (x: Int) => x > 1)
+
+  /** @group getParam */
+  @Since("2.0.0")
+  def getK: Int = $(k)
+
+  /**
+   * Validates and transforms the input schema.
+   * @param schema input schema
+   * @return output schema
+   */
+  protected def validateAndTransformSchema(schema: StructType): StructType = {
+    SchemaUtils.checkColumnType(schema, $(featuresCol), new VectorUDT)
+    SchemaUtils.appendColumn(schema, $(predictionCol), IntegerType)
+    SchemaUtils.appendColumn(schema, $(probabilityCol), new VectorUDT)
+  }
+}
+
+/**
+ * :: Experimental ::
+ * Model fitted by GaussianMixture.
+ * @param parentModel a model trained by spark.mllib.clustering.GaussianMixture.
+ */
+@Since("2.0.0")
+@Experimental
+class GaussianMixtureModel private[ml] (
+    @Since("2.0.0") override val uid: String,
+    private val parentModel: MLlibGMModel)
+  extends Model[GaussianMixtureModel] with GaussianMixtureParams with MLWritable {
+
+  @Since("2.0.0")
+  override def copy(extra: ParamMap): GaussianMixtureModel = {
+    val copied = new GaussianMixtureModel(uid, parentModel)
+    copyValues(copied, extra).setParent(this.parent)
+  }
+
+  @Since("2.0.0")
+  override def transform(dataset: DataFrame): DataFrame = {
+    val predUDF = udf((vector: Vector) => predict(vector))
+    val probUDF = udf((vector: Vector) => predictProbability(vector))
+    dataset.withColumn($(predictionCol), predUDF(col($(featuresCol))))
+      .withColumn($(probabilityCol), probUDF(col($(featuresCol))))
+  }
+
+  @Since("2.0.0")
+  override def transformSchema(schema: StructType): StructType = {
+    validateAndTransformSchema(schema)
+  }
+
+  private[clustering] def predict(features: Vector): Int = parentModel.predict(features)
+
+  private[clustering] def predictProbability(features: Vector): Vector = {
+    Vectors.dense(parentModel.predictSoft(features))
+  }
+
+  @Since("2.0.0")
+  def weights: Array[Double] = parentModel.weights
+
+  @Since("2.0.0")
+  def gaussians: Array[MultivariateGaussian] = parentModel.gaussians
+
+  @Since("2.0.0")
+  override def write: MLWriter = new GaussianMixtureModel.GaussianMixtureModelWriter(this)
+
+  private var trainingSummary: Option[GaussianMixtureSummary] = None
+
+  private[clustering] def setSummary(summary: GaussianMixtureSummary): this.type = {
+    this.trainingSummary = Some(summary)
+    this
+  }
+
+  /**
+   * Return true if there exists summary of model.
+   */
+  @Since("2.0.0")
+  def hasSummary: Boolean = trainingSummary.nonEmpty
+
+  /**
+   * Gets summary of model on training set. An exception is
+   * thrown if `trainingSummary == None`.
+   */
+  @Since("2.0.0")
+  def summary: GaussianMixtureSummary = trainingSummary.getOrElse {
+    throw new RuntimeException(
+      s"No training summary available for the ${this.getClass.getSimpleName}")
+  }
+}
+
+@Since("2.0.0")
+object GaussianMixtureModel extends MLReadable[GaussianMixtureModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[GaussianMixtureModel] = new GaussianMixtureModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): GaussianMixtureModel = super.load(path)
+
+  /** [[MLWriter]] instance for [[GaussianMixtureModel]] */
+  private[GaussianMixtureModel] class GaussianMixtureModelWriter(
+      instance: GaussianMixtureModel) extends MLWriter {
+
+    private case class Data(weights: Array[Double], mus: Array[Vector], sigmas: Array[Matrix])
+
+    override protected def saveImpl(path: String): Unit = {
+      // Save metadata and Params
+      DefaultParamsWriter.saveMetadata(instance, path, sc)
+      // Save model data: weights and gaussians
+      val weights = instance.weights
+      val gaussians = instance.gaussians
+      val mus = gaussians.map(_.mu)
+      val sigmas = gaussians.map(_.sigma)
+      val data = Data(weights, mus, sigmas)
+      val dataPath = new Path(path, "data").toString
+      sqlContext.createDataFrame(Seq(data)).repartition(1).write.parquet(dataPath)
+    }
+  }
+
+  private class GaussianMixtureModelReader extends MLReader[GaussianMixtureModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[GaussianMixtureModel].getName
+
+    override def load(path: String): GaussianMixtureModel = {
+      val metadata = DefaultParamsReader.loadMetadata(path, sc, className)
+
+      val dataPath = new Path(path, "data").toString
+      val row = sqlContext.read.parquet(dataPath).select("weights", "mus", "sigmas").head()
+      val weights = row.getSeq[Double](0).toArray
+      val mus = row.getSeq[Vector](1).toArray
+      val sigmas = row.getSeq[Matrix](2).toArray
+      require(mus.length == sigmas.length, "Length of Mu and Sigma array must match")
+      require(mus.length == weights.length, "Length of weight and Gaussian array must match")
+
+      val gaussians = (mus zip sigmas).map {
+        case (mu, sigma) =>
+          new MultivariateGaussian(mu, sigma)
+      }
+      val model = new GaussianMixtureModel(metadata.uid, new MLlibGMModel(weights, gaussians))
+
+      DefaultParamsReader.getAndSetParams(model, metadata)
+      model
+    }
+  }
+}
+
+/**
+ * :: Experimental ::
+ * GaussianMixture clustering.
+ */
+@Since("2.0.0")
+@Experimental
+class GaussianMixture @Since("2.0.0") (
+    @Since("2.0.0") override val uid: String)
+  extends Estimator[GaussianMixtureModel] with GaussianMixtureParams with DefaultParamsWritable {
+
+  setDefault(
+    k -> 2,
+    maxIter -> 100,
+    tol -> 0.01)
+
+  @Since("2.0.0")
+  override def copy(extra: ParamMap): GaussianMixture = defaultCopy(extra)
+
+  @Since("2.0.0")
+  def this() = this(Identifiable.randomUID("GaussianMixture"))
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setFeaturesCol(value: String): this.type = set(featuresCol, value)
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setPredictionCol(value: String): this.type = set(predictionCol, value)
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setProbabilityCol(value: String): this.type = set(probabilityCol, value)
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setK(value: Int): this.type = set(k, value)
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setMaxIter(value: Int): this.type = set(maxIter, value)
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setTol(value: Double): this.type = set(tol, value)
+
+  /** @group setParam */
+  @Since("2.0.0")
+  def setSeed(value: Long): this.type = set(seed, value)
+
+  @Since("2.0.0")
+  override def fit(dataset: DataFrame): GaussianMixtureModel = {
+    val rdd = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
+
+    val algo = new MLlibGM()
+      .setK($(k))
+      .setMaxIterations($(maxIter))
+      .setSeed($(seed))
+      .setConvergenceTol($(tol))
+    val parentModel = algo.run(rdd)
+    val model = copyValues(new GaussianMixtureModel(uid, parentModel).setParent(this))
+    val summary = new GaussianMixtureSummary(model.transform(dataset),
+      $(predictionCol), $(probabilityCol), $(featuresCol), $(k))
+    model.setSummary(summary)
+  }
+
+  @Since("2.0.0")
+  override def transformSchema(schema: StructType): StructType = {
+    validateAndTransformSchema(schema)
+  }
+}
+
+@Since("2.0.0")
+object GaussianMixture extends DefaultParamsReadable[GaussianMixture] {
+
+  @Since("2.0.0")
+  override def load(path: String): GaussianMixture = super.load(path)
+}
+
+/**
+ * :: Experimental ::
+ * Summary of GaussianMixture.
+ *
+ * @param predictions  [[DataFrame]] produced by [[GaussianMixtureModel.transform()]]
+ * @param predictionCol  Name for column of predicted clusters in `predictions`
+ * @param probabilityCol  Name for column of predicted probability of each cluster in `predictions`
+ * @param featuresCol  Name for column of features in `predictions`
+ * @param k  Number of clusters
+ */
+@Since("2.0.0")
+@Experimental
+class GaussianMixtureSummary private[clustering] (
+    @Since("2.0.0") @transient val predictions: DataFrame,
+    @Since("2.0.0") val predictionCol: String,
+    @Since("2.0.0") val probabilityCol: String,
+    @Since("2.0.0") val featuresCol: String,
+    @Since("2.0.0") val k: Int) extends Serializable {
+
+  /**
+   * Cluster centers of the transformed data.
+   */
+  @Since("2.0.0")
+  @transient lazy val cluster: DataFrame = predictions.select(predictionCol)
+
+  /**
+   * Probability of each cluster.
+   */
+  @Since("2.0.0")
+  @transient lazy val probability: DataFrame = predictions.select(probabilityCol)
+
+  /**
+   * Size of (number of data points in) each cluster.
+   */
+  @Since("2.0.0")
+  lazy val clusterSizes: Array[Long] = {
+    val sizes = Array.fill[Long](k)(0)
+    cluster.groupBy(predictionCol).count().select(predictionCol, "count").collect().foreach {
+      case Row(cluster: Int, count: Long) => sizes(cluster) = count
+    }
+    sizes
+  }
+}
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala
new file mode 100644
index 0000000000..8edd44e5f1
--- /dev/null
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala
@@ -0,0 +1,133 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.clustering
+
+import org.apache.spark.SparkFunSuite
+import org.apache.spark.ml.util.DefaultReadWriteTest
+import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.sql.DataFrame
+
+
+class GaussianMixtureSuite extends SparkFunSuite with MLlibTestSparkContext
+  with DefaultReadWriteTest {
+
+  final val k = 5
+  @transient var dataset: DataFrame = _
+
+  override def beforeAll(): Unit = {
+    super.beforeAll()
+
+    dataset = KMeansSuite.generateKMeansData(sqlContext, 50, 3, k)
+  }
+
+  test("default parameters") {
+    val gm = new GaussianMixture()
+
+    assert(gm.getK === 2)
+    assert(gm.getFeaturesCol === "features")
+    assert(gm.getPredictionCol === "prediction")
+    assert(gm.getMaxIter === 100)
+    assert(gm.getTol === 0.01)
+  }
+
+  test("set parameters") {
+    val gm = new GaussianMixture()
+      .setK(9)
+      .setFeaturesCol("test_feature")
+      .setPredictionCol("test_prediction")
+      .setProbabilityCol("test_probability")
+      .setMaxIter(33)
+      .setSeed(123)
+      .setTol(1e-3)
+
+    assert(gm.getK === 9)
+    assert(gm.getFeaturesCol === "test_feature")
+    assert(gm.getPredictionCol === "test_prediction")
+    assert(gm.getProbabilityCol === "test_probability")
+    assert(gm.getMaxIter === 33)
+    assert(gm.getSeed === 123)
+    assert(gm.getTol === 1e-3)
+  }
+
+  test("parameters validation") {
+    intercept[IllegalArgumentException] {
+      new GaussianMixture().setK(1)
+    }
+  }
+
+  test("fit, transform, and summary") {
+    val predictionColName = "gm_prediction"
+    val probabilityColName = "gm_probability"
+    val gm = new GaussianMixture().setK(k).setMaxIter(2).setPredictionCol(predictionColName)
+        .setProbabilityCol(probabilityColName).setSeed(1)
+    val model = gm.fit(dataset)
+    assert(model.hasParent)
+    assert(model.weights.length === k)
+    assert(model.gaussians.length === k)
+
+    val transformed = model.transform(dataset)
+    val expectedColumns = Array("features", predictionColName, probabilityColName)
+    expectedColumns.foreach { column =>
+      assert(transformed.columns.contains(column))
+    }
+
+    // Check validity of model summary
+    val numRows = dataset.count()
+    assert(model.hasSummary)
+    val summary: GaussianMixtureSummary = model.summary
+    assert(summary.predictionCol === predictionColName)
+    assert(summary.probabilityCol === probabilityColName)
+    assert(summary.featuresCol === "features")
+    assert(summary.predictions.count() === numRows)
+    for (c <- Array(predictionColName, probabilityColName, "features")) {
+      assert(summary.predictions.columns.contains(c))
+    }
+    assert(summary.cluster.columns === Array(predictionColName))
+    assert(summary.probability.columns === Array(probabilityColName))
+    val clusterSizes = summary.clusterSizes
+    assert(clusterSizes.length === k)
+    assert(clusterSizes.sum === numRows)
+    assert(clusterSizes.forall(_ >= 0))
+  }
+
+  test("read/write") {
+    def checkModelData(model: GaussianMixtureModel, model2: GaussianMixtureModel): Unit = {
+      assert(model.weights === model2.weights)
+      assert(model.gaussians.map(_.mu) === model2.gaussians.map(_.mu))
+      assert(model.gaussians.map(_.sigma) === model2.gaussians.map(_.sigma))
+    }
+    val gm = new GaussianMixture()
+    testEstimatorAndModelReadWrite(gm, dataset,
+      GaussianMixtureSuite.allParamSettings, checkModelData)
+  }
+}
+
+object GaussianMixtureSuite {
+  /**
+   * Mapping from all Params to valid settings which differ from the defaults.
+   * This is useful for tests which need to exercise all Params, such as save/load.
+   * This excludes input columns to simplify some tests.
+   */
+  val allParamSettings: Map[String, Any] = Map(
+    "predictionCol" -> "myPrediction",
+    "probabilityCol" -> "myProbability",
+    "k" -> 3,
+    "maxIter" -> 2,
+    "tol" -> 0.01
+  )
+}
-- 
cgit v1.2.3


From 9c6556c5f8ab013b36312db4bf02c4c6d965a535 Mon Sep 17 00:00:00 2001
From: Bryan Cutler <cutlerb@gmail.com>
Date: Wed, 6 Apr 2016 12:07:47 -0700
Subject: [SPARK-13430][PYSPARK][ML] Python API for training summaries of
 linear and logistic regression

## What changes were proposed in this pull request?

Adding Python API for training summaries of LogisticRegression and LinearRegression in PySpark ML.

## How was this patch tested?
Added unit tests to exercise the api calls for the summary classes.  Also, manually verified values are expected and match those from Scala directly.

Author: Bryan Cutler <cutlerb@gmail.com>

Closes #11621 from BryanCutler/pyspark-ml-summary-SPARK-13430.
---
 .../ml/classification/LogisticRegression.scala     |   8 +-
 .../spark/ml/regression/LinearRegression.scala     |  41 +++-
 project/MimaExcludes.scala                         |   3 +
 python/pyspark/ml/classification.py                | 218 +++++++++++++++++-
 python/pyspark/ml/regression.py                    | 245 ++++++++++++++++++++-
 python/pyspark/ml/tests.py                         |  87 +++++++-
 python/pyspark/ml/wrapper.py                       |  30 ++-
 7 files changed, 602 insertions(+), 30 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
index aeb94a6600..37182928cc 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
@@ -777,10 +777,10 @@ sealed trait LogisticRegressionSummary extends Serializable {
   /** Dataframe outputted by the model's `transform` method. */
   def predictions: DataFrame
 
-  /** Field in "predictions" which gives the calibrated probability of each instance as a vector. */
+  /** Field in "predictions" which gives the calibrated probability of each class as a vector. */
   def probabilityCol: String
 
-  /** Field in "predictions" which gives the true label of each instance. */
+  /** Field in "predictions" which gives the true label of each instance (if available). */
   def labelCol: String
 
   /** Field in "predictions" which gives the features of each instance as a vector. */
@@ -794,7 +794,7 @@ sealed trait LogisticRegressionSummary extends Serializable {
  *
  * @param predictions dataframe outputted by the model's `transform` method.
  * @param probabilityCol field in "predictions" which gives the calibrated probability of
- *                       each instance as a vector.
+ *                       each class as a vector.
  * @param labelCol field in "predictions" which gives the true label of each instance.
  * @param featuresCol field in "predictions" which gives the features of each instance as a vector.
  * @param objectiveHistory objective function (scaled loss + regularization) at each iteration.
@@ -818,7 +818,7 @@ class BinaryLogisticRegressionTrainingSummary private[classification] (
  *
  * @param predictions dataframe outputted by the model's `transform` method.
  * @param probabilityCol field in "predictions" which gives the calibrated probability of
- *                       each instance.
+ *                       each class as a vector.
  * @param labelCol field in "predictions" which gives the true label of each instance.
  * @param featuresCol field in "predictions" which gives the features of each instance as a vector.
  */
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
index 2633c06f40..9619e72a45 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
@@ -190,9 +190,9 @@ class LinearRegression @Since("1.3.0") (@Since("1.3.0") override val uid: String
         summaryModel.transform(dataset),
         predictionColName,
         $(labelCol),
+        $(featuresCol),
         summaryModel,
         model.diagInvAtWA.toArray,
-        $(featuresCol),
         Array(0D))
 
       return lrModel.setSummary(trainingSummary)
@@ -249,9 +249,9 @@ class LinearRegression @Since("1.3.0") (@Since("1.3.0") override val uid: String
           summaryModel.transform(dataset),
           predictionColName,
           $(labelCol),
+          $(featuresCol),
           model,
           Array(0D),
-          $(featuresCol),
           Array(0D))
         return copyValues(model.setSummary(trainingSummary))
       } else {
@@ -356,9 +356,9 @@ class LinearRegression @Since("1.3.0") (@Since("1.3.0") override val uid: String
       summaryModel.transform(dataset),
       predictionColName,
       $(labelCol),
+      $(featuresCol),
       model,
       Array(0D),
-      $(featuresCol),
       objectiveHistory)
     model.setSummary(trainingSummary)
   }
@@ -421,7 +421,7 @@ class LinearRegressionModel private[ml] (
     // Handle possible missing or invalid prediction columns
     val (summaryModel, predictionColName) = findSummaryModelAndPredictionCol()
     new LinearRegressionSummary(summaryModel.transform(dataset), predictionColName,
-      $(labelCol), summaryModel, Array(0D))
+      $(labelCol), $(featuresCol), summaryModel, Array(0D))
   }
 
   /**
@@ -511,7 +511,7 @@ object LinearRegressionModel extends MLReadable[LinearRegressionModel] {
 /**
  * :: Experimental ::
  * Linear regression training results. Currently, the training summary ignores the
- * training coefficients except for the objective trace.
+ * training weights except for the objective trace.
  *
  * @param predictions predictions outputted by the model's `transform` method.
  * @param objectiveHistory objective function (scaled loss + regularization) at each iteration.
@@ -522,13 +522,24 @@ class LinearRegressionTrainingSummary private[regression] (
     predictions: DataFrame,
     predictionCol: String,
     labelCol: String,
+    featuresCol: String,
     model: LinearRegressionModel,
     diagInvAtWA: Array[Double],
-    val featuresCol: String,
     val objectiveHistory: Array[Double])
-  extends LinearRegressionSummary(predictions, predictionCol, labelCol, model, diagInvAtWA) {
+  extends LinearRegressionSummary(
+    predictions,
+    predictionCol,
+    labelCol,
+    featuresCol,
+    model,
+    diagInvAtWA) {
 
-  /** Number of training iterations until termination */
+  /**
+   * Number of training iterations until termination
+   *
+   * This value is only available when using the "l-bfgs" solver.
+   * @see [[LinearRegression.solver]]
+   */
   @Since("1.5.0")
   val totalIterations = objectiveHistory.length
 
@@ -539,6 +550,10 @@ class LinearRegressionTrainingSummary private[regression] (
  * Linear regression results evaluated on a dataset.
  *
  * @param predictions predictions outputted by the model's `transform` method.
+ * @param predictionCol Field in "predictions" which gives the predicted value of the label at
+ *                      each instance.
+ * @param labelCol Field in "predictions" which gives the true label of each instance.
+ * @param featuresCol Field in "predictions" which gives the features of each instance as a vector.
  */
 @Since("1.5.0")
 @Experimental
@@ -546,6 +561,7 @@ class LinearRegressionSummary private[regression] (
     @transient val predictions: DataFrame,
     val predictionCol: String,
     val labelCol: String,
+    val featuresCol: String,
     val model: LinearRegressionModel,
     private val diagInvAtWA: Array[Double]) extends Serializable {
 
@@ -639,6 +655,9 @@ class LinearRegressionSummary private[regression] (
 
   /**
    * Standard error of estimated coefficients and intercept.
+   *
+   * This value is only available when using the "normal" solver.
+   * @see [[LinearRegression.solver]]
    */
   lazy val coefficientStandardErrors: Array[Double] = {
     if (diagInvAtWA.length == 1 && diagInvAtWA(0) == 0) {
@@ -660,6 +679,9 @@ class LinearRegressionSummary private[regression] (
 
   /**
    * T-statistic of estimated coefficients and intercept.
+   *
+   * This value is only available when using the "normal" solver.
+   * @see [[LinearRegression.solver]]
    */
   lazy val tValues: Array[Double] = {
     if (diagInvAtWA.length == 1 && diagInvAtWA(0) == 0) {
@@ -677,6 +699,9 @@ class LinearRegressionSummary private[regression] (
 
   /**
    * Two-sided p-value of estimated coefficients and intercept.
+   *
+   * This value is only available when using the "normal" solver.
+   * @see [[LinearRegression.solver]]
    */
   lazy val pValues: Array[Double] = {
     if (diagInvAtWA.length == 1 && diagInvAtWA(0) == 0) {
diff --git a/project/MimaExcludes.scala b/project/MimaExcludes.scala
index 9f245afd50..d916c49a6a 100644
--- a/project/MimaExcludes.scala
+++ b/project/MimaExcludes.scala
@@ -610,6 +610,9 @@ object MimaExcludes {
         // [SPARK-13674][SQL] Add wholestage codegen support to Sample
         ProblemFilters.exclude[IncompatibleMethTypeProblem]("org.apache.spark.util.random.PoissonSampler.this"),
         ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.util.random.PoissonSampler.this")
+      ) ++ Seq(
+        // [SPARK-13430][ML] moved featureCol from LinearRegressionModelSummary to LinearRegressionSummary
+        ProblemFilters.exclude[MissingMethodProblem]("org.apache.spark.ml.regression.LinearRegressionSummary.this")
       )
     case v if v.startsWith("1.6") =>
       Seq(
diff --git a/python/pyspark/ml/classification.py b/python/pyspark/ml/classification.py
index 067009559b..be7f9ea9ef 100644
--- a/python/pyspark/ml/classification.py
+++ b/python/pyspark/ml/classification.py
@@ -19,15 +19,18 @@ import warnings
 
 from pyspark import since
 from pyspark.ml.util import *
-from pyspark.ml.wrapper import JavaEstimator, JavaModel
+from pyspark.ml.wrapper import JavaEstimator, JavaModel, JavaCallable
 from pyspark.ml.param import TypeConverters
 from pyspark.ml.param.shared import *
 from pyspark.ml.regression import (
     RandomForestParams, TreeEnsembleParams, DecisionTreeModel, TreeEnsembleModels)
 from pyspark.mllib.common import inherit_doc
+from pyspark.sql import DataFrame
 
 
 __all__ = ['LogisticRegression', 'LogisticRegressionModel',
+           'LogisticRegressionSummary', 'LogisticRegressionTrainingSummary',
+           'BinaryLogisticRegressionSummary', 'BinaryLogisticRegressionTrainingSummary',
            'DecisionTreeClassifier', 'DecisionTreeClassificationModel',
            'GBTClassifier', 'GBTClassificationModel',
            'RandomForestClassifier', 'RandomForestClassificationModel',
@@ -233,6 +236,219 @@ class LogisticRegressionModel(JavaModel, JavaMLWritable, JavaMLReadable):
         """
         return self._call_java("intercept")
 
+    @property
+    @since("2.0.0")
+    def summary(self):
+        """
+        Gets summary (e.g. residuals, mse, r-squared ) of model on
+        training set. An exception is thrown if
+        `trainingSummary is None`.
+        """
+        java_blrt_summary = self._call_java("summary")
+        # Note: Once multiclass is added, update this to return correct summary
+        return BinaryLogisticRegressionTrainingSummary(java_blrt_summary)
+
+    @property
+    @since("2.0.0")
+    def hasSummary(self):
+        """
+        Indicates whether a training summary exists for this model
+        instance.
+        """
+        return self._call_java("hasSummary")
+
+    @since("2.0.0")
+    def evaluate(self, dataset):
+        """
+        Evaluates the model on a test dataset.
+
+        :param dataset:
+          Test dataset to evaluate model on, where dataset is an
+          instance of :py:class:`pyspark.sql.DataFrame`
+        """
+        if not isinstance(dataset, DataFrame):
+            raise ValueError("dataset must be a DataFrame but got %s." % type(dataset))
+        java_blr_summary = self._call_java("evaluate", dataset)
+        return BinaryLogisticRegressionSummary(java_blr_summary)
+
+
+class LogisticRegressionSummary(JavaCallable):
+    """
+    Abstraction for Logistic Regression Results for a given model.
+
+    .. versionadded:: 2.0.0
+    """
+
+    @property
+    @since("2.0.0")
+    def predictions(self):
+        """
+        Dataframe outputted by the model's `transform` method.
+        """
+        return self._call_java("predictions")
+
+    @property
+    @since("2.0.0")
+    def probabilityCol(self):
+        """
+        Field in "predictions" which gives the calibrated probability
+        of each class as a vector.
+        """
+        return self._call_java("probabilityCol")
+
+    @property
+    @since("2.0.0")
+    def labelCol(self):
+        """
+        Field in "predictions" which gives the true label of each
+        instance.
+        """
+        return self._call_java("labelCol")
+
+    @property
+    @since("2.0.0")
+    def featuresCol(self):
+        """
+        Field in "predictions" which gives the features of each instance
+        as a vector.
+        """
+        return self._call_java("featuresCol")
+
+
+@inherit_doc
+class LogisticRegressionTrainingSummary(LogisticRegressionSummary):
+    """
+    Abstraction for multinomial Logistic Regression Training results.
+    Currently, the training summary ignores the training weights except
+    for the objective trace.
+
+    .. versionadded:: 2.0.0
+    """
+
+    @property
+    @since("2.0.0")
+    def objectiveHistory(self):
+        """
+        Objective function (scaled loss + regularization) at each
+        iteration.
+        """
+        return self._call_java("objectiveHistory")
+
+    @property
+    @since("2.0.0")
+    def totalIterations(self):
+        """
+        Number of training iterations until termination.
+        """
+        return self._call_java("totalIterations")
+
+
+@inherit_doc
+class BinaryLogisticRegressionSummary(LogisticRegressionSummary):
+    """
+    .. note:: Experimental
+
+    Binary Logistic regression results for a given model.
+
+    .. versionadded:: 2.0.0
+    """
+
+    @property
+    @since("2.0.0")
+    def roc(self):
+        """
+        Returns the receiver operating characteristic (ROC) curve,
+        which is an Dataframe having two fields (FPR, TPR) with
+        (0.0, 0.0) prepended and (1.0, 1.0) appended to it.
+        Reference: http://en.wikipedia.org/wiki/Receiver_operating_characteristic
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LogisticRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("roc")
+
+    @property
+    @since("2.0.0")
+    def areaUnderROC(self):
+        """
+        Computes the area under the receiver operating characteristic
+        (ROC) curve.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LogisticRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("areaUnderROC")
+
+    @property
+    @since("2.0.0")
+    def pr(self):
+        """
+        Returns the precision-recall curve, which is an Dataframe
+        containing two fields recall, precision with (0.0, 1.0) prepended
+        to it.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LogisticRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("pr")
+
+    @property
+    @since("2.0.0")
+    def fMeasureByThreshold(self):
+        """
+        Returns a dataframe with two fields (threshold, F-Measure) curve
+        with beta = 1.0.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LogisticRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("fMeasureByThreshold")
+
+    @property
+    @since("2.0.0")
+    def precisionByThreshold(self):
+        """
+        Returns a dataframe with two fields (threshold, precision) curve.
+        Every possible probability obtained in transforming the dataset
+        are used as thresholds used in calculating the precision.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LogisticRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("precisionByThreshold")
+
+    @property
+    @since("2.0.0")
+    def recallByThreshold(self):
+        """
+        Returns a dataframe with two fields (threshold, recall) curve.
+        Every possible probability obtained in transforming the dataset
+        are used as thresholds used in calculating the recall.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LogisticRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("recallByThreshold")
+
+
+@inherit_doc
+class BinaryLogisticRegressionTrainingSummary(BinaryLogisticRegressionSummary,
+                                              LogisticRegressionTrainingSummary):
+    """
+    .. note:: Experimental
+
+    Binary Logistic regression training results for a given model.
+
+    .. versionadded:: 2.0.0
+    """
+    pass
+
 
 class TreeClassifierParams(object):
     """
diff --git a/python/pyspark/ml/regression.py b/python/pyspark/ml/regression.py
index de8a5e4bed..6cd1b4bf3a 100644
--- a/python/pyspark/ml/regression.py
+++ b/python/pyspark/ml/regression.py
@@ -20,8 +20,9 @@ import warnings
 from pyspark import since
 from pyspark.ml.param.shared import *
 from pyspark.ml.util import *
-from pyspark.ml.wrapper import JavaEstimator, JavaModel
+from pyspark.ml.wrapper import JavaEstimator, JavaModel, JavaCallable
 from pyspark.mllib.common import inherit_doc
+from pyspark.sql import DataFrame
 
 
 __all__ = ['AFTSurvivalRegression', 'AFTSurvivalRegressionModel',
@@ -29,6 +30,7 @@ __all__ = ['AFTSurvivalRegression', 'AFTSurvivalRegressionModel',
            'GBTRegressor', 'GBTRegressionModel',
            'IsotonicRegression', 'IsotonicRegressionModel',
            'LinearRegression', 'LinearRegressionModel',
+           'LinearRegressionSummary', 'LinearRegressionTrainingSummary',
            'RandomForestRegressor', 'RandomForestRegressionModel']
 
 
@@ -131,7 +133,6 @@ class LinearRegressionModel(JavaModel, JavaMLWritable, JavaMLReadable):
         """
         Model weights.
         """
-
         warnings.warn("weights is deprecated. Use coefficients instead.")
         return self._call_java("weights")
 
@@ -151,6 +152,246 @@ class LinearRegressionModel(JavaModel, JavaMLWritable, JavaMLReadable):
         """
         return self._call_java("intercept")
 
+    @property
+    @since("2.0.0")
+    def summary(self):
+        """
+        Gets summary (e.g. residuals, mse, r-squared ) of model on
+        training set. An exception is thrown if
+        `trainingSummary is None`.
+        """
+        java_lrt_summary = self._call_java("summary")
+        return LinearRegressionTrainingSummary(java_lrt_summary)
+
+    @property
+    @since("2.0.0")
+    def hasSummary(self):
+        """
+        Indicates whether a training summary exists for this model
+        instance.
+        """
+        return self._call_java("hasSummary")
+
+    @since("2.0.0")
+    def evaluate(self, dataset):
+        """
+        Evaluates the model on a test dataset.
+
+        :param dataset:
+          Test dataset to evaluate model on, where dataset is an
+          instance of :py:class:`pyspark.sql.DataFrame`
+        """
+        if not isinstance(dataset, DataFrame):
+            raise ValueError("dataset must be a DataFrame but got %s." % type(dataset))
+        java_lr_summary = self._call_java("evaluate", dataset)
+        return LinearRegressionSummary(java_lr_summary)
+
+
+class LinearRegressionSummary(JavaCallable):
+    """
+    .. note:: Experimental
+
+    Linear regression results evaluated on a dataset.
+
+    .. versionadded:: 2.0.0
+    """
+
+    @property
+    @since("2.0.0")
+    def predictions(self):
+        """
+        Dataframe outputted by the model's `transform` method.
+        """
+        return self._call_java("predictions")
+
+    @property
+    @since("2.0.0")
+    def predictionCol(self):
+        """
+        Field in "predictions" which gives the predicted value of
+        the label at each instance.
+        """
+        return self._call_java("predictionCol")
+
+    @property
+    @since("2.0.0")
+    def labelCol(self):
+        """
+        Field in "predictions" which gives the true label of each
+        instance.
+        """
+        return self._call_java("labelCol")
+
+    @property
+    @since("2.0.0")
+    def featuresCol(self):
+        """
+        Field in "predictions" which gives the features of each instance
+        as a vector.
+        """
+        return self._call_java("featuresCol")
+
+    @property
+    @since("2.0.0")
+    def explainedVariance(self):
+        """
+        Returns the explained variance regression score.
+        explainedVariance = 1 - variance(y - \hat{y}) / variance(y)
+        Reference: http://en.wikipedia.org/wiki/Explained_variation
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LinearRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("explainedVariance")
+
+    @property
+    @since("2.0.0")
+    def meanAbsoluteError(self):
+        """
+        Returns the mean absolute error, which is a risk function
+        corresponding to the expected value of the absolute error
+        loss or l1-norm loss.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LinearRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("meanAbsoluteError")
+
+    @property
+    @since("2.0.0")
+    def meanSquaredError(self):
+        """
+        Returns the mean squared error, which is a risk function
+        corresponding to the expected value of the squared error
+        loss or quadratic loss.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LinearRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("meanSquaredError")
+
+    @property
+    @since("2.0.0")
+    def rootMeanSquaredError(self):
+        """
+        Returns the root mean squared error, which is defined as the
+        square root of the mean squared error.
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LinearRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("rootMeanSquaredError")
+
+    @property
+    @since("2.0.0")
+    def r2(self):
+        """
+        Returns R^2^, the coefficient of determination.
+        Reference: http://en.wikipedia.org/wiki/Coefficient_of_determination
+
+        Note: This ignores instance weights (setting all to 1.0) from
+        `LinearRegression.weightCol`. This will change in later Spark
+        versions.
+        """
+        return self._call_java("r2")
+
+    @property
+    @since("2.0.0")
+    def residuals(self):
+        """
+        Residuals (label - predicted value)
+        """
+        return self._call_java("residuals")
+
+    @property
+    @since("2.0.0")
+    def numInstances(self):
+        """
+        Number of instances in DataFrame predictions
+        """
+        return self._call_java("numInstances")
+
+    @property
+    @since("2.0.0")
+    def devianceResiduals(self):
+        """
+        The weighted residuals, the usual residuals rescaled by the
+        square root of the instance weights.
+        """
+        return self._call_java("devianceResiduals")
+
+    @property
+    @since("2.0.0")
+    def coefficientStandardErrors(self):
+        """
+        Standard error of estimated coefficients and intercept.
+        This value is only available when using the "normal" solver.
+
+        .. seealso:: :py:attr:`LinearRegression.solver`
+        """
+        return self._call_java("coefficientStandardErrors")
+
+    @property
+    @since("2.0.0")
+    def tValues(self):
+        """
+        T-statistic of estimated coefficients and intercept.
+        This value is only available when using the "normal" solver.
+
+        .. seealso:: :py:attr:`LinearRegression.solver`
+        """
+        return self._call_java("tValues")
+
+    @property
+    @since("2.0.0")
+    def pValues(self):
+        """
+        Two-sided p-value of estimated coefficients and intercept.
+        This value is only available when using the "normal" solver.
+
+        .. seealso:: :py:attr:`LinearRegression.solver`
+        """
+        return self._call_java("pValues")
+
+
+@inherit_doc
+class LinearRegressionTrainingSummary(LinearRegressionSummary):
+    """
+    .. note:: Experimental
+
+    Linear regression training results. Currently, the training summary ignores the
+    training weights except for the objective trace.
+
+    .. versionadded:: 2.0.0
+    """
+
+    @property
+    @since("2.0.0")
+    def objectiveHistory(self):
+        """
+        Objective function (scaled loss + regularization) at each
+        iteration.
+        This value is only available when using the "l-bfgs" solver.
+
+        .. seealso:: :py:attr:`LinearRegression.solver`
+        """
+        return self._call_java("objectiveHistory")
+
+    @property
+    @since("2.0.0")
+    def totalIterations(self):
+        """
+        Number of training iterations until termination.
+        This value is only available when using the "l-bfgs" solver.
+
+        .. seealso:: :py:attr:`LinearRegression.solver`
+        """
+        return self._call_java("totalIterations")
+
 
 @inherit_doc
 class IsotonicRegression(JavaEstimator, HasFeaturesCol, HasLabelCol, HasPredictionCol,
diff --git a/python/pyspark/ml/tests.py b/python/pyspark/ml/tests.py
index e3f873e3a7..2dcd5eeb52 100644
--- a/python/pyspark/ml/tests.py
+++ b/python/pyspark/ml/tests.py
@@ -239,6 +239,17 @@ class OtherTestParams(HasMaxIter, HasInputCol, HasSeed):
         return self._set(**kwargs)
 
 
+class HasThrowableProperty(Params):
+
+    def __init__(self):
+        super(HasThrowableProperty, self).__init__()
+        self.p = Param(self, "none", "empty param")
+
+    @property
+    def test_property(self):
+        raise RuntimeError("Test property to raise error when invoked")
+
+
 class ParamTests(PySparkTestCase):
 
     def test_copy_new_parent(self):
@@ -749,15 +760,75 @@ class PersistenceTest(PySparkTestCase):
             pass
 
 
-class HasThrowableProperty(Params):
-
-    def __init__(self):
-        super(HasThrowableProperty, self).__init__()
-        self.p = Param(self, "none", "empty param")
+class TrainingSummaryTest(PySparkTestCase):
 
-    @property
-    def test_property(self):
-        raise RuntimeError("Test property to raise error when invoked")
+    def test_linear_regression_summary(self):
+        from pyspark.mllib.linalg import Vectors
+        sqlContext = SQLContext(self.sc)
+        df = sqlContext.createDataFrame([(1.0, 2.0, Vectors.dense(1.0)),
+                                         (0.0, 2.0, Vectors.sparse(1, [], []))],
+                                        ["label", "weight", "features"])
+        lr = LinearRegression(maxIter=5, regParam=0.0, solver="normal", weightCol="weight",
+                              fitIntercept=False)
+        model = lr.fit(df)
+        self.assertTrue(model.hasSummary)
+        s = model.summary
+        # test that api is callable and returns expected types
+        self.assertGreater(s.totalIterations, 0)
+        self.assertTrue(isinstance(s.predictions, DataFrame))
+        self.assertEqual(s.predictionCol, "prediction")
+        self.assertEqual(s.labelCol, "label")
+        self.assertEqual(s.featuresCol, "features")
+        objHist = s.objectiveHistory
+        self.assertTrue(isinstance(objHist, list) and isinstance(objHist[0], float))
+        self.assertAlmostEqual(s.explainedVariance, 0.25, 2)
+        self.assertAlmostEqual(s.meanAbsoluteError, 0.0)
+        self.assertAlmostEqual(s.meanSquaredError, 0.0)
+        self.assertAlmostEqual(s.rootMeanSquaredError, 0.0)
+        self.assertAlmostEqual(s.r2, 1.0, 2)
+        self.assertTrue(isinstance(s.residuals, DataFrame))
+        self.assertEqual(s.numInstances, 2)
+        devResiduals = s.devianceResiduals
+        self.assertTrue(isinstance(devResiduals, list) and isinstance(devResiduals[0], float))
+        coefStdErr = s.coefficientStandardErrors
+        self.assertTrue(isinstance(coefStdErr, list) and isinstance(coefStdErr[0], float))
+        tValues = s.tValues
+        self.assertTrue(isinstance(tValues, list) and isinstance(tValues[0], float))
+        pValues = s.pValues
+        self.assertTrue(isinstance(pValues, list) and isinstance(pValues[0], float))
+        # test evaluation (with training dataset) produces a summary with same values
+        # one check is enough to verify a summary is returned, Scala version runs full test
+        sameSummary = model.evaluate(df)
+        self.assertAlmostEqual(sameSummary.explainedVariance, s.explainedVariance)
+
+    def test_logistic_regression_summary(self):
+        from pyspark.mllib.linalg import Vectors
+        sqlContext = SQLContext(self.sc)
+        df = sqlContext.createDataFrame([(1.0, 2.0, Vectors.dense(1.0)),
+                                         (0.0, 2.0, Vectors.sparse(1, [], []))],
+                                        ["label", "weight", "features"])
+        lr = LogisticRegression(maxIter=5, regParam=0.01, weightCol="weight", fitIntercept=False)
+        model = lr.fit(df)
+        self.assertTrue(model.hasSummary)
+        s = model.summary
+        # test that api is callable and returns expected types
+        self.assertTrue(isinstance(s.predictions, DataFrame))
+        self.assertEqual(s.probabilityCol, "probability")
+        self.assertEqual(s.labelCol, "label")
+        self.assertEqual(s.featuresCol, "features")
+        objHist = s.objectiveHistory
+        self.assertTrue(isinstance(objHist, list) and isinstance(objHist[0], float))
+        self.assertGreater(s.totalIterations, 0)
+        self.assertTrue(isinstance(s.roc, DataFrame))
+        self.assertAlmostEqual(s.areaUnderROC, 1.0, 2)
+        self.assertTrue(isinstance(s.pr, DataFrame))
+        self.assertTrue(isinstance(s.fMeasureByThreshold, DataFrame))
+        self.assertTrue(isinstance(s.precisionByThreshold, DataFrame))
+        self.assertTrue(isinstance(s.recallByThreshold, DataFrame))
+        # test evaluation (with training dataset) produces a summary with same values
+        # one check is enough to verify a summary is returned, Scala version runs full test
+        sameSummary = model.evaluate(df)
+        self.assertAlmostEqual(sameSummary.areaUnderROC, s.areaUnderROC)
 
 
 if __name__ == "__main__":
diff --git a/python/pyspark/ml/wrapper.py b/python/pyspark/ml/wrapper.py
index ca93bf7d7d..a2cf2296fb 100644
--- a/python/pyspark/ml/wrapper.py
+++ b/python/pyspark/ml/wrapper.py
@@ -213,8 +213,30 @@ class JavaTransformer(Transformer, JavaWrapper):
         return DataFrame(self._java_obj.transform(dataset._jdf), dataset.sql_ctx)
 
 
+class JavaCallable(object):
+    """
+    Wrapper for a plain object in JVM to make Java calls, can be used
+    as a mixin to another class that defines a _java_obj wrapper
+    """
+    def __init__(self, java_obj=None, sc=None):
+        super(JavaCallable, self).__init__()
+        self._sc = sc if sc is not None else SparkContext._active_spark_context
+        # if this class is a mixin and _java_obj is already defined then don't initialize
+        if java_obj is not None or not hasattr(self, "_java_obj"):
+            self._java_obj = java_obj
+
+    def __del__(self):
+        if self._java_obj is not None:
+            self._sc._gateway.detach(self._java_obj)
+
+    def _call_java(self, name, *args):
+        m = getattr(self._java_obj, name)
+        java_args = [_py2java(self._sc, arg) for arg in args]
+        return _java2py(self._sc, m(*java_args))
+
+
 @inherit_doc
-class JavaModel(Model, JavaTransformer):
+class JavaModel(Model, JavaCallable, JavaTransformer):
     """
     Base class for :py:class:`Model`s that wrap Java/Scala
     implementations. Subclasses should inherit this class before
@@ -259,9 +281,3 @@ class JavaModel(Model, JavaTransformer):
             that._java_obj = self._java_obj.copy(self._empty_java_param_map())
             that._transfer_params_to_java()
         return that
-
-    def _call_java(self, name, *args):
-        m = getattr(self._java_obj, name)
-        sc = SparkContext._active_spark_context
-        java_args = [_py2java(sc, arg) for arg in args]
-        return _java2py(sc, m(*java_args))
-- 
cgit v1.2.3


From d717ae1fd74d125a9df21350a70e7c2b2d2b4786 Mon Sep 17 00:00:00 2001
From: Dongjoon Hyun <dongjoon@apache.org>
Date: Wed, 6 Apr 2016 16:02:55 -0700
Subject: [SPARK-14444][BUILD] Add a new scalastyle `NoScalaDoc` to prevent
 ScalaDoc-style multiline comments

## What changes were proposed in this pull request?

According to the [Spark Code Style Guide](https://cwiki.apache.org/confluence/display/SPARK/Spark+Code+Style+Guide#SparkCodeStyleGuide-Indentation), this PR adds a new scalastyle rule to prevent the followings.
```
/** In Spark, we don't use the ScalaDoc style so this
  * is not correct.
  */
```

## How was this patch tested?

Pass the Jenkins tests (including `lint-scala`).

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #12221 from dongjoon-hyun/SPARK-14444.
---
 core/src/main/scala/org/apache/spark/SparkConf.scala   |  6 ++++--
 .../org/apache/spark/deploy/SparkHadoopUtil.scala      | 12 ++++++------
 .../scala/org/apache/spark/partial/BoundedDouble.scala |  4 ++--
 .../apache/spark/examples/DriverSubmissionTest.scala   |  6 ++++--
 .../spark/streaming/flume/FlumeInputDStream.scala      |  6 ++++--
 .../mllib/stat/distribution/MultivariateGaussian.scala | 10 ++++++----
 scalastyle-config.xml                                  |  5 +++++
 .../apache/spark/sql/catalyst/ScalaReflection.scala    | 18 +++++++++---------
 .../catalyst/expressions/codegen/CodeGenerator.scala   |  8 ++++----
 .../apache/spark/sql/catalyst/plans/QueryPlan.scala    |  6 ++++--
 .../apache/spark/sql/RelationalGroupedDataset.scala    |  4 ++--
 .../apache/spark/sql/execution/WholeStageCodegen.scala |  4 ++--
 .../apache/spark/sql/execution/ui/SparkPlanGraph.scala |  4 ++--
 .../scala/org/apache/spark/sql/jdbc/JdbcDialects.scala | 10 +++++-----
 14 files changed, 59 insertions(+), 44 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/core/src/main/scala/org/apache/spark/SparkConf.scala b/core/src/main/scala/org/apache/spark/SparkConf.scala
index 5da2e98f1f..e0fd248c43 100644
--- a/core/src/main/scala/org/apache/spark/SparkConf.scala
+++ b/core/src/main/scala/org/apache/spark/SparkConf.scala
@@ -419,8 +419,10 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging {
    */
   private[spark] def getenv(name: String): String = System.getenv(name)
 
-  /** Checks for illegal or deprecated config settings. Throws an exception for the former. Not
-    * idempotent - may mutate this conf object to convert deprecated settings to supported ones. */
+  /**
+   * Checks for illegal or deprecated config settings. Throws an exception for the former. Not
+   * idempotent - may mutate this conf object to convert deprecated settings to supported ones.
+   */
   private[spark] def validateSettings() {
     if (contains("spark.local.dir")) {
       val msg = "In Spark 1.0 and later spark.local.dir will be overridden by the value set by " +
diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala b/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala
index 4e8e363635..41ac308808 100644
--- a/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala
@@ -76,9 +76,9 @@ class SparkHadoopUtil extends Logging {
 
 
   /**
-    * Appends S3-specific, spark.hadoop.*, and spark.buffer.size configurations to a Hadoop
-    * configuration.
-    */
+   * Appends S3-specific, spark.hadoop.*, and spark.buffer.size configurations to a Hadoop
+   * configuration.
+   */
   def appendS3AndSparkHadoopConfigurations(conf: SparkConf, hadoopConf: Configuration): Unit = {
     // Note: this null check is around more than just access to the "conf" object to maintain
     // the behavior of the old implementation of this code, for backwards compatibility.
@@ -108,9 +108,9 @@ class SparkHadoopUtil extends Logging {
   }
 
   /**
-    * Return an appropriate (subclass) of Configuration. Creating config can initializes some Hadoop
-    * subsystems.
-    */
+   * Return an appropriate (subclass) of Configuration. Creating config can initializes some Hadoop
+   * subsystems.
+   */
   def newConfiguration(conf: SparkConf): Configuration = {
     val hadoopConf = new Configuration()
     appendS3AndSparkHadoopConfigurations(conf, hadoopConf)
diff --git a/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala b/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala
index d06b2c67d2..c562c70aba 100644
--- a/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala
+++ b/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala
@@ -28,8 +28,8 @@ class BoundedDouble(val mean: Double, val confidence: Double, val low: Double, v
     this.mean.hashCode ^ this.confidence.hashCode ^ this.low.hashCode ^ this.high.hashCode
 
   /**
-    * Note that consistent with Double, any NaN value will make equality false
-    */
+   * Note that consistent with Double, any NaN value will make equality false
+   */
   override def equals(that: Any): Boolean =
     that match {
       case that: BoundedDouble => {
diff --git a/examples/src/main/scala/org/apache/spark/examples/DriverSubmissionTest.scala b/examples/src/main/scala/org/apache/spark/examples/DriverSubmissionTest.scala
index a2d59a1c95..d12ef642bd 100644
--- a/examples/src/main/scala/org/apache/spark/examples/DriverSubmissionTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/DriverSubmissionTest.scala
@@ -22,8 +22,10 @@ import scala.collection.JavaConverters._
 
 import org.apache.spark.util.Utils
 
-/** Prints out environmental information, sleeps, and then exits. Made to
-  * test driver submission in the standalone scheduler. */
+/**
+ * Prints out environmental information, sleeps, and then exits. Made to
+ * test driver submission in the standalone scheduler.
+ */
 object DriverSubmissionTest {
   def main(args: Array[String]) {
     if (args.length < 1) {
diff --git a/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala b/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala
index 6e7c3f358e..13aa817492 100644
--- a/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala
+++ b/external/flume/src/main/scala/org/apache/spark/streaming/flume/FlumeInputDStream.scala
@@ -130,8 +130,10 @@ class FlumeEventServer(receiver: FlumeReceiver) extends AvroSourceProtocol {
   }
 }
 
-/** A NetworkReceiver which listens for events using the
-  * Flume Avro interface. */
+/**
+ * A NetworkReceiver which listens for events using the
+ * Flume Avro interface.
+ */
 private[streaming]
 class FlumeReceiver(
     host: String,
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/stat/distribution/MultivariateGaussian.scala b/mllib/src/main/scala/org/apache/spark/mllib/stat/distribution/MultivariateGaussian.scala
index 052b5b1d65..6c6e9fb7c6 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/stat/distribution/MultivariateGaussian.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/stat/distribution/MultivariateGaussian.scala
@@ -61,15 +61,17 @@ class MultivariateGaussian @Since("1.3.0") (
    */
   private val (rootSigmaInv: DBM[Double], u: Double) = calculateCovarianceConstants
 
-  /** Returns density of this multivariate Gaussian at given point, x
-    */
+  /**
+   * Returns density of this multivariate Gaussian at given point, x
+   */
    @Since("1.3.0")
   def pdf(x: Vector): Double = {
     pdf(x.toBreeze)
   }
 
-  /** Returns the log-density of this multivariate Gaussian at given point, x
-    */
+  /**
+   * Returns the log-density of this multivariate Gaussian at given point, x
+   */
    @Since("1.3.0")
   def logpdf(x: Vector): Double = {
     logpdf(x.toBreeze)
diff --git a/scalastyle-config.xml b/scalastyle-config.xml
index 33c2cbd293..472a8f4084 100644
--- a/scalastyle-config.xml
+++ b/scalastyle-config.xml
@@ -223,6 +223,11 @@ This file is divided into 3 sections:
     ]]></customMessage>
   </check>
 
+  <check customId="NoScalaDoc" level="error" class="org.scalastyle.file.RegexChecker" enabled="true">
+    <parameters><parameter name="regex">(?m)^(\s*)/[*][*].*$(\r|)\n^\1  [*]</parameter></parameters>
+    <customMessage>Use Javadoc style indentation for multiline comments</customMessage>
+  </check>
+
   <!-- ================================================================================ -->
   <!--       rules we'd like to enforce, but haven't cleaned up the codebase yet        -->
   <!-- ================================================================================ -->
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala
index d241b8a79b..4795fc2557 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala
@@ -762,15 +762,15 @@ trait ScalaReflection {
   }
 
   /**
-    * Returns the full class name for a type. The returned name is the canonical
-    * Scala name, where each component is separated by a period. It is NOT the
-    * Java-equivalent runtime name (no dollar signs).
-    *
-    * In simple cases, both the Scala and Java names are the same, however when Scala
-    * generates constructs that do not map to a Java equivalent, such as singleton objects
-    * or nested classes in package objects, it uses the dollar sign ($) to create
-    * synthetic classes, emulating behaviour in Java bytecode.
-    */
+   * Returns the full class name for a type. The returned name is the canonical
+   * Scala name, where each component is separated by a period. It is NOT the
+   * Java-equivalent runtime name (no dollar signs).
+   *
+   * In simple cases, both the Scala and Java names are the same, however when Scala
+   * generates constructs that do not map to a Java equivalent, such as singleton objects
+   * or nested classes in package objects, it uses the dollar sign ($) to create
+   * synthetic classes, emulating behaviour in Java bytecode.
+   */
   def getClassNameFromType(tpe: `Type`): String = {
     tpe.erasure.typeSymbol.asClass.fullName
   }
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
index 1bebd4e904..ee7f4fadca 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
@@ -626,15 +626,15 @@ abstract class CodeGenerator[InType <: AnyRef, OutType <: AnyRef] extends Loggin
 
 object CodeGenerator extends Logging {
   /**
-    * Compile the Java source code into a Java class, using Janino.
-    */
+   * Compile the Java source code into a Java class, using Janino.
+   */
   def compile(code: String): GeneratedClass = {
     cache.get(code)
   }
 
   /**
-    * Compile the Java source code into a Java class, using Janino.
-    */
+   * Compile the Java source code into a Java class, using Janino.
+   */
   private[this] def doCompile(code: String): GeneratedClass = {
     val evaluator = new ClassBodyEvaluator()
     evaluator.setParentClassLoader(Utils.getContextOrSparkClassLoader)
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala
index 609a33e2f1..0a11574f44 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/QueryPlan.scala
@@ -211,8 +211,10 @@ abstract class QueryPlan[PlanType <: QueryPlan[PlanType]] extends TreeNode[PlanT
     if (changed) makeCopy(newArgs).asInstanceOf[this.type] else this
   }
 
-  /** Returns the result of running [[transformExpressions]] on this node
-    * and all its children. */
+  /**
+   * Returns the result of running [[transformExpressions]] on this node
+   * and all its children.
+   */
   def transformAllExpressions(rule: PartialFunction[Expression, Expression]): this.type = {
     transform {
       case q: QueryPlan[_] => q.transformExpressions(rule).asInstanceOf[PlanType]
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/RelationalGroupedDataset.scala b/sql/core/src/main/scala/org/apache/spark/sql/RelationalGroupedDataset.scala
index 91c02053ae..7dbf2e6c7c 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/RelationalGroupedDataset.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/RelationalGroupedDataset.scala
@@ -408,7 +408,7 @@ private[sql] object RelationalGroupedDataset {
   private[sql] object RollupType extends GroupType
 
   /**
-    * To indicate it's the PIVOT
-    */
+   * To indicate it's the PIVOT
+   */
   private[sql] case class PivotType(pivotCol: Expression, values: Seq[Literal]) extends GroupType
 }
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala
index 98129d6c52..c4594f0480 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/WholeStageCodegen.scala
@@ -312,8 +312,8 @@ case class WholeStageCodegen(child: SparkPlan) extends UnaryNode with CodegenSup
       }
 
       /** Codegened pipeline for:
-        * ${toCommentSafeString(child.treeString.trim)}
-        */
+       * ${toCommentSafeString(child.treeString.trim)}
+       */
       final class GeneratedIterator extends org.apache.spark.sql.execution.BufferedRowIterator {
 
         private Object[] references;
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala
index 012b125d6b..c6fcb6956c 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/ui/SparkPlanGraph.scala
@@ -167,8 +167,8 @@ private[ui] class SparkPlanGraphNode(
 }
 
 /**
-  * Represent a tree of SparkPlan for WholeStageCodegen.
-  */
+ * Represent a tree of SparkPlan for WholeStageCodegen.
+ */
 private[ui] class SparkPlanGraphCluster(
     id: Long,
     name: String,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala
index cfe4911cb7..948106fd06 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/jdbc/JdbcDialects.scala
@@ -100,11 +100,11 @@ abstract class JdbcDialect extends Serializable {
   }
 
   /**
-    * Override connection specific properties to run before a select is made.  This is in place to
-    * allow dialects that need special treatment to optimize behavior.
-    * @param connection The connection object
-    * @param properties The connection properties.  This is passed through from the relation.
-    */
+   * Override connection specific properties to run before a select is made.  This is in place to
+   * allow dialects that need special treatment to optimize behavior.
+   * @param connection The connection object
+   * @param properties The connection properties.  This is passed through from the relation.
+   */
   def beforeFetch(connection: Connection, properties: Map[String, String]): Unit = {
   }
 
-- 
cgit v1.2.3


From bb873754b4700104755ab969694bf30945557dc3 Mon Sep 17 00:00:00 2001
From: sethah <seth.hendrickson16@gmail.com>
Date: Wed, 6 Apr 2016 17:13:34 -0700
Subject: [SPARK-12382][ML] Remove mllib GBT implementation and wrap ml

## What changes were proposed in this pull request?

This patch removes the implementation of gradient boosted trees in mllib/tree/GradientBoostedTrees.scala and changes mllib GBTs to call the implementation in spark.ML.

Primary changes:
* Removed `boost` method in mllib GradientBoostedTrees.scala
* Created new test suite GradientBoostedTreesSuite in ML, which contains unit tests that were specific to GBT internals from mllib

Other changes:
* Added an `updatePrediction` method in GradientBoostedTrees package. This method is added to provide consistency for methods that build predictions from boosted models. There are several methods that hard code the method of predicting as: sum_{i=1}^{numTrees} (treePrediction*treeWeight). Calling this function ensures that test methods that check accuracy use the same prediction method that the algorithm uses during training
* Added methods that were previously only used in testing, but were public methods, to GradientBoostedTrees. This includes `computeError` (previously part  of `Loss` trait) and `evaluateEachIteration`. These are used in the new spark.ML unit tests. They are left in mllib as well so as to not break the API.

## How was this patch tested?

Existing unit tests which compare ML and MLlib ensure that mllib GBTs have not changed. Only a single unit test was moved to ML, which verifies that `runWithValidation` performs as expected.

Author: sethah <seth.hendrickson16@gmail.com>

Closes #12050 from sethah/SPARK-12382.
---
 .../spark/ml/tree/impl/GradientBoostedTrees.scala  | 115 ++++++++++++-
 .../org/apache/spark/ml/tree/treeModels.scala      |   4 +-
 .../spark/mllib/tree/GradientBoostedTrees.scala    | 182 +--------------------
 .../spark/ml/regression/GBTRegressorSuite.scala    |   2 +-
 .../ml/tree/impl/GradientBoostedTreesSuite.scala   |  85 ++++++++++
 .../mllib/tree/GradientBoostedTreesSuite.scala     |  45 +----
 6 files changed, 207 insertions(+), 226 deletions(-)
 create mode 100644 mllib/src/test/scala/org/apache/spark/ml/tree/impl/GradientBoostedTreesSuite.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
index 0749d93b7d..d365655674 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
@@ -19,7 +19,6 @@ package org.apache.spark.ml.tree.impl
 
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.regression.{DecisionTreeRegressionModel, DecisionTreeRegressor}
-import org.apache.spark.ml.tree.DecisionTreeModel
 import org.apache.spark.mllib.impl.PeriodicRDDCheckpointer
 import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.regression.LabeledPoint
@@ -30,7 +29,24 @@ import org.apache.spark.mllib.tree.loss.{Loss => OldLoss}
 import org.apache.spark.rdd.RDD
 import org.apache.spark.storage.StorageLevel
 
-private[ml] object GradientBoostedTrees extends Logging {
+
+/**
+ * A package that implements
+ * [[http://en.wikipedia.org/wiki/Gradient_boosting  Stochastic Gradient Boosting]]
+ * for regression and binary classification.
+ *
+ * The implementation is based upon:
+ *   J.H. Friedman.  "Stochastic Gradient Boosting."  1999.
+ *
+ * Notes on Gradient Boosting vs. TreeBoost:
+ *  - This implementation is for Stochastic Gradient Boosting, not for TreeBoost.
+ *  - Both algorithms learn tree ensembles by minimizing loss functions.
+ *  - TreeBoost (Friedman, 1999) additionally modifies the outputs at tree leaf nodes
+ *    based on the loss function, whereas the original gradient boosting method does not.
+ *     - When the loss is SquaredError, these methods give the same result, but they could differ
+ *       for other loss functions.
+ */
+private[spark] object GradientBoostedTrees extends Logging {
 
   /**
    * Method to train a gradient boosting model
@@ -107,7 +123,7 @@ private[ml] object GradientBoostedTrees extends Logging {
       initTree: DecisionTreeRegressionModel,
       loss: OldLoss): RDD[(Double, Double)] = {
     data.map { lp =>
-      val pred = initTreeWeight * initTree.rootNode.predictImpl(lp.features).prediction
+      val pred = updatePrediction(lp.features, 0.0, initTree, initTreeWeight)
       val error = loss.computeError(pred, lp.label)
       (pred, error)
     }
@@ -133,7 +149,7 @@ private[ml] object GradientBoostedTrees extends Logging {
 
     val newPredError = data.zip(predictionAndError).mapPartitions { iter =>
       iter.map { case (lp, (pred, error)) =>
-        val newPred = pred + tree.rootNode.predictImpl(lp.features).prediction * treeWeight
+        val newPred = updatePrediction(lp.features, pred, tree, treeWeight)
         val newError = loss.computeError(newPred, lp.label)
         (newPred, newError)
       }
@@ -141,6 +157,97 @@ private[ml] object GradientBoostedTrees extends Logging {
     newPredError
   }
 
+  /**
+   * Add prediction from a new boosting iteration to an existing prediction.
+   *
+   * @param features Vector of features representing a single data point.
+   * @param prediction The existing prediction.
+   * @param tree New Decision Tree model.
+   * @param weight Tree weight.
+   * @return Updated prediction.
+   */
+  def updatePrediction(
+      features: Vector,
+      prediction: Double,
+      tree: DecisionTreeRegressionModel,
+      weight: Double): Double = {
+    prediction + tree.rootNode.predictImpl(features).prediction * weight
+  }
+
+  /**
+   * Method to calculate error of the base learner for the gradient boosting calculation.
+   * Note: This method is not used by the gradient boosting algorithm but is useful for debugging
+   * purposes.
+   * @param data Training dataset: RDD of [[org.apache.spark.mllib.regression.LabeledPoint]].
+   * @param trees Boosted Decision Tree models
+   * @param treeWeights Learning rates at each boosting iteration.
+   * @param loss evaluation metric.
+   * @return Measure of model error on data
+   */
+  def computeError(
+      data: RDD[LabeledPoint],
+      trees: Array[DecisionTreeRegressionModel],
+      treeWeights: Array[Double],
+      loss: OldLoss): Double = {
+    data.map { lp =>
+      val predicted = trees.zip(treeWeights).foldLeft(0.0) { case (acc, (model, weight)) =>
+        updatePrediction(lp.features, acc, model, weight)
+      }
+      loss.computeError(predicted, lp.label)
+    }.mean()
+  }
+
+  /**
+   * Method to compute error or loss for every iteration of gradient boosting.
+   *
+   * @param data RDD of [[org.apache.spark.mllib.regression.LabeledPoint]]
+   * @param trees Boosted Decision Tree models
+   * @param treeWeights Learning rates at each boosting iteration.
+   * @param loss evaluation metric.
+   * @param algo algorithm for the ensemble, either Classification or Regression
+   * @return an array with index i having the losses or errors for the ensemble
+   *         containing the first i+1 trees
+   */
+  def evaluateEachIteration(
+      data: RDD[LabeledPoint],
+      trees: Array[DecisionTreeRegressionModel],
+      treeWeights: Array[Double],
+      loss: OldLoss,
+      algo: OldAlgo.Value): Array[Double] = {
+
+    val sc = data.sparkContext
+    val remappedData = algo match {
+      case OldAlgo.Classification => data.map(x => new LabeledPoint((x.label * 2) - 1, x.features))
+      case _ => data
+    }
+
+    val numIterations = trees.length
+    val evaluationArray = Array.fill(numIterations)(0.0)
+    val localTreeWeights = treeWeights
+
+    var predictionAndError = computeInitialPredictionAndError(
+      remappedData, localTreeWeights(0), trees(0), loss)
+
+    evaluationArray(0) = predictionAndError.values.mean()
+
+    val broadcastTrees = sc.broadcast(trees)
+    (1 until numIterations).foreach { nTree =>
+      predictionAndError = remappedData.zip(predictionAndError).mapPartitions { iter =>
+        val currentTree = broadcastTrees.value(nTree)
+        val currentTreeWeight = localTreeWeights(nTree)
+        iter.map { case (point, (pred, error)) =>
+          val newPred = updatePrediction(point.features, pred, currentTree, currentTreeWeight)
+          val newError = loss.computeError(newPred, point.label)
+          (newPred, newError)
+        }
+      }
+      evaluationArray(nTree) = predictionAndError.values.mean()
+    }
+
+    broadcastTrees.unpersist()
+    evaluationArray
+  }
+
   /**
    * Internal method for performing regression using trees as base learners.
    * @param input training dataset
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
index db0ff28d82..c4ab673d9a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
@@ -133,8 +133,8 @@ private[ml] object TreeEnsembleModel {
    * following the explanation of Gini importance from "Random Forests" documentation
    * by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
    *
-   *  For collections of trees, including boosting and bagging, Hastie et al.
-   *  propose to use the average of single tree importances across all trees in the ensemble.
+   * For collections of trees, including boosting and bagging, Hastie et al.
+   * propose to use the average of single tree importances across all trees in the ensemble.
    *
    * This feature importance is calculated as follows:
    *  - Average over trees:
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala
index 0f0c6b466d..7fe60e2d99 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/GradientBoostedTrees.scala
@@ -20,15 +20,11 @@ package org.apache.spark.mllib.tree
 import org.apache.spark.annotation.Since
 import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.internal.Logging
-import org.apache.spark.ml.tree.impl.TimeTracker
-import org.apache.spark.mllib.impl.PeriodicRDDCheckpointer
+import org.apache.spark.ml.tree.impl.{GradientBoostedTrees => NewGBT}
 import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.mllib.tree.configuration.Algo._
 import org.apache.spark.mllib.tree.configuration.BoostingStrategy
-import org.apache.spark.mllib.tree.impurity.Variance
-import org.apache.spark.mllib.tree.model.{DecisionTreeModel, GradientBoostedTreesModel}
+import org.apache.spark.mllib.tree.model.GradientBoostedTreesModel
 import org.apache.spark.rdd.RDD
-import org.apache.spark.storage.StorageLevel
 
 /**
  * A class that implements
@@ -70,17 +66,8 @@ class GradientBoostedTrees private[spark] (
   @Since("1.2.0")
   def run(input: RDD[LabeledPoint]): GradientBoostedTreesModel = {
     val algo = boostingStrategy.treeStrategy.algo
-    algo match {
-      case Regression =>
-        GradientBoostedTrees.boost(input, input, boostingStrategy, validate = false, seed)
-      case Classification =>
-        // Map labels to -1, +1 so binary classification can be treated as regression.
-        val remappedInput = input.map(x => new LabeledPoint((x.label * 2) - 1, x.features))
-        GradientBoostedTrees.boost(remappedInput, remappedInput, boostingStrategy, validate = false,
-          seed)
-      case _ =>
-        throw new IllegalArgumentException(s"$algo is not supported by the gradient boosting.")
-    }
+    val (trees, treeWeights) = NewGBT.run(input, boostingStrategy, seed.toLong)
+    new GradientBoostedTreesModel(algo, trees.map(_.toOld), treeWeights)
   }
 
   /**
@@ -107,20 +94,9 @@ class GradientBoostedTrees private[spark] (
       input: RDD[LabeledPoint],
       validationInput: RDD[LabeledPoint]): GradientBoostedTreesModel = {
     val algo = boostingStrategy.treeStrategy.algo
-    algo match {
-      case Regression =>
-        GradientBoostedTrees.boost(input, validationInput, boostingStrategy, validate = true, seed)
-      case Classification =>
-        // Map labels to -1, +1 so binary classification can be treated as regression.
-        val remappedInput = input.map(
-          x => new LabeledPoint((x.label * 2) - 1, x.features))
-        val remappedValidationInput = validationInput.map(
-          x => new LabeledPoint((x.label * 2) - 1, x.features))
-        GradientBoostedTrees.boost(remappedInput, remappedValidationInput, boostingStrategy,
-          validate = true, seed)
-      case _ =>
-        throw new IllegalArgumentException(s"$algo is not supported by the gradient boosting.")
-    }
+    val (trees, treeWeights) = NewGBT.runWithValidation(input, validationInput, boostingStrategy,
+      seed.toLong)
+    new GradientBoostedTreesModel(algo, trees.map(_.toOld), treeWeights)
   }
 
   /**
@@ -162,148 +138,4 @@ object GradientBoostedTrees extends Logging {
       boostingStrategy: BoostingStrategy): GradientBoostedTreesModel = {
     train(input.rdd, boostingStrategy)
   }
-
-  /**
-   * Internal method for performing regression using trees as base learners.
-   *
-   * @param input Training dataset.
-   * @param validationInput Validation dataset, ignored if validate is set to false.
-   * @param boostingStrategy Boosting parameters.
-   * @param validate Whether or not to use the validation dataset.
-   * @param seed Random seed.
-   * @return GradientBoostedTreesModel that can be used for prediction.
-   */
-  private def boost(
-      input: RDD[LabeledPoint],
-      validationInput: RDD[LabeledPoint],
-      boostingStrategy: BoostingStrategy,
-      validate: Boolean,
-      seed: Int): GradientBoostedTreesModel = {
-    val timer = new TimeTracker()
-    timer.start("total")
-    timer.start("init")
-
-    boostingStrategy.assertValid()
-
-    // Initialize gradient boosting parameters
-    val numIterations = boostingStrategy.numIterations
-    val baseLearners = new Array[DecisionTreeModel](numIterations)
-    val baseLearnerWeights = new Array[Double](numIterations)
-    val loss = boostingStrategy.loss
-    val learningRate = boostingStrategy.learningRate
-    // Prepare strategy for individual trees, which use regression with variance impurity.
-    val treeStrategy = boostingStrategy.treeStrategy.copy
-    val validationTol = boostingStrategy.validationTol
-    treeStrategy.algo = Regression
-    treeStrategy.impurity = Variance
-    treeStrategy.assertValid()
-
-    // Cache input
-    val persistedInput = if (input.getStorageLevel == StorageLevel.NONE) {
-      input.persist(StorageLevel.MEMORY_AND_DISK)
-      true
-    } else {
-      false
-    }
-
-    // Prepare periodic checkpointers
-    val predErrorCheckpointer = new PeriodicRDDCheckpointer[(Double, Double)](
-      treeStrategy.getCheckpointInterval, input.sparkContext)
-    val validatePredErrorCheckpointer = new PeriodicRDDCheckpointer[(Double, Double)](
-      treeStrategy.getCheckpointInterval, input.sparkContext)
-
-    timer.stop("init")
-
-    logDebug("##########")
-    logDebug("Building tree 0")
-    logDebug("##########")
-
-    // Initialize tree
-    timer.start("building tree 0")
-    val firstTreeModel = new DecisionTree(treeStrategy, seed).run(input)
-    val firstTreeWeight = 1.0
-    baseLearners(0) = firstTreeModel
-    baseLearnerWeights(0) = firstTreeWeight
-
-    var predError: RDD[(Double, Double)] = GradientBoostedTreesModel.
-      computeInitialPredictionAndError(input, firstTreeWeight, firstTreeModel, loss)
-    predErrorCheckpointer.update(predError)
-    logDebug("error of gbt = " + predError.values.mean())
-
-    // Note: A model of type regression is used since we require raw prediction
-    timer.stop("building tree 0")
-
-    var validatePredError: RDD[(Double, Double)] = GradientBoostedTreesModel.
-      computeInitialPredictionAndError(validationInput, firstTreeWeight, firstTreeModel, loss)
-    if (validate) validatePredErrorCheckpointer.update(validatePredError)
-    var bestValidateError = if (validate) validatePredError.values.mean() else 0.0
-    var bestM = 1
-
-    var m = 1
-    var doneLearning = false
-    while (m < numIterations && !doneLearning) {
-      // Update data with pseudo-residuals
-      val data = predError.zip(input).map { case ((pred, _), point) =>
-        LabeledPoint(-loss.gradient(pred, point.label), point.features)
-      }
-
-      timer.start(s"building tree $m")
-      logDebug("###################################################")
-      logDebug("Gradient boosting tree iteration " + m)
-      logDebug("###################################################")
-      val model = new DecisionTree(treeStrategy, seed + m).run(data)
-      timer.stop(s"building tree $m")
-      // Update partial model
-      baseLearners(m) = model
-      // Note: The setting of baseLearnerWeights is incorrect for losses other than SquaredError.
-      //       Technically, the weight should be optimized for the particular loss.
-      //       However, the behavior should be reasonable, though not optimal.
-      baseLearnerWeights(m) = learningRate
-
-      predError = GradientBoostedTreesModel.updatePredictionError(
-        input, predError, baseLearnerWeights(m), baseLearners(m), loss)
-      predErrorCheckpointer.update(predError)
-      logDebug("error of gbt = " + predError.values.mean())
-
-      if (validate) {
-        // Stop training early if
-        // 1. Reduction in error is less than the validationTol or
-        // 2. If the error increases, that is if the model is overfit.
-        // We want the model returned corresponding to the best validation error.
-
-        validatePredError = GradientBoostedTreesModel.updatePredictionError(
-          validationInput, validatePredError, baseLearnerWeights(m), baseLearners(m), loss)
-        validatePredErrorCheckpointer.update(validatePredError)
-        val currentValidateError = validatePredError.values.mean()
-        if (bestValidateError - currentValidateError < validationTol * Math.max(
-          currentValidateError, 0.01)) {
-          doneLearning = true
-        } else if (currentValidateError < bestValidateError) {
-          bestValidateError = currentValidateError
-          bestM = m + 1
-        }
-      }
-      m += 1
-    }
-
-    timer.stop("total")
-
-    logInfo("Internal timing for DecisionTree:")
-    logInfo(s"$timer")
-
-    predErrorCheckpointer.deleteAllCheckpoints()
-    validatePredErrorCheckpointer.deleteAllCheckpoints()
-    if (persistedInput) input.unpersist()
-
-    if (validate) {
-      new GradientBoostedTreesModel(
-        boostingStrategy.treeStrategy.algo,
-        baseLearners.slice(0, bestM),
-        baseLearnerWeights.slice(0, bestM))
-    } else {
-      new GradientBoostedTreesModel(
-        boostingStrategy.treeStrategy.algo, baseLearners, baseLearnerWeights)
-    }
-  }
-
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
index 914818f41f..3c11631f98 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
@@ -53,7 +53,7 @@ class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
       sc.parallelize(EnsembleTestHelper.generateOrderedLabeledPoints(numFeatures = 20, 80), 2)
   }
 
-  test("Regression with continuous features: SquaredError") {
+  test("Regression with continuous features") {
     val categoricalFeatures = Map.empty[Int, Int]
     GBTRegressor.supportedLossTypes.foreach { loss =>
       testCombinations.foreach {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/GradientBoostedTreesSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/GradientBoostedTreesSuite.scala
new file mode 100644
index 0000000000..fecf372c3d
--- /dev/null
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/GradientBoostedTreesSuite.scala
@@ -0,0 +1,85 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.tree.impl
+
+import org.apache.spark.SparkFunSuite
+import org.apache.spark.internal.Logging
+import org.apache.spark.mllib.regression.LabeledPoint
+import org.apache.spark.mllib.tree.{GradientBoostedTreesSuite => OldGBTSuite}
+import org.apache.spark.mllib.tree.configuration.{BoostingStrategy, Strategy}
+import org.apache.spark.mllib.tree.configuration.Algo._
+import org.apache.spark.mllib.tree.impurity.Variance
+import org.apache.spark.mllib.tree.loss.{AbsoluteError, LogLoss, SquaredError}
+import org.apache.spark.mllib.util.MLlibTestSparkContext
+
+/**
+ * Test suite for [[GradientBoostedTrees]].
+ */
+class GradientBoostedTreesSuite extends SparkFunSuite with MLlibTestSparkContext with Logging {
+
+  test("runWithValidation stops early and performs better on a validation dataset") {
+    // Set numIterations large enough so that it stops early.
+    val numIterations = 20
+    val trainRdd = sc.parallelize(OldGBTSuite.trainData, 2)
+    val validateRdd = sc.parallelize(OldGBTSuite.validateData, 2)
+    val trainDF = sqlContext.createDataFrame(trainRdd)
+    val validateDF = sqlContext.createDataFrame(validateRdd)
+
+    val algos = Array(Regression, Regression, Classification)
+    val losses = Array(SquaredError, AbsoluteError, LogLoss)
+    algos.zip(losses).foreach { case (algo, loss) =>
+      val treeStrategy = new Strategy(algo = algo, impurity = Variance, maxDepth = 2,
+        categoricalFeaturesInfo = Map.empty)
+      val boostingStrategy =
+        new BoostingStrategy(treeStrategy, loss, numIterations, validationTol = 0.0)
+      val (validateTrees, validateTreeWeights) = GradientBoostedTrees
+        .runWithValidation(trainRdd, validateRdd, boostingStrategy, 42L)
+      val numTrees = validateTrees.length
+      assert(numTrees !== numIterations)
+
+      // Test that it performs better on the validation dataset.
+      val (trees, treeWeights) = GradientBoostedTrees.run(trainRdd, boostingStrategy, 42L)
+      val (errorWithoutValidation, errorWithValidation) = {
+        if (algo == Classification) {
+          val remappedRdd = validateRdd.map(x => new LabeledPoint(2 * x.label - 1, x.features))
+          (GradientBoostedTrees.computeError(remappedRdd, trees, treeWeights, loss),
+            GradientBoostedTrees.computeError(remappedRdd, validateTrees,
+              validateTreeWeights, loss))
+        } else {
+          (GradientBoostedTrees.computeError(validateRdd, trees, treeWeights, loss),
+            GradientBoostedTrees.computeError(validateRdd, validateTrees,
+              validateTreeWeights, loss))
+        }
+      }
+      assert(errorWithValidation <= errorWithoutValidation)
+
+      // Test that results from evaluateEachIteration comply with runWithValidation.
+      // Note that convergenceTol is set to 0.0
+      val evaluationArray = GradientBoostedTrees
+        .evaluateEachIteration(validateRdd, trees, treeWeights, loss, algo)
+      assert(evaluationArray.length === numIterations)
+      assert(evaluationArray(numTrees) > evaluationArray(numTrees - 1))
+      var i = 1
+      while (i < numTrees) {
+        assert(evaluationArray(i) <= evaluationArray(i - 1))
+        i += 1
+      }
+    }
+  }
+
+}
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala
index 747c267b4f..c61f89322d 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala
@@ -158,49 +158,6 @@ class GradientBoostedTreesSuite extends SparkFunSuite with MLlibTestSparkContext
     }
   }
 
-  test("runWithValidation stops early and performs better on a validation dataset") {
-    // Set numIterations large enough so that it stops early.
-    val numIterations = 20
-    val trainRdd = sc.parallelize(GradientBoostedTreesSuite.trainData, 2)
-    val validateRdd = sc.parallelize(GradientBoostedTreesSuite.validateData, 2)
-
-    val algos = Array(Regression, Regression, Classification)
-    val losses = Array(SquaredError, AbsoluteError, LogLoss)
-    algos.zip(losses).foreach { case (algo, loss) =>
-      val treeStrategy = new Strategy(algo = algo, impurity = Variance, maxDepth = 2,
-        categoricalFeaturesInfo = Map.empty)
-      val boostingStrategy =
-        new BoostingStrategy(treeStrategy, loss, numIterations, validationTol = 0.0)
-      val gbtValidate = new GradientBoostedTrees(boostingStrategy, seed = 0)
-        .runWithValidation(trainRdd, validateRdd)
-      val numTrees = gbtValidate.numTrees
-      assert(numTrees !== numIterations)
-
-      // Test that it performs better on the validation dataset.
-      val gbt = new GradientBoostedTrees(boostingStrategy, seed = 0).run(trainRdd)
-      val (errorWithoutValidation, errorWithValidation) = {
-        if (algo == Classification) {
-          val remappedRdd = validateRdd.map(x => new LabeledPoint(2 * x.label - 1, x.features))
-          (loss.computeError(gbt, remappedRdd), loss.computeError(gbtValidate, remappedRdd))
-        } else {
-          (loss.computeError(gbt, validateRdd), loss.computeError(gbtValidate, validateRdd))
-        }
-      }
-      assert(errorWithValidation <= errorWithoutValidation)
-
-      // Test that results from evaluateEachIteration comply with runWithValidation.
-      // Note that convergenceTol is set to 0.0
-      val evaluationArray = gbt.evaluateEachIteration(validateRdd, loss)
-      assert(evaluationArray.length === numIterations)
-      assert(evaluationArray(numTrees) > evaluationArray(numTrees - 1))
-      var i = 1
-      while (i < numTrees) {
-        assert(evaluationArray(i) <= evaluationArray(i - 1))
-        i += 1
-      }
-    }
-  }
-
   test("Checkpointing") {
     val tempDir = Utils.createTempDir()
     val path = tempDir.toURI.toString
@@ -220,7 +177,7 @@ class GradientBoostedTreesSuite extends SparkFunSuite with MLlibTestSparkContext
 
 }
 
-private object GradientBoostedTreesSuite {
+private[spark] object GradientBoostedTreesSuite {
 
   // Combinations for estimators, learning rates and subsamplingRate
   val testCombinations = Array((10, 1.0, 1.0), (10, 0.1, 1.0), (10, 0.5, 0.75), (10, 0.1, 0.75))
-- 
cgit v1.2.3


From 953ff897e422570a329d0aec98d573d3fb66ab9a Mon Sep 17 00:00:00 2001
From: "Joseph K. Bradley" <joseph@databricks.com>
Date: Thu, 7 Apr 2016 19:48:33 -0700
Subject: [SPARK-13048][ML][MLLIB] keepLastCheckpoint option for LDA EM
 optimizer

## What changes were proposed in this pull request?

The EMLDAOptimizer should generally not delete its last checkpoint since that can cause failures when DistributedLDAModel methods are called (if any partitions need to be recovered from the checkpoint).

This PR adds a "deleteLastCheckpoint" option which defaults to false.  This is a change in behavior from Spark 1.6, in that the last checkpoint will not be removed by default.

This involves adding the deleteLastCheckpoint option to both spark.ml and spark.mllib, and modifying PeriodicCheckpointer to support the option.

This also:
* Makes MLlibTestSparkContext extend TempDirectory and set the checkpointDir to tempDir
* Updates LibSVMRelationSuite because of a name conflict with "tempDir" (and fixes a bug where it failed to delete a temp directory)
* Adds a MIMA exclude for DistributedLDAModel constructor, which is already ```private[clustering]```

## How was this patch tested?

Added 2 new unit tests to spark.ml LDASuite, which calls into spark.mllib.

Author: Joseph K. Bradley <joseph@databricks.com>

Closes #12166 from jkbradley/emlda-save-checkpoint.
---
 .../scala/org/apache/spark/ml/clustering/LDA.scala | 79 ++++++++++++++++++++--
 .../apache/spark/mllib/clustering/LDAModel.scala   | 13 ++--
 .../spark/mllib/clustering/LDAOptimizer.scala      | 34 ++++++++--
 .../spark/mllib/impl/PeriodicCheckpointer.scala    | 41 ++++++++---
 .../org/apache/spark/ml/clustering/LDASuite.scala  | 28 ++++++++
 .../ml/source/libsvm/LibSVMRelationSuite.scala     | 15 ++--
 .../spark/mllib/util/MLlibTestSparkContext.scala   | 13 +++-
 project/MimaExcludes.scala                         |  3 +
 8 files changed, 194 insertions(+), 32 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
index 60cc345565..727b724708 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
@@ -17,18 +17,19 @@
 
 package org.apache.spark.ml.clustering
 
-import org.apache.hadoop.fs.Path
+import org.apache.hadoop.fs.{FileSystem, Path}
 
-import org.apache.spark.annotation.{Experimental, Since}
+import org.apache.spark.annotation.{DeveloperApi, Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared.{HasCheckpointInterval, HasFeaturesCol, HasMaxIter, HasSeed}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.clustering.{DistributedLDAModel => OldDistributedLDAModel,
-    EMLDAOptimizer => OldEMLDAOptimizer, LDA => OldLDA, LDAModel => OldLDAModel,
-    LDAOptimizer => OldLDAOptimizer, LocalLDAModel => OldLocalLDAModel,
-    OnlineLDAOptimizer => OldOnlineLDAOptimizer}
+  EMLDAOptimizer => OldEMLDAOptimizer, LDA => OldLDA, LDAModel => OldLDAModel,
+  LDAOptimizer => OldLDAOptimizer, LocalLDAModel => OldLocalLDAModel,
+  OnlineLDAOptimizer => OldOnlineLDAOptimizer}
+import org.apache.spark.mllib.impl.PeriodicCheckpointer
 import org.apache.spark.mllib.linalg.{Matrix, Vector, Vectors, VectorUDT}
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Row, SQLContext}
@@ -41,6 +42,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
 
   /**
    * Param for the number of topics (clusters) to infer. Must be > 1. Default: 10.
+   *
    * @group param
    */
   @Since("1.6.0")
@@ -173,6 +175,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * This uses a variational approximation following Hoffman et al. (2010), where the approximate
    * distribution is called "gamma."  Technically, this method returns this approximation "gamma"
    * for each document.
+   *
    * @group param
    */
   @Since("1.6.0")
@@ -191,6 +194,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * iterations count less.
    * This is called "tau0" in the Online LDA paper (Hoffman et al., 2010)
    * Default: 1024, following Hoffman et al.
+   *
    * @group expertParam
    */
   @Since("1.6.0")
@@ -207,6 +211,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * This should be between (0.5, 1.0] to guarantee asymptotic convergence.
    * This is called "kappa" in the Online LDA paper (Hoffman et al., 2010).
    * Default: 0.51, based on Hoffman et al.
+   *
    * @group expertParam
    */
   @Since("1.6.0")
@@ -230,6 +235,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    *       [[org.apache.spark.mllib.clustering.OnlineLDAOptimizer]].
    *
    * Default: 0.05, i.e., 5% of total documents.
+   *
    * @group param
    */
   @Since("1.6.0")
@@ -246,6 +252,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * document-topic distribution) will be optimized during training.
    * Setting this to true will make the model more expressive and fit the training data better.
    * Default: false
+   *
    * @group expertParam
    */
   @Since("1.6.0")
@@ -257,8 +264,32 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
   @Since("1.6.0")
   def getOptimizeDocConcentration: Boolean = $(optimizeDocConcentration)
 
+  /**
+   * For EM optimizer, if using checkpointing, this indicates whether to keep the last
+   * checkpoint. If false, then the checkpoint will be deleted. Deleting the checkpoint can
+   * cause failures if a data partition is lost, so set this bit with care.
+   * Note that checkpoints will be cleaned up via reference counting, regardless.
+   *
+   * See [[DistributedLDAModel.getCheckpointFiles]] for getting remaining checkpoints and
+   * [[DistributedLDAModel.deleteCheckpointFiles]] for removing remaining checkpoints.
+   *
+   * Default: true
+   *
+   * @group expertParam
+   */
+  @Since("2.0.0")
+  final val keepLastCheckpoint = new BooleanParam(this, "keepLastCheckpoint",
+    "For EM optimizer, if using checkpointing, this indicates whether to keep the last" +
+      " checkpoint. If false, then the checkpoint will be deleted. Deleting the checkpoint can" +
+      " cause failures if a data partition is lost, so set this bit with care.")
+
+  /** @group expertGetParam */
+  @Since("2.0.0")
+  def getKeepLastCheckpoint: Boolean = $(keepLastCheckpoint)
+
   /**
    * Validates and transforms the input schema.
+   *
    * @param schema input schema
    * @return output schema
    */
@@ -303,6 +334,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
         .setOptimizeDocConcentration($(optimizeDocConcentration))
     case "em" =>
       new OldEMLDAOptimizer()
+        .setKeepLastCheckpoint($(keepLastCheckpoint))
   }
 }
 
@@ -341,6 +373,7 @@ sealed abstract class LDAModel private[ml] (
   /**
    * The features for LDA should be a [[Vector]] representing the word counts in a document.
    * The vector should be of length vocabSize, with counts for each term (word).
+   *
    * @group setParam
    */
   @Since("1.6.0")
@@ -619,6 +652,35 @@ class DistributedLDAModel private[ml] (
   @Since("1.6.0")
   lazy val logPrior: Double = oldDistributedModel.logPrior
 
+  private var _checkpointFiles: Array[String] = oldDistributedModel.checkpointFiles
+
+  /**
+   * If using checkpointing and [[LDA.keepLastCheckpoint]] is set to true, then there may be
+   * saved checkpoint files.  This method is provided so that users can manage those files.
+   *
+   * Note that removing the checkpoints can cause failures if a partition is lost and is needed
+   * by certain [[DistributedLDAModel]] methods.  Reference counting will clean up the checkpoints
+   * when this model and derivative data go out of scope.
+   *
+   * @return  Checkpoint files from training
+   */
+  @DeveloperApi
+  @Since("2.0.0")
+  def getCheckpointFiles: Array[String] = _checkpointFiles
+
+  /**
+   * Remove any remaining checkpoint files from training.
+   *
+   * @see [[getCheckpointFiles]]
+   */
+  @DeveloperApi
+  @Since("2.0.0")
+  def deleteCheckpointFiles(): Unit = {
+    val fs = FileSystem.get(sqlContext.sparkContext.hadoopConfiguration)
+    _checkpointFiles.foreach(PeriodicCheckpointer.removeCheckpointFile(_, fs))
+    _checkpointFiles = Array.empty[String]
+  }
+
   @Since("1.6.0")
   override def write: MLWriter = new DistributedLDAModel.DistributedWriter(this)
 }
@@ -696,11 +758,12 @@ class LDA @Since("1.6.0") (
 
   setDefault(maxIter -> 20, k -> 10, optimizer -> "online", checkpointInterval -> 10,
     learningOffset -> 1024, learningDecay -> 0.51, subsamplingRate -> 0.05,
-    optimizeDocConcentration -> true)
+    optimizeDocConcentration -> true, keepLastCheckpoint -> true)
 
   /**
    * The features for LDA should be a [[Vector]] representing the word counts in a document.
    * The vector should be of length vocabSize, with counts for each term (word).
+   *
    * @group setParam
    */
   @Since("1.6.0")
@@ -758,6 +821,10 @@ class LDA @Since("1.6.0") (
   @Since("1.6.0")
   def setOptimizeDocConcentration(value: Boolean): this.type = set(optimizeDocConcentration, value)
 
+  /** @group expertSetParam */
+  @Since("2.0.0")
+  def setKeepLastCheckpoint(value: Boolean): this.type = set(keepLastCheckpoint, value)
+
   @Since("1.6.0")
   override def copy(extra: ParamMap): LDA = defaultCopy(extra)
 
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAModel.scala
index 25d67a3756..27b4004927 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAModel.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAModel.scala
@@ -534,7 +534,8 @@ class DistributedLDAModel private[clustering] (
     @Since("1.5.0") override val docConcentration: Vector,
     @Since("1.5.0") override val topicConcentration: Double,
     private[spark] val iterationTimes: Array[Double],
-    override protected[clustering] val gammaShape: Double = 100)
+    override protected[clustering] val gammaShape: Double = DistributedLDAModel.defaultGammaShape,
+    private[spark] val checkpointFiles: Array[String] = Array.empty[String])
   extends LDAModel {
 
   import LDA._
@@ -806,11 +807,9 @@ class DistributedLDAModel private[clustering] (
 
   override protected def formatVersion = "1.0"
 
-  /**
-   * Java-friendly version of [[topicDistributions]]
-   */
   @Since("1.5.0")
   override def save(sc: SparkContext, path: String): Unit = {
+    // Note: This intentionally does not save checkpointFiles.
     DistributedLDAModel.SaveLoadV1_0.save(
       sc, path, graph, globalTopicTotals, k, vocabSize, docConcentration, topicConcentration,
       iterationTimes, gammaShape)
@@ -822,6 +821,12 @@ class DistributedLDAModel private[clustering] (
 @Since("1.5.0")
 object DistributedLDAModel extends Loader[DistributedLDAModel] {
 
+  /**
+   * The [[DistributedLDAModel]] constructor's default arguments assume gammaShape = 100
+   * to ensure equivalence in LDAModel.toLocal conversion.
+   */
+  private[clustering] val defaultGammaShape: Double = 100
+
   private object SaveLoadV1_0 {
 
     val thisFormatVersion = "1.0"
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala
index 2b404a8651..6418f0d3b3 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala
@@ -80,9 +80,29 @@ final class EMLDAOptimizer extends LDAOptimizer {
 
   import LDA._
 
+  // Adjustable parameters
+  private var keepLastCheckpoint: Boolean = true
+
   /**
-   * The following fields will only be initialized through the initialize() method
+   * If using checkpointing, this indicates whether to keep the last checkpoint (vs clean up).
+   */
+  @Since("2.0.0")
+  def getKeepLastCheckpoint: Boolean = this.keepLastCheckpoint
+
+  /**
+   * If using checkpointing, this indicates whether to keep the last checkpoint (vs clean up).
+   * Deleting the checkpoint can cause failures if a data partition is lost, so set this bit with
+   * care.  Note that checkpoints will be cleaned up via reference counting, regardless.
+   *
+   * Default: true
    */
+  @Since("2.0.0")
+  def setKeepLastCheckpoint(keepLastCheckpoint: Boolean): this.type = {
+    this.keepLastCheckpoint = keepLastCheckpoint
+    this
+  }
+
+  // The following fields will only be initialized through the initialize() method
   private[clustering] var graph: Graph[TopicCounts, TokenCount] = null
   private[clustering] var k: Int = 0
   private[clustering] var vocabSize: Int = 0
@@ -208,12 +228,18 @@ final class EMLDAOptimizer extends LDAOptimizer {
 
   override private[clustering] def getLDAModel(iterationTimes: Array[Double]): LDAModel = {
     require(graph != null, "graph is null, EMLDAOptimizer not initialized.")
-    this.graphCheckpointer.deleteAllCheckpoints()
+    val checkpointFiles: Array[String] = if (keepLastCheckpoint) {
+      this.graphCheckpointer.deleteAllCheckpointsButLast()
+      this.graphCheckpointer.getAllCheckpointFiles
+    } else {
+      this.graphCheckpointer.deleteAllCheckpoints()
+      Array.empty[String]
+    }
     // The constructor's default arguments assume gammaShape = 100 to ensure equivalence in
-    // LDAModel.toLocal conversion
+    // LDAModel.toLocal conversion.
     new DistributedLDAModel(this.graph, this.globalTopicTotals, this.k, this.vocabSize,
       Vectors.dense(Array.fill(this.k)(this.docConcentration)), this.topicConcentration,
-      iterationTimes)
+      iterationTimes, DistributedLDAModel.defaultGammaShape, checkpointFiles)
   }
 }
 
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala b/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala
index 391f89aa14..cbc8f60112 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala
@@ -133,6 +133,24 @@ private[mllib] abstract class PeriodicCheckpointer[T](
     }
   }
 
+  /**
+   * Call this at the end to delete any remaining checkpoint files, except for the last checkpoint.
+   * Note that there may not be any checkpoints at all.
+   */
+  def deleteAllCheckpointsButLast(): Unit = {
+    while (checkpointQueue.size > 1) {
+      removeCheckpointFile()
+    }
+  }
+
+  /**
+   * Get all current checkpoint files.
+   * This is useful in combination with [[deleteAllCheckpointsButLast()]].
+   */
+  def getAllCheckpointFiles: Array[String] = {
+    checkpointQueue.flatMap(getCheckpointFiles).toArray
+  }
+
   /**
    * Dequeue the oldest checkpointed Dataset, and remove its checkpoint files.
    * This prints a warning but does not fail if the files cannot be removed.
@@ -141,15 +159,20 @@ private[mllib] abstract class PeriodicCheckpointer[T](
     val old = checkpointQueue.dequeue()
     // Since the old checkpoint is not deleted by Spark, we manually delete it.
     val fs = FileSystem.get(sc.hadoopConfiguration)
-    getCheckpointFiles(old).foreach { checkpointFile =>
-      try {
-        fs.delete(new Path(checkpointFile), true)
-      } catch {
-        case e: Exception =>
-          logWarning("PeriodicCheckpointer could not remove old checkpoint file: " +
-            checkpointFile)
-      }
-    }
+    getCheckpointFiles(old).foreach(PeriodicCheckpointer.removeCheckpointFile(_, fs))
   }
+}
+
+private[spark] object PeriodicCheckpointer extends Logging {
 
+  /** Delete a checkpoint file, and log a warning if deletion fails. */
+  def removeCheckpointFile(path: String, fs: FileSystem): Unit = {
+    try {
+      fs.delete(new Path(path), true)
+    } catch {
+      case e: Exception =>
+        logWarning("PeriodicCheckpointer could not remove old checkpoint file: " +
+          path)
+    }
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala
index dd3f4c6e53..a1c93891c7 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala
@@ -17,6 +17,8 @@
 
 package org.apache.spark.ml.clustering
 
+import org.apache.hadoop.fs.{FileSystem, Path}
+
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
@@ -261,4 +263,30 @@ class LDASuite extends SparkFunSuite with MLlibTestSparkContext with DefaultRead
     testEstimatorAndModelReadWrite(lda, dataset,
       LDASuite.allParamSettings ++ Map("optimizer" -> "em"), checkModelData)
   }
+
+  test("EM LDA checkpointing: save last checkpoint") {
+    // Checkpoint dir is set by MLlibTestSparkContext
+    val lda = new LDA().setK(2).setSeed(1).setOptimizer("em").setMaxIter(3).setCheckpointInterval(1)
+    val model_ = lda.fit(dataset)
+    assert(model_.isInstanceOf[DistributedLDAModel])
+    val model = model_.asInstanceOf[DistributedLDAModel]
+
+    // There should be 1 checkpoint remaining.
+    assert(model.getCheckpointFiles.length === 1)
+    val fs = FileSystem.get(sqlContext.sparkContext.hadoopConfiguration)
+    assert(fs.exists(new Path(model.getCheckpointFiles.head)))
+    model.deleteCheckpointFiles()
+    assert(model.getCheckpointFiles.isEmpty)
+  }
+
+  test("EM LDA checkpointing: remove last checkpoint") {
+    // Checkpoint dir is set by MLlibTestSparkContext
+    val lda = new LDA().setK(2).setSeed(1).setOptimizer("em").setMaxIter(3).setCheckpointInterval(1)
+      .setKeepLastCheckpoint(false)
+    val model_ = lda.fit(dataset)
+    assert(model_.isInstanceOf[DistributedLDAModel])
+    val model = model_.asInstanceOf[DistributedLDAModel]
+
+    assert(model.getCheckpointFiles.isEmpty)
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/source/libsvm/LibSVMRelationSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/source/libsvm/LibSVMRelationSuite.scala
index 114a238462..0bd14978b2 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/source/libsvm/LibSVMRelationSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/source/libsvm/LibSVMRelationSuite.scala
@@ -28,8 +28,9 @@ import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.sql.SaveMode
 import org.apache.spark.util.Utils
 
+
 class LibSVMRelationSuite extends SparkFunSuite with MLlibTestSparkContext {
-  var tempDir: File = _
+  // Path for dataset
   var path: String = _
 
   override def beforeAll(): Unit = {
@@ -40,15 +41,15 @@ class LibSVMRelationSuite extends SparkFunSuite with MLlibTestSparkContext {
         |0
         |0 2:4.0 4:5.0 6:6.0
       """.stripMargin
-    tempDir = Utils.createTempDir()
-    val file = new File(tempDir, "part-00000")
+    val dir = Utils.createDirectory(tempDir.getCanonicalPath, "data")
+    val file = new File(dir, "part-00000")
     Files.write(lines, file, StandardCharsets.UTF_8)
-    path = tempDir.toURI.toString
+    path = dir.toURI.toString
   }
 
   override def afterAll(): Unit = {
     try {
-      Utils.deleteRecursively(tempDir)
+      Utils.deleteRecursively(new File(path))
     } finally {
       super.afterAll()
     }
@@ -86,7 +87,7 @@ class LibSVMRelationSuite extends SparkFunSuite with MLlibTestSparkContext {
 
   test("write libsvm data and read it again") {
     val df = sqlContext.read.format("libsvm").load(path)
-    val tempDir2 = Utils.createTempDir()
+    val tempDir2 = new File(tempDir, "read_write_test")
     val writepath = tempDir2.toURI.toString
     // TODO: Remove requirement to coalesce by supporting multiple reads.
     df.coalesce(1).write.format("libsvm").mode(SaveMode.Overwrite).save(writepath)
@@ -99,7 +100,7 @@ class LibSVMRelationSuite extends SparkFunSuite with MLlibTestSparkContext {
 
   test("write libsvm data failed due to invalid schema") {
     val df = sqlContext.read.format("text").load(path)
-    val e = intercept[SparkException] {
+    intercept[SparkException] {
       df.write.format("libsvm").save(path + "_2")
     }
   }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/util/MLlibTestSparkContext.scala b/mllib/src/test/scala/org/apache/spark/mllib/util/MLlibTestSparkContext.scala
index ebcd591465..cb1efd5251 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/util/MLlibTestSparkContext.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/util/MLlibTestSparkContext.scala
@@ -17,14 +17,20 @@
 
 package org.apache.spark.mllib.util
 
-import org.scalatest.{BeforeAndAfterAll, Suite}
+import java.io.File
+
+import org.apache.hadoop.fs.Path
+import org.scalatest.Suite
 
 import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.ml.util.TempDirectory
 import org.apache.spark.sql.SQLContext
+import org.apache.spark.util.Utils
 
-trait MLlibTestSparkContext extends BeforeAndAfterAll { self: Suite =>
+trait MLlibTestSparkContext extends TempDirectory { self: Suite =>
   @transient var sc: SparkContext = _
   @transient var sqlContext: SQLContext = _
+  @transient var checkpointDir: String = _
 
   override def beforeAll() {
     super.beforeAll()
@@ -35,10 +41,13 @@ trait MLlibTestSparkContext extends BeforeAndAfterAll { self: Suite =>
     SQLContext.clearActive()
     sqlContext = new SQLContext(sc)
     SQLContext.setActive(sqlContext)
+    checkpointDir = Utils.createDirectory(tempDir.getCanonicalPath, "checkpoints").toString
+    sc.setCheckpointDir(checkpointDir)
   }
 
   override def afterAll() {
     try {
+      Utils.deleteRecursively(new File(checkpointDir))
       sqlContext = null
       SQLContext.clearActive()
       if (sc != null) {
diff --git a/project/MimaExcludes.scala b/project/MimaExcludes.scala
index fbadc563b8..a53161dc9a 100644
--- a/project/MimaExcludes.scala
+++ b/project/MimaExcludes.scala
@@ -614,6 +614,9 @@ object MimaExcludes {
       ) ++ Seq(
         // [SPARK-13430][ML] moved featureCol from LinearRegressionModelSummary to LinearRegressionSummary
         ProblemFilters.exclude[MissingMethodProblem]("org.apache.spark.ml.regression.LinearRegressionSummary.this")
+      ) ++ Seq(
+        // [SPARK-13048][ML][MLLIB] keepLastCheckpoint option for LDA EM optimizer
+        ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.mllib.clustering.DistributedLDAModel.this")
       )
     case v if v.startsWith("1.6") =>
       Seq(
-- 
cgit v1.2.3


From 56af8e85cca056096fe4e765d8d287e0f9efc0d2 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Fri, 8 Apr 2016 11:49:44 -0700
Subject: [SPARK-14298][ML][MLLIB] LDA should support disable checkpoint

## What changes were proposed in this pull request?
In the doc of [```checkpointInterval```](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/ml/param/shared/sharedParams.scala#L241), we told users that they can disable checkpoint by setting ```checkpointInterval = -1```. But we did not handle this situation for LDA actually, we should fix this bug.
## How was this patch tested?
Existing tests.

cc jkbradley

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12089 from yanboliang/spark-14298.
---
 .../scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala    | 6 ++++--
 .../org/apache/spark/mllib/impl/PeriodicGraphCheckpointer.scala     | 3 ++-
 2 files changed, 6 insertions(+), 3 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala b/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala
index cbc8f60112..5c12c9305b 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicCheckpointer.scala
@@ -52,7 +52,8 @@ import org.apache.spark.storage.StorageLevel
  *  - This class removes checkpoint files once later Datasets have been checkpointed.
  *    However, references to the older Datasets will still return isCheckpointed = true.
  *
- * @param checkpointInterval  Datasets will be checkpointed at this interval
+ * @param checkpointInterval  Datasets will be checkpointed at this interval.
+ *                            If this interval was set as -1, then checkpointing will be disabled.
  * @param sc  SparkContext for the Datasets given to this checkpointer
  * @tparam T  Dataset type, such as RDD[Double]
  */
@@ -89,7 +90,8 @@ private[mllib] abstract class PeriodicCheckpointer[T](
     updateCount += 1
 
     // Handle checkpointing (after persisting)
-    if ((updateCount % checkpointInterval) == 0 && sc.getCheckpointDir.nonEmpty) {
+    if (checkpointInterval != -1 && (updateCount % checkpointInterval) == 0
+      && sc.getCheckpointDir.nonEmpty) {
       // Add new checkpoint before removing old checkpoints.
       checkpoint(newData)
       checkpointQueue.enqueue(newData)
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicGraphCheckpointer.scala b/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicGraphCheckpointer.scala
index 11a059536c..20db6084d0 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicGraphCheckpointer.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/impl/PeriodicGraphCheckpointer.scala
@@ -69,7 +69,8 @@ import org.apache.spark.storage.StorageLevel
  *  // checkpointed: graph4
  * }}}
  *
- * @param checkpointInterval Graphs will be checkpointed at this interval
+ * @param checkpointInterval Graphs will be checkpointed at this interval.
+ *                           If this interval was set as -1, then checkpointing will be disabled.
  * @tparam VD  Vertex descriptor type
  * @tparam ED  Edge descriptor type
  *
-- 
cgit v1.2.3


From d7af736b2cf6c392b87e7b45c2d2219ef06979eb Mon Sep 17 00:00:00 2001
From: "Joseph K. Bradley" <joseph@databricks.com>
Date: Fri, 8 Apr 2016 20:15:44 -0700
Subject: [SPARK-14498][ML][PYTHON][SQL] Many cleanups to ML and ML-related
 docs
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## What changes were proposed in this pull request?

Cleanups to documentation.  No changes to code.
* GBT docs: Move Scala doc for private object GradientBoostedTrees to public docs for GBTClassifier,Regressor
* GLM regParam: needs doc saying it is for L2 only
* TrainValidationSplitModel: add .. versionadded:: 2.0.0
* Rename “_transformer_params_from_java” to “_transfer_params_from_java”
* LogReg Summary classes: “probability” col should not say “calibrated”
* LR summaries: coefficientStandardErrors —> document that intercept stderr comes last.  Same for t,p-values
* approxCountDistinct: Document meaning of “rsd" argument.
* LDA: note which params are for online LDA only

## How was this patch tested?

Doc build

Author: Joseph K. Bradley <joseph@databricks.com>

Closes #12266 from jkbradley/ml-doc-cleanups.
---
 .../spark/ml/classification/GBTClassifier.scala    | 10 +++++++
 .../ml/classification/LogisticRegression.scala     | 12 ++++----
 .../scala/org/apache/spark/ml/clustering/LDA.scala | 34 +++++++++++++++-------
 .../apache/spark/ml/regression/GBTRegressor.scala  | 12 ++++++++
 .../regression/GeneralizedLinearRegression.scala   | 17 +++++++++--
 .../spark/ml/regression/LinearRegression.scala     | 19 ++++++++----
 .../spark/ml/tree/impl/GradientBoostedTrees.scala  | 16 ----------
 python/pyspark/ml/classification.py                |  2 +-
 python/pyspark/ml/regression.py                    |  9 ++++++
 python/pyspark/ml/tuning.py                        |  2 ++
 python/pyspark/ml/wrapper.py                       |  2 +-
 .../scala/org/apache/spark/sql/functions.scala     |  4 +++
 12 files changed, 97 insertions(+), 42 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
index bee90fb3a5..a2150fbcc3 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -42,6 +42,16 @@ import org.apache.spark.sql.functions._
  * learning algorithm for classification.
  * It supports binary labels, as well as both continuous and categorical features.
  * Note: Multiclass labels are not currently supported.
+ *
+ * The implementation is based upon: J.H. Friedman. "Stochastic Gradient Boosting." 1999.
+ *
+ * Notes on Gradient Boosting vs. TreeBoost:
+ *  - This implementation is for Stochastic Gradient Boosting, not for TreeBoost.
+ *  - Both algorithms learn tree ensembles by minimizing loss functions.
+ *  - TreeBoost (Friedman, 1999) additionally modifies the outputs at tree leaf nodes
+ *    based on the loss function, whereas the original gradient boosting method does not.
+ *  - We expect to implement TreeBoost in the future:
+ *    [https://issues.apache.org/jira/browse/SPARK-4240]
  */
 @Since("1.4.0")
 @Experimental
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
index 37182928cc..268c3e32c3 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
@@ -774,10 +774,10 @@ sealed trait LogisticRegressionTrainingSummary extends LogisticRegressionSummary
  */
 sealed trait LogisticRegressionSummary extends Serializable {
 
-  /** Dataframe outputted by the model's `transform` method. */
+  /** Dataframe output by the model's `transform` method. */
   def predictions: DataFrame
 
-  /** Field in "predictions" which gives the calibrated probability of each class as a vector. */
+  /** Field in "predictions" which gives the probability of each class as a vector. */
   def probabilityCol: String
 
   /** Field in "predictions" which gives the true label of each instance (if available). */
@@ -792,8 +792,8 @@ sealed trait LogisticRegressionSummary extends Serializable {
  * :: Experimental ::
  * Logistic regression training results.
  *
- * @param predictions dataframe outputted by the model's `transform` method.
- * @param probabilityCol field in "predictions" which gives the calibrated probability of
+ * @param predictions dataframe output by the model's `transform` method.
+ * @param probabilityCol field in "predictions" which gives the probability of
  *                       each class as a vector.
  * @param labelCol field in "predictions" which gives the true label of each instance.
  * @param featuresCol field in "predictions" which gives the features of each instance as a vector.
@@ -816,8 +816,8 @@ class BinaryLogisticRegressionTrainingSummary private[classification] (
  * :: Experimental ::
  * Binary Logistic regression results for a given model.
  *
- * @param predictions dataframe outputted by the model's `transform` method.
- * @param probabilityCol field in "predictions" which gives the calibrated probability of
+ * @param predictions dataframe output by the model's `transform` method.
+ * @param probabilityCol field in "predictions" which gives the probability of
  *                       each class as a vector.
  * @param labelCol field in "predictions" which gives the true label of each instance.
  * @param featuresCol field in "predictions" which gives the features of each instance as a vector.
diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
index 727b724708..89a7a4ccf6 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
@@ -190,6 +190,8 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
   def getTopicDistributionCol: String = $(topicDistributionCol)
 
   /**
+   * For Online optimizer only: [[optimizer]] = "online".
+   *
    * A (positive) learning parameter that downweights early iterations. Larger values make early
    * iterations count less.
    * This is called "tau0" in the Online LDA paper (Hoffman et al., 2010)
@@ -198,8 +200,9 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * @group expertParam
    */
   @Since("1.6.0")
-  final val learningOffset = new DoubleParam(this, "learningOffset", "A (positive) learning" +
-    " parameter that downweights early iterations. Larger values make early iterations count less.",
+  final val learningOffset = new DoubleParam(this, "learningOffset", "(For online optimizer)" +
+    " A (positive) learning parameter that downweights early iterations. Larger values make early" +
+    " iterations count less.",
     ParamValidators.gt(0))
 
   /** @group expertGetParam */
@@ -207,6 +210,8 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
   def getLearningOffset: Double = $(learningOffset)
 
   /**
+   * For Online optimizer only: [[optimizer]] = "online".
+   *
    * Learning rate, set as an exponential decay rate.
    * This should be between (0.5, 1.0] to guarantee asymptotic convergence.
    * This is called "kappa" in the Online LDA paper (Hoffman et al., 2010).
@@ -215,15 +220,17 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * @group expertParam
    */
   @Since("1.6.0")
-  final val learningDecay = new DoubleParam(this, "learningDecay", "Learning rate, set as an" +
-    " exponential decay rate. This should be between (0.5, 1.0] to guarantee asymptotic" +
-    " convergence.", ParamValidators.gt(0))
+  final val learningDecay = new DoubleParam(this, "learningDecay", "(For online optimizer)" +
+    " Learning rate, set as an exponential decay rate. This should be between (0.5, 1.0] to" +
+    " guarantee asymptotic convergence.", ParamValidators.gt(0))
 
   /** @group expertGetParam */
   @Since("1.6.0")
   def getLearningDecay: Double = $(learningDecay)
 
   /**
+   * For Online optimizer only: [[optimizer]] = "online".
+   *
    * Fraction of the corpus to be sampled and used in each iteration of mini-batch gradient descent,
    * in range (0, 1].
    *
@@ -239,8 +246,9 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    * @group param
    */
   @Since("1.6.0")
-  final val subsamplingRate = new DoubleParam(this, "subsamplingRate", "Fraction of the corpus" +
-    " to be sampled and used in each iteration of mini-batch gradient descent, in range (0, 1].",
+  final val subsamplingRate = new DoubleParam(this, "subsamplingRate", "(For online optimizer)" +
+    " Fraction of the corpus to be sampled and used in each iteration of mini-batch" +
+    " gradient descent, in range (0, 1].",
     ParamValidators.inRange(0.0, 1.0, lowerInclusive = false, upperInclusive = true))
 
   /** @group getParam */
@@ -248,6 +256,8 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
   def getSubsamplingRate: Double = $(subsamplingRate)
 
   /**
+   * For Online optimizer only (currently): [[optimizer]] = "online".
+   *
    * Indicates whether the docConcentration (Dirichlet parameter for
    * document-topic distribution) will be optimized during training.
    * Setting this to true will make the model more expressive and fit the training data better.
@@ -257,15 +267,17 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    */
   @Since("1.6.0")
   final val optimizeDocConcentration = new BooleanParam(this, "optimizeDocConcentration",
-    "Indicates whether the docConcentration (Dirichlet parameter for document-topic" +
-      " distribution) will be optimized during training.")
+    "(For online optimizer only, currently) Indicates whether the docConcentration" +
+      " (Dirichlet parameter for document-topic distribution) will be optimized during training.")
 
   /** @group expertGetParam */
   @Since("1.6.0")
   def getOptimizeDocConcentration: Boolean = $(optimizeDocConcentration)
 
   /**
-   * For EM optimizer, if using checkpointing, this indicates whether to keep the last
+   * For EM optimizer only: [[optimizer]] = "em".
+   *
+   * If using checkpointing, this indicates whether to keep the last
    * checkpoint. If false, then the checkpoint will be deleted. Deleting the checkpoint can
    * cause failures if a data partition is lost, so set this bit with care.
    * Note that checkpoints will be cleaned up via reference counting, regardless.
@@ -279,7 +291,7 @@ private[clustering] trait LDAParams extends Params with HasFeaturesCol with HasM
    */
   @Since("2.0.0")
   final val keepLastCheckpoint = new BooleanParam(this, "keepLastCheckpoint",
-    "For EM optimizer, if using checkpointing, this indicates whether to keep the last" +
+    "(For EM optimizer) If using checkpointing, this indicates whether to keep the last" +
       " checkpoint. If false, then the checkpoint will be deleted. Deleting the checkpoint can" +
       " cause failures if a data partition is lost, so set this bit with care.")
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
index cef7c643d7..8eb2984f7b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -41,6 +41,18 @@ import org.apache.spark.sql.functions._
  * [[http://en.wikipedia.org/wiki/Gradient_boosting Gradient-Boosted Trees (GBTs)]]
  * learning algorithm for regression.
  * It supports both continuous and categorical features.
+ *
+ * The implementation is based upon: J.H. Friedman. "Stochastic Gradient Boosting." 1999.
+ *
+ * Notes on Gradient Boosting vs. TreeBoost:
+ *  - This implementation is for Stochastic Gradient Boosting, not for TreeBoost.
+ *  - Both algorithms learn tree ensembles by minimizing loss functions.
+ *  - TreeBoost (Friedman, 1999) additionally modifies the outputs at tree leaf nodes
+ *    based on the loss function, whereas the original gradient boosting method does not.
+ *     - When the loss is SquaredError, these methods give the same result, but they could differ
+ *       for other loss functions.
+ *  - We expect to implement TreeBoost in the future:
+ *    [https://issues.apache.org/jira/browse/SPARK-4240]
  */
 @Since("1.4.0")
 @Experimental
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
index a40d3731cb..05bf64591b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
@@ -165,7 +165,11 @@ class GeneralizedLinearRegression @Since("2.0.0") (@Since("2.0.0") override val
   setDefault(tol -> 1E-6)
 
   /**
-   * Sets the regularization parameter.
+   * Sets the regularization parameter for L2 regularization.
+   * The regularization term is
+   * {{{
+   *   0.5 * regParam * L2norm(coefficients)^2
+   * }}}
    * Default is 0.0.
    * @group setParam
    */
@@ -772,7 +776,7 @@ object GeneralizedLinearRegressionModel extends MLReadable[GeneralizedLinearRegr
  * :: Experimental ::
  * Summarizing Generalized Linear regression Fits.
  *
- * @param predictions predictions outputted by the model's `transform` method
+ * @param predictions predictions output by the model's `transform` method
  * @param predictionCol field in "predictions" which gives the prediction value of each instance
  * @param model the model that should be summarized
  * @param diagInvAtWA diagonal of matrix (A^T * W * A)^-1 in the last iteration
@@ -933,6 +937,9 @@ class GeneralizedLinearRegressionSummary private[regression] (
 
   /**
    * Standard error of estimated coefficients and intercept.
+   *
+   * If [[GeneralizedLinearRegression.fitIntercept]] is set to true,
+   * then the last element returned corresponds to the intercept.
    */
   @Since("2.0.0")
   lazy val coefficientStandardErrors: Array[Double] = {
@@ -941,6 +948,9 @@ class GeneralizedLinearRegressionSummary private[regression] (
 
   /**
    * T-statistic of estimated coefficients and intercept.
+   *
+   * If [[GeneralizedLinearRegression.fitIntercept]] is set to true,
+   * then the last element returned corresponds to the intercept.
    */
   @Since("2.0.0")
   lazy val tValues: Array[Double] = {
@@ -954,6 +964,9 @@ class GeneralizedLinearRegressionSummary private[regression] (
 
   /**
    * Two-sided p-value of estimated coefficients and intercept.
+   *
+   * If [[GeneralizedLinearRegression.fitIntercept]] is set to true,
+   * then the last element returned corresponds to the intercept.
    */
   @Since("2.0.0")
   lazy val pValues: Array[Double] = {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
index 9619e72a45..aacff4ea47 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
@@ -513,7 +513,7 @@ object LinearRegressionModel extends MLReadable[LinearRegressionModel] {
  * Linear regression training results. Currently, the training summary ignores the
  * training weights except for the objective trace.
  *
- * @param predictions predictions outputted by the model's `transform` method.
+ * @param predictions predictions output by the model's `transform` method.
  * @param objectiveHistory objective function (scaled loss + regularization) at each iteration.
  */
 @Since("1.5.0")
@@ -549,7 +549,7 @@ class LinearRegressionTrainingSummary private[regression] (
  * :: Experimental ::
  * Linear regression results evaluated on a dataset.
  *
- * @param predictions predictions outputted by the model's `transform` method.
+ * @param predictions predictions output by the model's `transform` method.
  * @param predictionCol Field in "predictions" which gives the predicted value of the label at
  *                      each instance.
  * @param labelCol Field in "predictions" which gives the true label of each instance.
@@ -655,8 +655,11 @@ class LinearRegressionSummary private[regression] (
 
   /**
    * Standard error of estimated coefficients and intercept.
-   *
    * This value is only available when using the "normal" solver.
+   *
+   * If [[LinearRegression.fitIntercept]] is set to true,
+   * then the last element returned corresponds to the intercept.
+   *
    * @see [[LinearRegression.solver]]
    */
   lazy val coefficientStandardErrors: Array[Double] = {
@@ -679,8 +682,11 @@ class LinearRegressionSummary private[regression] (
 
   /**
    * T-statistic of estimated coefficients and intercept.
-   *
    * This value is only available when using the "normal" solver.
+   *
+   * If [[LinearRegression.fitIntercept]] is set to true,
+   * then the last element returned corresponds to the intercept.
+   *
    * @see [[LinearRegression.solver]]
    */
   lazy val tValues: Array[Double] = {
@@ -699,8 +705,11 @@ class LinearRegressionSummary private[regression] (
 
   /**
    * Two-sided p-value of estimated coefficients and intercept.
-   *
    * This value is only available when using the "normal" solver.
+   *
+   * If [[LinearRegression.fitIntercept]] is set to true,
+   * then the last element returned corresponds to the intercept.
+   *
    * @see [[LinearRegression.solver]]
    */
   lazy val pValues: Array[Double] = {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
index d365655674..b6334762c7 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/GradientBoostedTrees.scala
@@ -30,22 +30,6 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.storage.StorageLevel
 
 
-/**
- * A package that implements
- * [[http://en.wikipedia.org/wiki/Gradient_boosting  Stochastic Gradient Boosting]]
- * for regression and binary classification.
- *
- * The implementation is based upon:
- *   J.H. Friedman.  "Stochastic Gradient Boosting."  1999.
- *
- * Notes on Gradient Boosting vs. TreeBoost:
- *  - This implementation is for Stochastic Gradient Boosting, not for TreeBoost.
- *  - Both algorithms learn tree ensembles by minimizing loss functions.
- *  - TreeBoost (Friedman, 1999) additionally modifies the outputs at tree leaf nodes
- *    based on the loss function, whereas the original gradient boosting method does not.
- *     - When the loss is SquaredError, these methods give the same result, but they could differ
- *       for other loss functions.
- */
 private[spark] object GradientBoostedTrees extends Logging {
 
   /**
diff --git a/python/pyspark/ml/classification.py b/python/pyspark/ml/classification.py
index d98919b3c6..e64c7a392b 100644
--- a/python/pyspark/ml/classification.py
+++ b/python/pyspark/ml/classification.py
@@ -291,7 +291,7 @@ class LogisticRegressionSummary(JavaCallable):
     @since("2.0.0")
     def probabilityCol(self):
         """
-        Field in "predictions" which gives the calibrated probability
+        Field in "predictions" which gives the probability
         of each class as a vector.
         """
         return self._call_java("probabilityCol")
diff --git a/python/pyspark/ml/regression.py b/python/pyspark/ml/regression.py
index f6c5d130dd..1c18df3b27 100644
--- a/python/pyspark/ml/regression.py
+++ b/python/pyspark/ml/regression.py
@@ -331,6 +331,9 @@ class LinearRegressionSummary(JavaCallable):
         Standard error of estimated coefficients and intercept.
         This value is only available when using the "normal" solver.
 
+        If :py:attr:`LinearRegression.fitIntercept` is set to True,
+        then the last element returned corresponds to the intercept.
+
         .. seealso:: :py:attr:`LinearRegression.solver`
         """
         return self._call_java("coefficientStandardErrors")
@@ -342,6 +345,9 @@ class LinearRegressionSummary(JavaCallable):
         T-statistic of estimated coefficients and intercept.
         This value is only available when using the "normal" solver.
 
+        If :py:attr:`LinearRegression.fitIntercept` is set to True,
+        then the last element returned corresponds to the intercept.
+
         .. seealso:: :py:attr:`LinearRegression.solver`
         """
         return self._call_java("tValues")
@@ -353,6 +359,9 @@ class LinearRegressionSummary(JavaCallable):
         Two-sided p-value of estimated coefficients and intercept.
         This value is only available when using the "normal" solver.
 
+        If :py:attr:`LinearRegression.fitIntercept` is set to True,
+        then the last element returned corresponds to the intercept.
+
         .. seealso:: :py:attr:`LinearRegression.solver`
         """
         return self._call_java("pValues")
diff --git a/python/pyspark/ml/tuning.py b/python/pyspark/ml/tuning.py
index da00f317b3..ea8c61b7ef 100644
--- a/python/pyspark/ml/tuning.py
+++ b/python/pyspark/ml/tuning.py
@@ -588,6 +588,8 @@ class TrainValidationSplit(Estimator, ValidatorParams, MLReadable, MLWritable):
 class TrainValidationSplitModel(Model, ValidatorParams, MLReadable, MLWritable):
     """
     Model from train validation split.
+
+    .. versionadded:: 2.0.0
     """
 
     def __init__(self, bestModel):
diff --git a/python/pyspark/ml/wrapper.py b/python/pyspark/ml/wrapper.py
index a2cf2296fb..bbeb6cfe6f 100644
--- a/python/pyspark/ml/wrapper.py
+++ b/python/pyspark/ml/wrapper.py
@@ -249,7 +249,7 @@ class JavaModel(Model, JavaCallable, JavaTransformer):
         """
         Initialize this instance with a Java model object.
         Subclasses should call this constructor, initialize params,
-        and then call _transformer_params_from_java.
+        and then call _transfer_params_from_java.
 
         This instance can be instantiated without specifying java_model,
         it will be assigned after that, but this scenario only used by
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
index 5bc0034cb0..223122300d 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala
@@ -154,6 +154,8 @@ object functions {
   /**
    * Aggregate function: returns the approximate number of distinct items in a group.
    *
+   * @param rsd maximum estimation error allowed (default = 0.05)
+   *
    * @group agg_funcs
    * @since 1.3.0
    */
@@ -164,6 +166,8 @@ object functions {
   /**
    * Aggregate function: returns the approximate number of distinct items in a group.
    *
+   * @param rsd maximum estimation error allowed (default = 0.05)
+   *
    * @group agg_funcs
    * @since 1.3.0
    */
-- 
cgit v1.2.3


From a9b8b655b25f4ed519037faaf7601a3d9842547f Mon Sep 17 00:00:00 2001
From: "wm624@hotmail.com" <wm624@hotmail.com>
Date: Sat, 9 Apr 2016 09:57:07 +0200
Subject: [SPARK-14392][ML] CountVectorizer Estimator should include binary
 toggle Param

## What changes were proposed in this pull request?

CountVectorizerModel has a binary toggle param. This PR is to add binary toggle param for estimator CountVectorizer. As discussed in the JIRA, instead of adding a param into CountVerctorizer, I moved the binary param to CountVectorizerParams. Therefore, the estimator inherits the binary param.

## How was this patch tested?

Add a new test case, which fits the model with binary flag set to true and then check the trained model's all non-zero counts is set to 1.0.

All tests in CounterVectorizerSuite.scala are passed.

Author: wm624@hotmail.com <wm624@hotmail.com>

Closes #12200 from wangmiao1981/binary_param.
---
 .../apache/spark/ml/feature/CountVectorizer.scala  | 35 +++++++++++-----------
 .../spark/ml/feature/CountVectorizerSuite.scala    | 19 ++++++++++--
 2 files changed, 34 insertions(+), 20 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
index 5694b3890f..f1be971a6a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
@@ -100,6 +100,21 @@ private[feature] trait CountVectorizerParams extends Params with HasInputCol wit
 
   /** @group getParam */
   def getMinTF: Double = $(minTF)
+
+  /**
+   * Binary toggle to control the output vector values.
+   * If True, all nonzero counts (after minTF filter applied) are set to 1. This is useful for
+   * discrete probabilistic models that model binary events rather than integer counts.
+   * Default: false
+   * @group param
+   */
+  val binary: BooleanParam =
+    new BooleanParam(this, "binary", "If True, all non zero counts are set to 1.")
+
+  /** @group getParam */
+  def getBinary: Boolean = $(binary)
+
+  setDefault(binary -> false)
 }
 
 /**
@@ -127,6 +142,9 @@ class CountVectorizer(override val uid: String)
   /** @group setParam */
   def setMinTF(value: Double): this.type = set(minTF, value)
 
+  /** @group setParam */
+  def setBinary(value: Boolean): this.type = set(binary, value)
+
   setDefault(vocabSize -> (1 << 18), minDF -> 1)
 
   override def fit(dataset: DataFrame): CountVectorizerModel = {
@@ -206,26 +224,9 @@ class CountVectorizerModel(override val uid: String, val vocabulary: Array[Strin
   /** @group setParam */
   def setMinTF(value: Double): this.type = set(minTF, value)
 
-  /**
-   * Binary toggle to control the output vector values.
-   * If True, all nonzero counts (after minTF filter applied) are set to 1. This is useful for
-   * discrete probabilistic models that model binary events rather than integer counts.
-   * Default: false
-   * @group param
-   */
-  val binary: BooleanParam =
-    new BooleanParam(this, "binary", "If True, all non zero counts are set to 1. " +
-      "This is useful for discrete probabilistic models that model binary events rather " +
-      "than integer counts")
-
-  /** @group getParam */
-  def getBinary: Boolean = $(binary)
-
   /** @group setParam */
   def setBinary(value: Boolean): this.type = set(binary, value)
 
-  setDefault(binary -> false)
-
   /** Dictionary created from [[vocabulary]] and its indices, broadcast once for [[transform()]] */
   private var broadcastDict: Option[Broadcast[Map[String, Int]]] = None
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala
index 04f165c5f1..ff0de06e27 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala
@@ -168,21 +168,34 @@ class CountVectorizerSuite extends SparkFunSuite with MLlibTestSparkContext
     }
   }
 
-  test("CountVectorizerModel with binary") {
+  test("CountVectorizerModel and CountVectorizer with binary") {
     val df = sqlContext.createDataFrame(Seq(
-      (0, split("a a a b b c"), Vectors.sparse(4, Seq((0, 1.0), (1, 1.0), (2, 1.0)))),
+      (0, split("a a a a b b b b c d"),
+      Vectors.sparse(4, Seq((0, 1.0), (1, 1.0), (2, 1.0), (3, 1.0)))),
       (1, split("c c c"), Vectors.sparse(4, Seq((2, 1.0)))),
       (2, split("a"), Vectors.sparse(4, Seq((0, 1.0))))
     )).toDF("id", "words", "expected")
 
-    val cv = new CountVectorizerModel(Array("a", "b", "c", "d"))
+    // CountVectorizer test
+    val cv = new CountVectorizer()
       .setInputCol("words")
       .setOutputCol("features")
       .setBinary(true)
+      .fit(df)
     cv.transform(df).select("features", "expected").collect().foreach {
       case Row(features: Vector, expected: Vector) =>
         assert(features ~== expected absTol 1e-14)
     }
+
+    // CountVectorizerModel test
+    val cv2 = new CountVectorizerModel(cv.vocabulary)
+      .setInputCol("words")
+      .setOutputCol("features")
+      .setBinary(true)
+    cv2.transform(df).select("features", "expected").collect().foreach {
+      case Row(features: Vector, expected: Vector) =>
+        assert(features ~== expected absTol 1e-14)
+    }
   }
 
   test("CountVectorizer read/write") {
-- 
cgit v1.2.3


From f4344582ba28983bf3892d08e11236f090f5bf92 Mon Sep 17 00:00:00 2001
From: fwang1 <desperado.wf@gmail.com>
Date: Sun, 10 Apr 2016 01:13:25 -0700
Subject: [SPARK-14497][ML] Use top instead of sortBy() to get top N frequent
 words as dict in ConutVectorizer

## What changes were proposed in this pull request?

Replace sortBy() with top() to calculate the top N frequent words as dictionary.

## How was this patch tested?
existing unit tests.  The terms with same TF would be sorted in descending order. The test would fail if hardcode the terms with same TF the dictionary like "c", "d"...

Author: fwang1 <desperado.wf@gmail.com>

Closes #12265 from lionelfeng/master.
---
 .../org/apache/spark/ml/feature/CountVectorizer.scala      | 14 ++++----------
 .../org/apache/spark/ml/feature/CountVectorizerSuite.scala |  7 ++++---
 2 files changed, 8 insertions(+), 13 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
index f1be971a6a..00abbbe29c 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
@@ -170,16 +170,10 @@ class CountVectorizer(override val uid: String)
       (word, count)
     }.cache()
     val fullVocabSize = wordCounts.count()
-    val vocab: Array[String] = {
-      val tmpSortedWC: Array[(String, Long)] = if (fullVocabSize <= vocSize) {
-        // Use all terms
-        wordCounts.collect().sortBy(-_._2)
-      } else {
-        // Sort terms to select vocab
-        wordCounts.sortBy(_._2, ascending = false).take(vocSize)
-      }
-      tmpSortedWC.map(_._1)
-    }
+
+    val vocab = wordCounts
+      .top(math.min(fullVocabSize, vocSize).toInt)(Ordering.by(_._2))
+      .map(_._1)
 
     require(vocab.length > 0, "The vocabulary size should be > 0. Lower minDF as necessary.")
     copyValues(new CountVectorizerModel(uid, vocab).setParent(this))
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala
index ff0de06e27..7641e3b8cf 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/CountVectorizerSuite.scala
@@ -59,14 +59,15 @@ class CountVectorizerSuite extends SparkFunSuite with MLlibTestSparkContext
       (0, split("a b c d e"),
         Vectors.sparse(5, Seq((0, 1.0), (1, 1.0), (2, 1.0), (3, 1.0), (4, 1.0)))),
       (1, split("a a a a a a"), Vectors.sparse(5, Seq((0, 6.0)))),
-      (2, split("c"), Vectors.sparse(5, Seq((2, 1.0)))),
-      (3, split("b b b b b"), Vectors.sparse(5, Seq((1, 5.0)))))
+      (2, split("c c"), Vectors.sparse(5, Seq((2, 2.0)))),
+      (3, split("d"), Vectors.sparse(5, Seq((3, 1.0)))),
+      (4, split("b b b b b"), Vectors.sparse(5, Seq((1, 5.0)))))
     ).toDF("id", "words", "expected")
     val cv = new CountVectorizer()
       .setInputCol("words")
       .setOutputCol("features")
       .fit(df)
-    assert(cv.vocabulary === Array("a", "b", "c", "d", "e"))
+    assert(cv.vocabulary.toSet === Set("a", "b", "c", "d", "e"))
 
     cv.transform(df).select("features", "expected").collect().foreach {
       case Row(features: Vector, expected: Vector) =>
-- 
cgit v1.2.3


From 1c751fcf488189e5176546fe0d00f560ffcf1cec Mon Sep 17 00:00:00 2001
From: Xiangrui Meng <meng@databricks.com>
Date: Mon, 11 Apr 2016 09:28:28 -0700
Subject: [SPARK-14500] [ML] Accept Dataset[_] instead of DataFrame in MLlib
 APIs

## What changes were proposed in this pull request?

This PR updates MLlib APIs to accept `Dataset[_]` as input where `DataFrame` was the input type. This PR doesn't change the output type. In Java, `Dataset[_]` maps to `Dataset<?>`, which includes `Dataset<Row>`. Some implementations were changed in order to return `DataFrame`. Tests and examples were updated. Note that this is a breaking change for subclasses of Transformer/Estimator.

Lol, we don't have to rename the input argument, which has been `dataset` since Spark 1.2.

TODOs:
- [x] update MiMaExcludes (seems all covered by explicit filters from SPARK-13920)
- [x] Python
- [x] add a new test to accept Dataset[LabeledPoint]
- [x] remove unused imports of Dataset

## How was this patch tested?

Exiting unit tests with some modifications.

cc: rxin jkbradley

Author: Xiangrui Meng <meng@databricks.com>

Closes #12274 from mengxr/SPARK-14500.
---
 .../spark/examples/ml/JavaDeveloperApiExample.java |  2 +-
 .../spark/examples/ml/DeveloperApiExample.scala    |  4 ++--
 .../main/scala/org/apache/spark/ml/Estimator.scala | 16 +++++++++------
 .../main/scala/org/apache/spark/ml/Pipeline.scala  | 12 +++++------
 .../main/scala/org/apache/spark/ml/Predictor.scala | 14 ++++++-------
 .../scala/org/apache/spark/ml/Transformer.scala    | 15 ++++++++------
 .../spark/ml/classification/Classifier.scala       |  6 +++---
 .../ml/classification/DecisionTreeClassifier.scala |  4 ++--
 .../spark/ml/classification/GBTClassifier.scala    |  6 +++---
 .../ml/classification/LogisticRegression.scala     |  8 ++++----
 .../MultilayerPerceptronClassifier.scala           |  4 ++--
 .../spark/ml/classification/NaiveBayes.scala       |  4 ++--
 .../apache/spark/ml/classification/OneVsRest.scala | 10 ++++-----
 .../classification/ProbabilisticClassifier.scala   |  6 +++---
 .../ml/classification/RandomForestClassifier.scala |  6 +++---
 .../spark/ml/clustering/BisectingKMeans.scala      |  8 ++++----
 .../spark/ml/clustering/GaussianMixture.scala      |  6 +++---
 .../org/apache/spark/ml/clustering/KMeans.scala    | 14 ++++++-------
 .../scala/org/apache/spark/ml/clustering/LDA.scala | 24 +++++++++++-----------
 .../evaluation/BinaryClassificationEvaluator.scala |  6 +++---
 .../org/apache/spark/ml/evaluation/Evaluator.scala | 10 ++++-----
 .../MulticlassClassificationEvaluator.scala        |  6 +++---
 .../spark/ml/evaluation/RegressionEvaluator.scala  |  6 +++---
 .../org/apache/spark/ml/feature/Binarizer.scala    |  3 ++-
 .../org/apache/spark/ml/feature/Bucketizer.scala   |  3 ++-
 .../apache/spark/ml/feature/ChiSqSelector.scala    |  6 ++++--
 .../apache/spark/ml/feature/CountVectorizer.scala  |  8 +++++---
 .../org/apache/spark/ml/feature/HashingTF.scala    |  5 +++--
 .../scala/org/apache/spark/ml/feature/IDF.scala    |  6 ++++--
 .../org/apache/spark/ml/feature/Interaction.scala  |  6 +++---
 .../org/apache/spark/ml/feature/MaxAbsScaler.scala |  6 ++++--
 .../org/apache/spark/ml/feature/MinMaxScaler.scala |  6 ++++--
 .../apache/spark/ml/feature/OneHotEncoder.scala    |  5 +++--
 .../scala/org/apache/spark/ml/feature/PCA.scala    |  6 ++++--
 .../spark/ml/feature/QuantileDiscretizer.scala     | 13 +++++++-----
 .../org/apache/spark/ml/feature/RFormula.scala     | 18 ++++++++--------
 .../apache/spark/ml/feature/SQLTransformer.scala   |  6 +++---
 .../apache/spark/ml/feature/StandardScaler.scala   |  6 ++++--
 .../apache/spark/ml/feature/StopWordsRemover.scala |  5 +++--
 .../apache/spark/ml/feature/StringIndexer.scala    | 13 +++++++-----
 .../apache/spark/ml/feature/VectorAssembler.scala  |  7 ++++---
 .../apache/spark/ml/feature/VectorIndexer.scala    |  8 +++++---
 .../org/apache/spark/ml/feature/VectorSlicer.scala |  5 +++--
 .../org/apache/spark/ml/feature/Word2Vec.scala     |  8 +++++---
 .../spark/ml/r/AFTSurvivalRegressionWrapper.scala  |  4 ++--
 .../org/apache/spark/ml/r/KMeansWrapper.scala      |  4 ++--
 .../org/apache/spark/ml/r/NaiveBayesWrapper.scala  |  4 ++--
 .../org/apache/spark/ml/recommendation/ALS.scala   | 10 ++++-----
 .../ml/regression/AFTSurvivalRegression.scala      | 12 +++++------
 .../ml/regression/DecisionTreeRegressor.scala      | 11 +++++-----
 .../apache/spark/ml/regression/GBTRegressor.scala  |  6 +++---
 .../regression/GeneralizedLinearRegression.scala   |  4 ++--
 .../spark/ml/regression/IsotonicRegression.scala   | 12 +++++------
 .../spark/ml/regression/LinearRegression.scala     |  6 +++---
 .../ml/regression/RandomForestRegressor.scala      |  6 +++---
 .../apache/spark/ml/tuning/CrossValidator.scala    | 12 +++++------
 .../spark/ml/tuning/TrainValidationSplit.scala     | 11 +++++-----
 .../org/apache/spark/mllib/linalg/Vectors.scala    |  2 +-
 .../scala/org/apache/spark/ml/PipelineSuite.scala  | 12 ++++++++---
 .../classification/LogisticRegressionSuite.scala   |  4 ++--
 .../MultilayerPerceptronClassifierSuite.scala      |  4 ++--
 .../spark/ml/classification/NaiveBayesSuite.scala  |  4 ++--
 .../spark/ml/classification/OneVsRestSuite.scala   |  6 +++---
 .../spark/ml/clustering/BisectingKMeansSuite.scala |  4 ++--
 .../spark/ml/clustering/GaussianMixtureSuite.scala |  4 ++--
 .../apache/spark/ml/clustering/KMeansSuite.scala   |  4 ++--
 .../org/apache/spark/ml/clustering/LDASuite.scala  |  4 ++--
 .../org/apache/spark/ml/feature/NGramSuite.scala   |  4 ++--
 .../spark/ml/feature/StopWordsRemoverSuite.scala   |  4 ++--
 .../spark/ml/feature/StringIndexerSuite.scala      |  2 +-
 .../apache/spark/ml/feature/TokenizerSuite.scala   |  4 ++--
 .../GeneralizedLinearRegressionSuite.scala         |  8 ++++++++
 .../spark/ml/tuning/CrossValidatorSuite.scala      |  8 ++++----
 .../ml/tuning/TrainValidationSplitSuite.scala      |  6 +++---
 .../spark/ml/util/DefaultReadWriteTest.scala       |  4 ++--
 75 files changed, 296 insertions(+), 240 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java
index fbd8817669..0ba94786d4 100644
--- a/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java
+++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaDeveloperApiExample.java
@@ -146,7 +146,7 @@ class MyJavaLogisticRegression
 
   // This method is used by fit().
   // In Java, we have to make it public since Java does not understand Scala's protected modifier.
-  public MyJavaLogisticRegressionModel train(Dataset<Row> dataset) {
+  public MyJavaLogisticRegressionModel train(Dataset<?> dataset) {
     // Extract columns from data using helper method.
     JavaRDD<LabeledPoint> oldDataset = extractLabeledPoints(dataset).toJavaRDD();
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala
index c1f63c6a1d..8d127f9b35 100644
--- a/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/ml/DeveloperApiExample.scala
@@ -24,7 +24,7 @@ import org.apache.spark.ml.param.{IntParam, ParamMap}
 import org.apache.spark.ml.util.Identifiable
 import org.apache.spark.mllib.linalg.{BLAS, Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.sql.{DataFrame, Row, SQLContext}
+import org.apache.spark.sql.{DataFrame, Dataset, Row, SQLContext}
 
 /**
  * A simple example demonstrating how to write your own learning algorithm using Estimator,
@@ -120,7 +120,7 @@ private class MyLogisticRegression(override val uid: String)
   def setMaxIter(value: Int): this.type = set(maxIter, value)
 
   // This method is used by fit()
-  override protected def train(dataset: DataFrame): MyLogisticRegressionModel = {
+  override protected def train(dataset: Dataset[_]): MyLogisticRegressionModel = {
     // Extract columns from data using helper method.
     val oldDataset = extractLabeledPoints(dataset)
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala b/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala
index 57e416591d..1247882d6c 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Estimator.scala
@@ -19,9 +19,9 @@ package org.apache.spark.ml
 
 import scala.annotation.varargs
 
-import org.apache.spark.annotation.DeveloperApi
+import org.apache.spark.annotation.{DeveloperApi, Since}
 import org.apache.spark.ml.param.{ParamMap, ParamPair}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.Dataset
 
 /**
  * :: DeveloperApi ::
@@ -39,8 +39,9 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
    *                        Estimator's embedded ParamMap.
    * @return fitted model
    */
+  @Since("2.0.0")
   @varargs
-  def fit(dataset: DataFrame, firstParamPair: ParamPair[_], otherParamPairs: ParamPair[_]*): M = {
+  def fit(dataset: Dataset[_], firstParamPair: ParamPair[_], otherParamPairs: ParamPair[_]*): M = {
     val map = new ParamMap()
       .put(firstParamPair)
       .put(otherParamPairs: _*)
@@ -55,14 +56,16 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
    *                 These values override any specified in this Estimator's embedded ParamMap.
    * @return fitted model
    */
-  def fit(dataset: DataFrame, paramMap: ParamMap): M = {
+  @Since("2.0.0")
+  def fit(dataset: Dataset[_], paramMap: ParamMap): M = {
     copy(paramMap).fit(dataset)
   }
 
   /**
    * Fits a model to the input data.
    */
-  def fit(dataset: DataFrame): M
+  @Since("2.0.0")
+  def fit(dataset: Dataset[_]): M
 
   /**
    * Fits multiple models to the input data with multiple sets of parameters.
@@ -74,7 +77,8 @@ abstract class Estimator[M <: Model[M]] extends PipelineStage {
    *                  These values override any specified in this Estimator's embedded ParamMap.
    * @return fitted models, matching the input parameter maps
    */
-  def fit(dataset: DataFrame, paramMaps: Array[ParamMap]): Seq[M] = {
+  @Since("2.0.0")
+  def fit(dataset: Dataset[_], paramMaps: Array[ParamMap]): Seq[M] = {
     paramMaps.map(fit(dataset, _))
   }
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala b/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
index afefaaa883..82066726a0 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
@@ -31,7 +31,7 @@ import org.apache.spark.annotation.{DeveloperApi, Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.param.{Param, ParamMap, Params}
 import org.apache.spark.ml.util._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.StructType
 
 /**
@@ -123,8 +123,8 @@ class Pipeline @Since("1.4.0") (
    * @param dataset input dataset
    * @return fitted pipeline
    */
-  @Since("1.2.0")
-  override def fit(dataset: DataFrame): PipelineModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): PipelineModel = {
     transformSchema(dataset.schema, logging = true)
     val theStages = $(stages)
     // Search for the last estimator.
@@ -291,10 +291,10 @@ class PipelineModel private[ml] (
     this(uid, stages.asScala.toArray)
   }
 
-  @Since("1.2.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
-    stages.foldLeft(dataset)((cur, transformer) => transformer.transform(cur))
+    stages.foldLeft(dataset.toDF)((cur, transformer) => transformer.transform(cur))
   }
 
   @Since("1.2.0")
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala b/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
index d23ae6f794..81140d1f7b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
@@ -24,7 +24,7 @@ import org.apache.spark.ml.util.SchemaUtils
 import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DataType, DoubleType, StructType}
 
@@ -83,7 +83,7 @@ abstract class Predictor[
   /** @group setParam */
   def setPredictionCol(value: String): Learner = set(predictionCol, value).asInstanceOf[Learner]
 
-  override def fit(dataset: DataFrame): M = {
+  override def fit(dataset: Dataset[_]): M = {
     // This handles a few items such as schema validation.
     // Developers only need to implement train().
     transformSchema(dataset.schema, logging = true)
@@ -100,7 +100,7 @@ abstract class Predictor[
    * @param dataset  Training dataset
    * @return  Fitted model
    */
-  protected def train(dataset: DataFrame): M
+  protected def train(dataset: Dataset[_]): M
 
   /**
    * Returns the SQL DataType corresponding to the FeaturesType type parameter.
@@ -120,7 +120,7 @@ abstract class Predictor[
    * Extract [[labelCol]] and [[featuresCol]] from the given dataset,
    * and put it in an RDD with strong types.
    */
-  protected def extractLabeledPoints(dataset: DataFrame): RDD[LabeledPoint] = {
+  protected def extractLabeledPoints(dataset: Dataset[_]): RDD[LabeledPoint] = {
     dataset.select(col($(labelCol)).cast(DoubleType), col($(featuresCol))).rdd.map {
       case Row(label: Double, features: Vector) => LabeledPoint(label, features)
     }
@@ -171,18 +171,18 @@ abstract class PredictionModel[FeaturesType, M <: PredictionModel[FeaturesType,
    * @param dataset input dataset
    * @return transformed dataset with [[predictionCol]] of type [[Double]]
    */
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     if ($(predictionCol).nonEmpty) {
       transformImpl(dataset)
     } else {
       this.logWarning(s"$uid: Predictor.transform() was called as NOOP" +
         " since no output columns were set.")
-      dataset
+      dataset.toDF
     }
   }
 
-  protected def transformImpl(dataset: DataFrame): DataFrame = {
+  protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val predictUDF = udf { (features: Any) =>
       predict(features.asInstanceOf[FeaturesType])
     }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala b/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala
index 2538c0f477..a3a2b55adc 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala
@@ -19,11 +19,11 @@ package org.apache.spark.ml
 
 import scala.annotation.varargs
 
-import org.apache.spark.annotation.DeveloperApi
+import org.apache.spark.annotation.{DeveloperApi, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 
@@ -41,9 +41,10 @@ abstract class Transformer extends PipelineStage {
    * @param otherParamPairs other param pairs, overwrite embedded params
    * @return transformed dataset
    */
+  @Since("2.0.0")
   @varargs
   def transform(
-      dataset: DataFrame,
+      dataset: Dataset[_],
       firstParamPair: ParamPair[_],
       otherParamPairs: ParamPair[_]*): DataFrame = {
     val map = new ParamMap()
@@ -58,14 +59,16 @@ abstract class Transformer extends PipelineStage {
    * @param paramMap additional parameters, overwrite embedded params
    * @return transformed dataset
    */
-  def transform(dataset: DataFrame, paramMap: ParamMap): DataFrame = {
+  @Since("2.0.0")
+  def transform(dataset: Dataset[_], paramMap: ParamMap): DataFrame = {
     this.copy(paramMap).transform(dataset)
   }
 
   /**
    * Transforms the input dataset.
    */
-  def transform(dataset: DataFrame): DataFrame
+  @Since("2.0.0")
+  def transform(dataset: Dataset[_]): DataFrame
 
   override def copy(extra: ParamMap): Transformer
 }
@@ -113,7 +116,7 @@ abstract class UnaryTransformer[IN, OUT, T <: UnaryTransformer[IN, OUT, T]]
     StructType(outputFields)
   }
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     val transformUDF = udf(this.createTransformFunc, outputDataType)
     dataset.withColumn($(outputCol), transformUDF(dataset($(inputCol))))
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala
index 8186afc17a..473e801794 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala
@@ -22,7 +22,7 @@ import org.apache.spark.ml.{PredictionModel, Predictor, PredictorParams}
 import org.apache.spark.ml.param.shared.HasRawPredictionCol
 import org.apache.spark.ml.util.SchemaUtils
 import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DataType, StructType}
 
@@ -92,7 +92,7 @@ abstract class ClassificationModel[FeaturesType, M <: ClassificationModel[Featur
    * @param dataset input dataset
    * @return transformed dataset
    */
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
 
     // Output selected columns only.
@@ -123,7 +123,7 @@ abstract class ClassificationModel[FeaturesType, M <: ClassificationModel[Featur
       logWarning(s"$uid: ClassificationModel.transform() was called as NOOP" +
         " since no output columns were set.")
     }
-    outputData
+    outputData.toDF
   }
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
index 4525bf71f6..300ae4339c 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/DecisionTreeClassifier.scala
@@ -32,7 +32,7 @@ import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, Strategy => OldStrategy}
 import org.apache.spark.mllib.tree.model.{DecisionTreeModel => OldDecisionTreeModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 
 /**
@@ -82,7 +82,7 @@ final class DecisionTreeClassifier @Since("1.4.0") (
   @Since("1.6.0")
   override def setSeed(value: Long): this.type = super.setSeed(value)
 
-  override protected def train(dataset: DataFrame): DecisionTreeClassificationModel = {
+  override protected def train(dataset: Dataset[_]): DecisionTreeClassificationModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
     val numClasses: Int = MetadataUtils.getNumClasses(dataset.schema($(labelCol))) match {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
index a2150fbcc3..46e8b89d01 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -33,7 +33,7 @@ import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
 import org.apache.spark.mllib.tree.loss.{LogLoss => OldLogLoss, Loss => OldLoss}
 import org.apache.spark.mllib.tree.model.{GradientBoostedTreesModel => OldGBTModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 
 /**
@@ -149,7 +149,7 @@ final class GBTClassifier @Since("1.4.0") (
     }
   }
 
-  override protected def train(dataset: DataFrame): GBTClassificationModel = {
+  override protected def train(dataset: Dataset[_]): GBTClassificationModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
     val numClasses: Int = MetadataUtils.getNumClasses(dataset.schema($(labelCol))) match {
@@ -220,7 +220,7 @@ final class GBTClassificationModel private[ml](
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
 
-  override protected def transformImpl(dataset: DataFrame): DataFrame = {
+  override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sqlContext.sparkContext.broadcast(this)
     val predictUDF = udf { (features: Any) =>
       bcastModel.value.predict(features.asInstanceOf[Vector])
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
index 268c3e32c3..4a3fe5c663 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
@@ -36,7 +36,7 @@ import org.apache.spark.mllib.linalg.BLAS._
 import org.apache.spark.mllib.stat.MultivariateOnlineSummarizer
 import org.apache.spark.mllib.util.MLUtils
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.{col, lit}
 import org.apache.spark.sql.types.DoubleType
 import org.apache.spark.storage.StorageLevel
@@ -257,12 +257,12 @@ class LogisticRegression @Since("1.2.0") (
     this
   }
 
-  override protected[spark] def train(dataset: DataFrame): LogisticRegressionModel = {
+  override protected[spark] def train(dataset: Dataset[_]): LogisticRegressionModel = {
     val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE
     train(dataset, handlePersistence)
   }
 
-  protected[spark] def train(dataset: DataFrame, handlePersistence: Boolean):
+  protected[spark] def train(dataset: Dataset[_], handlePersistence: Boolean):
       LogisticRegressionModel = {
     val w = if ($(weightCol).isEmpty) lit(1.0) else col($(weightCol))
     val instances: RDD[Instance] =
@@ -544,7 +544,7 @@ class LogisticRegressionModel private[spark] (
    * @param dataset Test dataset to evaluate model on.
    */
   @Since("2.0.0")
-  def evaluate(dataset: DataFrame): LogisticRegressionSummary = {
+  def evaluate(dataset: Dataset[_]): LogisticRegressionSummary = {
     // Handle possible missing or invalid prediction columns
     val (summaryModel, probabilityColName) = findSummaryModelAndProbabilityCol()
     new BinaryLogisticRegressionSummary(summaryModel.transform(dataset),
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
index 79bb2a8855..9ff5252e4f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
@@ -29,7 +29,7 @@ import org.apache.spark.ml.param.shared.{HasMaxIter, HasSeed, HasStepSize, HasTo
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 /** Params for Multilayer Perceptron. */
 private[ml] trait MultilayerPerceptronParams extends PredictorParams
@@ -199,7 +199,7 @@ class MultilayerPerceptronClassifier @Since("1.5.0") (
    * @param dataset Training dataset
    * @return Fitted model
    */
-  override protected def train(dataset: DataFrame): MultilayerPerceptronClassificationModel = {
+  override protected def train(dataset: Dataset[_]): MultilayerPerceptronClassificationModel = {
     val myLayers = $(layers)
     val labels = myLayers.last
     val lpData = extractLabeledPoints(dataset)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/NaiveBayes.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/NaiveBayes.scala
index 483ef0d88c..267d63b51e 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/NaiveBayes.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/NaiveBayes.scala
@@ -29,7 +29,7 @@ import org.apache.spark.mllib.classification.{NaiveBayesModel => OldNaiveBayesMo
 import org.apache.spark.mllib.linalg._
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 /**
  * Params for Naive Bayes Classifiers.
@@ -101,7 +101,7 @@ class NaiveBayes @Since("1.5.0") (
   def setModelType(value: String): this.type = set(modelType, value)
   setDefault(modelType -> OldNaiveBayes.Multinomial)
 
-  override protected def train(dataset: DataFrame): NaiveBayesModel = {
+  override protected def train(dataset: Dataset[_]): NaiveBayesModel = {
     val oldDataset: RDD[LabeledPoint] = extractLabeledPoints(dataset)
     val oldModel = OldNaiveBayes.train(oldDataset, $(smoothing), $(modelType))
     NaiveBayesModel.fromOld(oldModel, this)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
index 263d54ce4d..4de1b877b0 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
@@ -33,7 +33,7 @@ import org.apache.spark.ml.attribute._
 import org.apache.spark.ml.param.{Param, ParamMap, ParamPair, Params}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.Vector
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 import org.apache.spark.storage.StorageLevel
@@ -140,8 +140,8 @@ final class OneVsRestModel private[ml] (
     validateAndTransformSchema(schema, fitting = false, getClassifier.featuresDataType)
   }
 
-  @Since("1.4.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     // Check schema
     transformSchema(dataset.schema, logging = true)
 
@@ -293,8 +293,8 @@ final class OneVsRest @Since("1.4.0") (
     validateAndTransformSchema(schema, fitting = true, getClassifier.featuresDataType)
   }
 
-  @Since("1.4.0")
-  override def fit(dataset: DataFrame): OneVsRestModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): OneVsRestModel = {
     transformSchema(dataset.schema)
 
     // determine number of classes either from metadata if provided, or via computation.
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala
index 865614aa5c..d00fee12b0 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala
@@ -21,7 +21,7 @@ import org.apache.spark.annotation.DeveloperApi
 import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util.SchemaUtils
 import org.apache.spark.mllib.linalg.{DenseVector, Vector, Vectors, VectorUDT}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DataType, StructType}
 
@@ -95,7 +95,7 @@ abstract class ProbabilisticClassificationModel[
    * @param dataset input dataset
    * @return transformed dataset
    */
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     if (isDefined(thresholds)) {
       require($(thresholds).length == numClasses, this.getClass.getSimpleName +
@@ -145,7 +145,7 @@ abstract class ProbabilisticClassificationModel[
       this.logWarning(s"$uid: ProbabilisticClassificationModel.transform() was called as NOOP" +
         " since no output columns were set.")
     }
-    outputData
+    outputData.toDF
   }
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
index cb42532271..9d80b8eb68 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
@@ -31,7 +31,7 @@ import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
 import org.apache.spark.mllib.tree.model.{RandomForestModel => OldRandomForestModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 
 
@@ -98,7 +98,7 @@ final class RandomForestClassifier @Since("1.4.0") (
   override def setFeatureSubsetStrategy(value: String): this.type =
     super.setFeatureSubsetStrategy(value)
 
-  override protected def train(dataset: DataFrame): RandomForestClassificationModel = {
+  override protected def train(dataset: Dataset[_]): RandomForestClassificationModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
     val numClasses: Int = MetadataUtils.getNumClasses(dataset.schema($(labelCol))) match {
@@ -180,7 +180,7 @@ final class RandomForestClassificationModel private[ml] (
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
 
-  override protected def transformImpl(dataset: DataFrame): DataFrame = {
+  override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sqlContext.sparkContext.broadcast(this)
     val predictUDF = udf { (features: Any) =>
       bcastModel.value.predict(features.asInstanceOf[Vector])
diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala
index 55f751c57f..6cc9117da3 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/BisectingKMeans.scala
@@ -27,7 +27,7 @@ import org.apache.spark.ml.util._
 import org.apache.spark.mllib.clustering.
   {BisectingKMeans => MLlibBisectingKMeans, BisectingKMeansModel => MLlibBisectingKMeansModel}
 import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.{col, udf}
 import org.apache.spark.sql.types.{IntegerType, StructType}
 
@@ -92,7 +92,7 @@ class BisectingKMeansModel private[ml] (
   }
 
   @Since("2.0.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val predictUDF = udf((vector: Vector) => predict(vector))
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
@@ -112,7 +112,7 @@ class BisectingKMeansModel private[ml] (
    * centers.
    */
   @Since("2.0.0")
-  def computeCost(dataset: DataFrame): Double = {
+  def computeCost(dataset: Dataset[_]): Double = {
     SchemaUtils.checkColumnType(dataset.schema, $(featuresCol), new VectorUDT)
     val data = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
     parentModel.computeCost(data)
@@ -215,7 +215,7 @@ class BisectingKMeans @Since("2.0.0") (
   def setMinDivisibleClusterSize(value: Double): this.type = set(minDivisibleClusterSize, value)
 
   @Since("2.0.0")
-  override def fit(dataset: DataFrame): BisectingKMeansModel = {
+  override def fit(dataset: Dataset[_]): BisectingKMeansModel = {
     val rdd = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
 
     val bkm = new MLlibBisectingKMeans()
diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala
index 120bf3cf9d..ead8ad7806 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala
@@ -27,7 +27,7 @@ import org.apache.spark.ml.util._
 import org.apache.spark.mllib.clustering.{GaussianMixture => MLlibGM, GaussianMixtureModel => MLlibGMModel}
 import org.apache.spark.mllib.linalg._
 import org.apache.spark.mllib.stat.distribution.MultivariateGaussian
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.{col, udf}
 import org.apache.spark.sql.types.{IntegerType, StructType}
 
@@ -80,7 +80,7 @@ class GaussianMixtureModel private[ml] (
   }
 
   @Since("2.0.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val predUDF = udf((vector: Vector) => predict(vector))
     val probUDF = udf((vector: Vector) => predictProbability(vector))
     dataset.withColumn($(predictionCol), predUDF(col($(featuresCol))))
@@ -238,7 +238,7 @@ class GaussianMixture @Since("2.0.0") (
   def setSeed(value: Long): this.type = set(seed, value)
 
   @Since("2.0.0")
-  override def fit(dataset: DataFrame): GaussianMixtureModel = {
+  override def fit(dataset: Dataset[_]): GaussianMixtureModel = {
     val rdd = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
 
     val algo = new MLlibGM()
diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
index a8beef8b12..d716bc6887 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
@@ -27,7 +27,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.clustering.{KMeans => MLlibKMeans, KMeansModel => MLlibKMeansModel}
 import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.{col, udf}
 import org.apache.spark.sql.types.{IntegerType, StructType}
 
@@ -105,8 +105,8 @@ class KMeansModel private[ml] (
     copyValues(copied, extra)
   }
 
-  @Since("1.5.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val predictUDF = udf((vector: Vector) => predict(vector))
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
@@ -126,8 +126,8 @@ class KMeansModel private[ml] (
    * model on the given data.
    */
   // TODO: Replace the temp fix when we have proper evaluators defined for clustering.
-  @Since("1.6.0")
-  def computeCost(dataset: DataFrame): Double = {
+  @Since("2.0.0")
+  def computeCost(dataset: Dataset[_]): Double = {
     SchemaUtils.checkColumnType(dataset.schema, $(featuresCol), new VectorUDT)
     val data = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
     parentModel.computeCost(data)
@@ -254,8 +254,8 @@ class KMeans @Since("1.5.0") (
   @Since("1.5.0")
   def setSeed(value: Long): this.type = set(seed, value)
 
-  @Since("1.5.0")
-  override def fit(dataset: DataFrame): KMeansModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): KMeansModel = {
     val rdd = dataset.select(col($(featuresCol))).rdd.map { case Row(point: Vector) => point }
 
     val algo = new MLlibKMeans()
diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
index 89a7a4ccf6..c57ceba4a9 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/LDA.scala
@@ -32,7 +32,7 @@ import org.apache.spark.mllib.clustering.{DistributedLDAModel => OldDistributedL
 import org.apache.spark.mllib.impl.PeriodicCheckpointer
 import org.apache.spark.mllib.linalg.{Matrix, Vector, Vectors, VectorUDT}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row, SQLContext}
+import org.apache.spark.sql.{DataFrame, Dataset, Row, SQLContext}
 import org.apache.spark.sql.functions.{col, monotonicallyIncreasingId, udf}
 import org.apache.spark.sql.types.StructType
 
@@ -402,15 +402,15 @@ sealed abstract class LDAModel private[ml] (
    *          is set to "em"), this involves collecting a large [[topicsMatrix]] to the driver.
    *          This implementation may be changed in the future.
    */
-  @Since("1.6.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     if ($(topicDistributionCol).nonEmpty) {
       val t = udf(oldLocalModel.getTopicDistributionMethod(sqlContext.sparkContext))
-      dataset.withColumn($(topicDistributionCol), t(col($(featuresCol))))
+      dataset.withColumn($(topicDistributionCol), t(col($(featuresCol)))).toDF
     } else {
       logWarning("LDAModel.transform was called without any output columns. Set an output column" +
         " such as topicDistributionCol to produce results.")
-      dataset
+      dataset.toDF
     }
   }
 
@@ -455,8 +455,8 @@ sealed abstract class LDAModel private[ml] (
    * @param dataset  test corpus to use for calculating log likelihood
    * @return variational lower bound on the log likelihood of the entire corpus
    */
-  @Since("1.6.0")
-  def logLikelihood(dataset: DataFrame): Double = {
+  @Since("2.0.0")
+  def logLikelihood(dataset: Dataset[_]): Double = {
     val oldDataset = LDA.getOldDataset(dataset, $(featuresCol))
     oldLocalModel.logLikelihood(oldDataset)
   }
@@ -472,8 +472,8 @@ sealed abstract class LDAModel private[ml] (
    * @param dataset test corpus to use for calculating perplexity
    * @return Variational upper bound on log perplexity per token.
    */
-  @Since("1.6.0")
-  def logPerplexity(dataset: DataFrame): Double = {
+  @Since("2.0.0")
+  def logPerplexity(dataset: Dataset[_]): Double = {
     val oldDataset = LDA.getOldDataset(dataset, $(featuresCol))
     oldLocalModel.logPerplexity(oldDataset)
   }
@@ -840,8 +840,8 @@ class LDA @Since("1.6.0") (
   @Since("1.6.0")
   override def copy(extra: ParamMap): LDA = defaultCopy(extra)
 
-  @Since("1.6.0")
-  override def fit(dataset: DataFrame): LDAModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): LDAModel = {
     transformSchema(dataset.schema, logging = true)
     val oldLDA = new OldLDA()
       .setK($(k))
@@ -873,7 +873,7 @@ class LDA @Since("1.6.0") (
 private[clustering] object LDA extends DefaultParamsReadable[LDA] {
 
   /** Get dataset for spark.mllib LDA */
-  def getOldDataset(dataset: DataFrame, featuresCol: String): RDD[(Long, Vector)] = {
+  def getOldDataset(dataset: Dataset[_], featuresCol: String): RDD[(Long, Vector)] = {
     dataset
       .withColumn("docId", monotonicallyIncreasingId())
       .select("docId", featuresCol)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala b/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala
index 337ffbe90f..bde8c275fd 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala
@@ -23,7 +23,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util.{DefaultParamsReadable, DefaultParamsWritable, Identifiable, SchemaUtils}
 import org.apache.spark.mllib.evaluation.BinaryClassificationMetrics
 import org.apache.spark.mllib.linalg.{Vector, VectorUDT}
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{Dataset, Row}
 import org.apache.spark.sql.types.DoubleType
 
 /**
@@ -69,8 +69,8 @@ class BinaryClassificationEvaluator @Since("1.4.0") (@Since("1.4.0") override va
 
   setDefault(metricName -> "areaUnderROC")
 
-  @Since("1.2.0")
-  override def evaluate(dataset: DataFrame): Double = {
+  @Since("2.0.0")
+  override def evaluate(dataset: Dataset[_]): Double = {
     val schema = dataset.schema
     SchemaUtils.checkColumnTypes(schema, $(rawPredictionCol), Seq(DoubleType, new VectorUDT))
     SchemaUtils.checkColumnType(schema, $(labelCol), DoubleType)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/evaluation/Evaluator.scala b/mllib/src/main/scala/org/apache/spark/ml/evaluation/Evaluator.scala
index 0f22cca3a7..5f765c071b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/evaluation/Evaluator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/evaluation/Evaluator.scala
@@ -19,7 +19,7 @@ package org.apache.spark.ml.evaluation
 
 import org.apache.spark.annotation.{DeveloperApi, Since}
 import org.apache.spark.ml.param.{ParamMap, Params}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.Dataset
 
 /**
  * :: DeveloperApi ::
@@ -36,8 +36,8 @@ abstract class Evaluator extends Params {
    * @param paramMap parameter map that specifies the input columns and output metrics
    * @return metric
    */
-  @Since("1.5.0")
-  def evaluate(dataset: DataFrame, paramMap: ParamMap): Double = {
+  @Since("2.0.0")
+  def evaluate(dataset: Dataset[_], paramMap: ParamMap): Double = {
     this.copy(paramMap).evaluate(dataset)
   }
 
@@ -46,8 +46,8 @@ abstract class Evaluator extends Params {
    * @param dataset a dataset that contains labels/observations and predictions.
    * @return metric
    */
-  @Since("1.5.0")
-  def evaluate(dataset: DataFrame): Double
+  @Since("2.0.0")
+  def evaluate(dataset: Dataset[_]): Double
 
   /**
    * Indicates whether the metric returned by [[evaluate()]] should be maximized (true, default)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/evaluation/MulticlassClassificationEvaluator.scala b/mllib/src/main/scala/org/apache/spark/ml/evaluation/MulticlassClassificationEvaluator.scala
index 55ff44323a..3acfc221c9 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/evaluation/MulticlassClassificationEvaluator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/evaluation/MulticlassClassificationEvaluator.scala
@@ -22,7 +22,7 @@ import org.apache.spark.ml.param.{Param, ParamMap, ParamValidators}
 import org.apache.spark.ml.param.shared.{HasLabelCol, HasPredictionCol}
 import org.apache.spark.ml.util.{DefaultParamsReadable, DefaultParamsWritable, Identifiable, SchemaUtils}
 import org.apache.spark.mllib.evaluation.MulticlassMetrics
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{Dataset, Row}
 import org.apache.spark.sql.types.DoubleType
 
 /**
@@ -68,8 +68,8 @@ class MulticlassClassificationEvaluator @Since("1.5.0") (@Since("1.5.0") overrid
 
   setDefault(metricName -> "f1")
 
-  @Since("1.5.0")
-  override def evaluate(dataset: DataFrame): Double = {
+  @Since("2.0.0")
+  override def evaluate(dataset: Dataset[_]): Double = {
     val schema = dataset.schema
     SchemaUtils.checkColumnType(schema, $(predictionCol), DoubleType)
     SchemaUtils.checkColumnType(schema, $(labelCol), DoubleType)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala b/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala
index 9976d7ed43..4134e2dbc5 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala
@@ -22,7 +22,7 @@ import org.apache.spark.ml.param.{Param, ParamMap, ParamValidators}
 import org.apache.spark.ml.param.shared.{HasLabelCol, HasPredictionCol}
 import org.apache.spark.ml.util.{DefaultParamsReadable, DefaultParamsWritable, Identifiable}
 import org.apache.spark.mllib.evaluation.RegressionMetrics
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DoubleType, FloatType}
 
@@ -70,8 +70,8 @@ final class RegressionEvaluator @Since("1.4.0") (@Since("1.4.0") override val ui
 
   setDefault(metricName -> "rmse")
 
-  @Since("1.4.0")
-  override def evaluate(dataset: DataFrame): Double = {
+  @Since("2.0.0")
+  override def evaluate(dataset: Dataset[_]): Double = {
     val schema = dataset.schema
     val predictionColName = $(predictionCol)
     val predictionType = schema($(predictionCol)).dataType
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Binarizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Binarizer.scala
index 2f8e3a0371..898ac2cc89 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/Binarizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/Binarizer.scala
@@ -64,7 +64,8 @@ final class Binarizer(override val uid: String)
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val outputSchema = transformSchema(dataset.schema, logging = true)
     val schema = dataset.schema
     val inputType = schema($(inputCol)).dataType
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Bucketizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Bucketizer.scala
index 33abc7c99d..10e622ace6 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/Bucketizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/Bucketizer.scala
@@ -68,7 +68,8 @@ final class Bucketizer(override val uid: String)
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema)
     val bucketizer = udf { feature: Double =>
       Bucketizer.binarySearchForBuckets($(splits), feature)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/ChiSqSelector.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/ChiSqSelector.scala
index b9e9d56853..cfecae7e0b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/ChiSqSelector.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/ChiSqSelector.scala
@@ -77,7 +77,8 @@ final class ChiSqSelector(override val uid: String)
   /** @group setParam */
   def setLabelCol(value: String): this.type = set(labelCol, value)
 
-  override def fit(dataset: DataFrame): ChiSqSelectorModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): ChiSqSelectorModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(labelCol), $(featuresCol)).rdd.map {
       case Row(label: Double, features: Vector) =>
@@ -127,7 +128,8 @@ final class ChiSqSelectorModel private[ml] (
   /** @group setParam */
   def setLabelCol(value: String): this.type = set(labelCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val transformedSchema = transformSchema(dataset.schema, logging = true)
     val newField = transformedSchema.last
     val selector = udf { chiSqSelector.transform _ }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
index 00abbbe29c..922670a41b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizer.scala
@@ -26,7 +26,7 @@ import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{Vectors, VectorUDT}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 import org.apache.spark.util.collection.OpenHashMap
@@ -147,7 +147,8 @@ class CountVectorizer(override val uid: String)
 
   setDefault(vocabSize -> (1 << 18), minDF -> 1)
 
-  override def fit(dataset: DataFrame): CountVectorizerModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): CountVectorizerModel = {
     transformSchema(dataset.schema, logging = true)
     val vocSize = $(vocabSize)
     val input = dataset.select($(inputCol)).rdd.map(_.getAs[Seq[String]](0))
@@ -224,7 +225,8 @@ class CountVectorizerModel(override val uid: String, val vocabulary: Array[Strin
   /** Dictionary created from [[vocabulary]] and its indices, broadcast once for [[transform()]] */
   private var broadcastDict: Option[Broadcast[Map[String, Int]]] = None
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     if (broadcastDict.isEmpty) {
       val dict = vocabulary.zipWithIndex.toMap
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
index 0f7ae5a100..467ad73074 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
@@ -24,7 +24,7 @@ import org.apache.spark.ml.param.{BooleanParam, IntParam, ParamMap, ParamValidat
 import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.feature
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions.{col, udf}
 import org.apache.spark.sql.types.{ArrayType, StructType}
 
@@ -77,7 +77,8 @@ class HashingTF(override val uid: String)
   /** @group setParam */
   def setBinary(value: Boolean): this.type = set(binary, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val outputSchema = transformSchema(dataset.schema)
     val hashingTF = new feature.HashingTF($(numFeatures)).setBinary($(binary))
     val t = udf { terms: Seq[_] => hashingTF.transform(terms) }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/IDF.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/IDF.scala
index f36cf503a0..5075b78c98 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/IDF.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/IDF.scala
@@ -76,7 +76,8 @@ final class IDF(override val uid: String) extends Estimator[IDFModel] with IDFBa
   /** @group setParam */
   def setMinDocFreq(value: Int): this.type = set(minDocFreq, value)
 
-  override def fit(dataset: DataFrame): IDFModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): IDFModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(inputCol)).rdd.map { case Row(v: Vector) => v }
     val idf = new feature.IDF($(minDocFreq)).fit(input)
@@ -115,7 +116,8 @@ class IDFModel private[ml] (
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     val idf = udf { vec: Vector => idfModel.transform(vec) }
     dataset.withColumn($(outputCol), idf(col($(inputCol))))
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Interaction.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Interaction.scala
index d3fe6e528f..9ca34e9ae2 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/Interaction.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/Interaction.scala
@@ -27,7 +27,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.ml.Transformer
 import org.apache.spark.mllib.linalg.{Vector, Vectors, VectorUDT}
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 
@@ -68,8 +68,8 @@ class Interaction @Since("1.6.0") (override val uid: String) extends Transformer
     StructType(schema.fields :+ StructField($(outputCol), new VectorUDT, false))
   }
 
-  @Since("1.6.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val inputFeatures = $(inputCols).map(c => dataset.schema(c))
     val featureEncoders = getFeatureEncoders(inputFeatures)
     val featureAttrs = getFeatureAttrs(inputFeatures)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/MaxAbsScaler.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/MaxAbsScaler.scala
index 7de5a4d5d3..e9df600c8a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/MaxAbsScaler.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/MaxAbsScaler.scala
@@ -66,7 +66,8 @@ class MaxAbsScaler @Since("2.0.0") (override val uid: String)
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def fit(dataset: DataFrame): MaxAbsScalerModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): MaxAbsScalerModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(inputCol)).rdd.map { case Row(v: Vector) => v }
     val summary = Statistics.colStats(input)
@@ -111,7 +112,8 @@ class MaxAbsScalerModel private[ml] (
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     // TODO: this looks hack, we may have to handle sparse and dense vectors separately.
     val maxAbsUnzero = Vectors.dense(maxAbs.toArray.map(x => if (x == 0) 1 else x))
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/MinMaxScaler.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/MinMaxScaler.scala
index b13684a1cb..125becbb8a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/MinMaxScaler.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/MinMaxScaler.scala
@@ -103,7 +103,8 @@ class MinMaxScaler(override val uid: String)
   /** @group setParam */
   def setMax(value: Double): this.type = set(max, value)
 
-  override def fit(dataset: DataFrame): MinMaxScalerModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): MinMaxScalerModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(inputCol)).rdd.map { case Row(v: Vector) => v }
     val summary = Statistics.colStats(input)
@@ -154,7 +155,8 @@ class MinMaxScalerModel private[ml] (
   /** @group setParam */
   def setMax(value: Double): this.type = set(max, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val originalRange = (originalMax.toBreeze - originalMin.toBreeze).toArray
     val minArray = originalMin.toArray
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/OneHotEncoder.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/OneHotEncoder.scala
index 4f67042629..99357793db 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/OneHotEncoder.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/OneHotEncoder.scala
@@ -24,7 +24,7 @@ import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.Vectors
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions.{col, udf}
 import org.apache.spark.sql.types.{DoubleType, NumericType, StructType}
 
@@ -121,7 +121,8 @@ class OneHotEncoder(override val uid: String) extends Transformer
     StructType(outputFields)
   }
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     // schema transformation
     val inputColName = $(inputCol)
     val outputColName = $(outputCol)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/PCA.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/PCA.scala
index 305c3d187f..9cf722e121 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/PCA.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/PCA.scala
@@ -68,7 +68,8 @@ class PCA (override val uid: String) extends Estimator[PCAModel] with PCAParams
   /**
    * Computes a [[PCAModel]] that contains the principal components of the input vectors.
    */
-  override def fit(dataset: DataFrame): PCAModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): PCAModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(inputCol)).rdd.map { case Row(v: Vector) => v}
     val pca = new feature.PCA(k = $(k))
@@ -124,7 +125,8 @@ class PCAModel private[ml] (
    * NOTE: Vectors to be transformed must be the same length
    * as the source vectors given to [[PCA.fit()]].
    */
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     val pcaModel = new feature.PCAModel($(k), pc, explainedVariance)
     val pcaOp = udf { pcaModel.transform _ }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala
index e486e92c12..efe8b93d82 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala
@@ -23,10 +23,10 @@ import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml._
 import org.apache.spark.ml.attribute.NominalAttribute
-import org.apache.spark.ml.param.{IntParam, _}
+import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol, HasSeed}
 import org.apache.spark.ml.util._
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{Dataset, Row}
 import org.apache.spark.sql.types.{DoubleType, StructType}
 import org.apache.spark.util.random.XORShiftRandom
 
@@ -87,7 +87,8 @@ final class QuantileDiscretizer(override val uid: String)
     StructType(outputFields)
   }
 
-  override def fit(dataset: DataFrame): Bucketizer = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): Bucketizer = {
     val samples = QuantileDiscretizer
       .getSampledInput(dataset.select($(inputCol)), $(numBuckets), $(seed))
       .map { case Row(feature: Double) => feature }
@@ -112,13 +113,15 @@ object QuantileDiscretizer extends DefaultParamsReadable[QuantileDiscretizer] wi
   /**
    * Sampling from the given dataset to collect quantile statistics.
    */
-  private[feature] def getSampledInput(dataset: DataFrame, numBins: Int, seed: Long): Array[Row] = {
+  private[feature]
+  def getSampledInput(dataset: Dataset[_], numBins: Int, seed: Long): Array[Row] = {
     val totalSamples = dataset.count()
     require(totalSamples > 0,
       "QuantileDiscretizer requires non-empty input dataset but was given an empty input.")
     val requiredSamples = math.max(numBins * numBins, minSamplesRequired)
     val fraction = math.min(requiredSamples.toDouble / totalSamples, 1.0)
-    dataset.sample(withReplacement = false, fraction, new XORShiftRandom(seed).nextInt()).collect()
+    dataset.toDF.sample(withReplacement = false, fraction, new XORShiftRandom(seed).nextInt())
+      .collect()
   }
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/RFormula.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/RFormula.scala
index 12a76dbbfb..3ac6c77669 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/RFormula.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/RFormula.scala
@@ -29,7 +29,7 @@ import org.apache.spark.ml.param.{Param, ParamMap}
 import org.apache.spark.ml.param.shared.{HasFeaturesCol, HasLabelCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.VectorUDT
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types._
 
 /**
@@ -103,7 +103,8 @@ class RFormula(override val uid: String)
     RFormulaParser.parse($(formula)).hasIntercept
   }
 
-  override def fit(dataset: DataFrame): RFormulaModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): RFormulaModel = {
     require(isDefined(formula), "Formula must be defined first.")
     val parsedFormula = RFormulaParser.parse($(formula))
     val resolvedFormula = parsedFormula.resolve(dataset.schema)
@@ -204,7 +205,8 @@ class RFormulaModel private[feature](
     private[ml] val pipelineModel: PipelineModel)
   extends Model[RFormulaModel] with RFormulaBase with MLWritable {
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     checkCanTransform(dataset.schema)
     transformLabel(pipelineModel.transform(dataset))
   }
@@ -232,10 +234,10 @@ class RFormulaModel private[feature](
 
   override def toString: String = s"RFormulaModel($resolvedFormula) (uid=$uid)"
 
-  private def transformLabel(dataset: DataFrame): DataFrame = {
+  private def transformLabel(dataset: Dataset[_]): DataFrame = {
     val labelName = resolvedFormula.label
     if (hasLabelCol(dataset.schema)) {
-      dataset
+      dataset.toDF
     } else if (dataset.schema.exists(_.name == labelName)) {
       dataset.schema(labelName).dataType match {
         case _: NumericType | BooleanType =>
@@ -246,7 +248,7 @@ class RFormulaModel private[feature](
     } else {
       // Ignore the label field. This is a hack so that this transformer can also work on test
       // datasets in a Pipeline.
-      dataset
+      dataset.toDF
     }
   }
 
@@ -323,7 +325,7 @@ private class ColumnPruner(override val uid: String, val columnsToPrune: Set[Str
   def this(columnsToPrune: Set[String]) =
     this(Identifiable.randomUID("columnPruner"), columnsToPrune)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val columnsToKeep = dataset.columns.filter(!columnsToPrune.contains(_))
     dataset.select(columnsToKeep.map(dataset.col): _*)
   }
@@ -396,7 +398,7 @@ private class VectorAttributeRewriter(
   def this(vectorCol: String, prefixesToRewrite: Map[String, String]) =
     this(Identifiable.randomUID("vectorAttrRewriter"), vectorCol, prefixesToRewrite)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val metadata = {
       val group = AttributeGroup.fromStructField(dataset.schema(vectorCol))
       val attrs = group.attributes.get.map { attr =>
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala
index e0ca45b9a6..95fe942c6b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala
@@ -22,7 +22,7 @@ import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.ml.param.{Param, ParamMap}
 import org.apache.spark.ml.Transformer
 import org.apache.spark.ml.util._
-import org.apache.spark.sql.{DataFrame, Row, SQLContext}
+import org.apache.spark.sql.{DataFrame, Dataset, Row, SQLContext}
 import org.apache.spark.sql.types.StructType
 
 /**
@@ -63,8 +63,8 @@ class SQLTransformer @Since("1.6.0") (override val uid: String) extends Transfor
 
   private val tableIdentifier: String = "__THIS__"
 
-  @Since("1.6.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val tableName = Identifiable.randomUID(uid)
     dataset.registerTempTable(tableName)
     val realStatement = $(statement).replace(tableIdentifier, tableName)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/StandardScaler.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/StandardScaler.scala
index 26ee8e1bf1..118a6e3e6a 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/StandardScaler.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/StandardScaler.scala
@@ -85,7 +85,8 @@ class StandardScaler(override val uid: String) extends Estimator[StandardScalerM
   /** @group setParam */
   def setWithStd(value: Boolean): this.type = set(withStd, value)
 
-  override def fit(dataset: DataFrame): StandardScalerModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): StandardScalerModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(inputCol)).rdd.map { case Row(v: Vector) => v }
     val scaler = new feature.StandardScaler(withMean = $(withMean), withStd = $(withStd))
@@ -135,7 +136,8 @@ class StandardScalerModel private[ml] (
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     val scaler = new feature.StandardScalerModel(std, mean, $(withStd), $(withMean))
     val scale = udf { scaler.transform _ }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/StopWordsRemover.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/StopWordsRemover.scala
index 0a0e0b0960..b96bc48566 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/StopWordsRemover.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/StopWordsRemover.scala
@@ -22,7 +22,7 @@ import org.apache.spark.ml.Transformer
 import org.apache.spark.ml.param.{BooleanParam, ParamMap, StringArrayParam}
 import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions.{col, udf}
 import org.apache.spark.sql.types.{ArrayType, StringType, StructType}
 
@@ -125,7 +125,8 @@ class StopWordsRemover(override val uid: String)
 
   setDefault(stopWords -> StopWords.English, caseSensitive -> false)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val outputSchema = transformSchema(dataset.schema)
     val t = if ($(caseSensitive)) {
         val stopWordsSet = $(stopWords).toSet
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/StringIndexer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/StringIndexer.scala
index faa0f6f407..7e0d374f02 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/StringIndexer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/StringIndexer.scala
@@ -26,7 +26,7 @@ import org.apache.spark.ml.attribute.{Attribute, NominalAttribute}
 import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 import org.apache.spark.util.collection.OpenHashMap
@@ -80,7 +80,8 @@ class StringIndexer(override val uid: String) extends Estimator[StringIndexerMod
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
 
-  override def fit(dataset: DataFrame): StringIndexerModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): StringIndexerModel = {
     val counts = dataset.select(col($(inputCol)).cast(StringType))
       .rdd
       .map(_.getString(0))
@@ -144,11 +145,12 @@ class StringIndexerModel (
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     if (!dataset.schema.fieldNames.contains($(inputCol))) {
       logInfo(s"Input column ${$(inputCol)} does not exist during transformation. " +
         "Skip StringIndexerModel.")
-      return dataset
+      return dataset.toDF
     }
     validateAndTransformSchema(dataset.schema)
 
@@ -286,7 +288,8 @@ class IndexToString private[ml] (override val uid: String)
     StructType(outputFields)
   }
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val inputColSchema = dataset.schema($(inputCol))
     // If the labels array is empty use column metadata
     val values = if ($(labels).isEmpty) {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala
index 957e8e7a59..4d3e46e488 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala
@@ -27,7 +27,7 @@ import org.apache.spark.ml.param.ParamMap
 import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{Vector, Vectors, VectorUDT}
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 
@@ -47,10 +47,11 @@ class VectorAssembler(override val uid: String)
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     // Schema transformation.
     val schema = dataset.schema
-    lazy val first = dataset.first()
+    lazy val first = dataset.toDF.first()
     val attrs = $(inputCols).flatMap { c =>
       val field = schema(c)
       val index = schema.fieldIndex(c)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorIndexer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorIndexer.scala
index bf4aef2a74..68b699d569 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorIndexer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorIndexer.scala
@@ -31,7 +31,7 @@ import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{DenseVector, SparseVector, Vector, VectorUDT}
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.udf
 import org.apache.spark.sql.types.{StructField, StructType}
 import org.apache.spark.util.collection.OpenHashSet
@@ -108,7 +108,8 @@ class VectorIndexer(override val uid: String) extends Estimator[VectorIndexerMod
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def fit(dataset: DataFrame): VectorIndexerModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): VectorIndexerModel = {
     transformSchema(dataset.schema, logging = true)
     val firstRow = dataset.select($(inputCol)).take(1)
     require(firstRow.length == 1, s"VectorIndexer cannot be fit on an empty dataset.")
@@ -345,7 +346,8 @@ class VectorIndexerModel private[ml] (
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     val newField = prepOutputField(dataset.schema)
     val transformUDF = udf { (vector: Vector) => transformFunc(vector) }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorSlicer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorSlicer.scala
index b60e82de00..7a9468b87b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorSlicer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorSlicer.scala
@@ -24,7 +24,7 @@ import org.apache.spark.ml.param.{IntArrayParam, ParamMap, StringArrayParam}
 import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.StructType
 
@@ -89,7 +89,8 @@ final class VectorSlicer(override val uid: String)
   /** @group setParam */
   def setOutputCol(value: String): this.type = set(outputCol, value)
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     // Validity checks
     transformSchema(dataset.schema)
     val inputAttr = AttributeGroup.fromStructField(dataset.schema($(inputCol)))
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Word2Vec.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Word2Vec.scala
index 95bae1c8a3..a72692960f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/Word2Vec.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/Word2Vec.scala
@@ -27,7 +27,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.feature
 import org.apache.spark.mllib.linalg.{BLAS, Vector, Vectors, VectorUDT}
-import org.apache.spark.sql.{DataFrame, SQLContext}
+import org.apache.spark.sql.{DataFrame, Dataset, SQLContext}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types._
 
@@ -135,7 +135,8 @@ final class Word2Vec(override val uid: String) extends Estimator[Word2VecModel]
   /** @group setParam */
   def setMinCount(value: Int): this.type = set(minCount, value)
 
-  override def fit(dataset: DataFrame): Word2VecModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): Word2VecModel = {
     transformSchema(dataset.schema, logging = true)
     val input = dataset.select($(inputCol)).rdd.map(_.getAs[Seq[String]](0))
     val wordVectors = new feature.Word2Vec()
@@ -219,7 +220,8 @@ class Word2VecModel private[ml] (
    * Transform a sentence column to a vector column to represent the whole sentence. The transform
    * is performed by averaging all word vectors it contains.
    */
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     val vectors = wordVectors.getVectors
       .mapValues(vv => Vectors.dense(vv.map(_.toDouble)))
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala
index 40590e71c4..2ae411555f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala
@@ -22,7 +22,7 @@ import org.apache.spark.ml.{Pipeline, PipelineModel}
 import org.apache.spark.ml.attribute.AttributeGroup
 import org.apache.spark.ml.feature.RFormula
 import org.apache.spark.ml.regression.{AFTSurvivalRegression, AFTSurvivalRegressionModel}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 private[r] class AFTSurvivalRegressionWrapper private (
     pipeline: PipelineModel,
@@ -43,7 +43,7 @@ private[r] class AFTSurvivalRegressionWrapper private (
     features ++ Array("Log(scale)")
   }
 
-  def transform(dataset: DataFrame): DataFrame = {
+  def transform(dataset: Dataset[_]): DataFrame = {
     pipeline.transform(dataset)
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
index ed735a4ea3..ee513579ce 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
@@ -21,7 +21,7 @@ import org.apache.spark.ml.{Pipeline, PipelineModel}
 import org.apache.spark.ml.attribute.AttributeGroup
 import org.apache.spark.ml.clustering.{KMeans, KMeansModel}
 import org.apache.spark.ml.feature.VectorAssembler
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 private[r] class KMeansWrapper private (
     pipeline: PipelineModel) {
@@ -52,7 +52,7 @@ private[r] class KMeansWrapper private (
     }
   }
 
-  def transform(dataset: DataFrame): DataFrame = {
+  def transform(dataset: Dataset[_]): DataFrame = {
     pipeline.transform(dataset).drop(kMeansModel.getFeaturesCol)
   }
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala
index 07383d393d..2cd709d2ee 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala
@@ -21,7 +21,7 @@ import org.apache.spark.ml.{Pipeline, PipelineModel}
 import org.apache.spark.ml.attribute.{Attribute, AttributeGroup, NominalAttribute}
 import org.apache.spark.ml.classification.{NaiveBayes, NaiveBayesModel}
 import org.apache.spark.ml.feature.{IndexToString, RFormula}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 private[r] class NaiveBayesWrapper private (
     pipeline: PipelineModel,
@@ -36,7 +36,7 @@ private[r] class NaiveBayesWrapper private (
 
   lazy val tables: Array[Double] = naiveBayesModel.theta.toArray.map(math.exp)
 
-  def transform(dataset: DataFrame): DataFrame = {
+  def transform(dataset: Dataset[_]): DataFrame = {
     pipeline.transform(dataset).drop(PREDICTED_LABEL_INDEX_COL)
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala b/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala
index 4a3ad662a0..36dce01590 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala
@@ -40,7 +40,7 @@ import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.CholeskyDecomposition
 import org.apache.spark.mllib.optimization.NNLS
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DoubleType, FloatType, IntegerType, StructType}
 import org.apache.spark.storage.StorageLevel
@@ -200,8 +200,8 @@ class ALSModel private[ml] (
   @Since("1.3.0")
   def setPredictionCol(value: String): this.type = set(predictionCol, value)
 
-  @Since("1.3.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     // Register a UDF for DataFrame, and then
     // create a new column named map(predictionCol) by running the predict UDF.
     val predict = udf { (userFeatures: Seq[Float], itemFeatures: Seq[Float]) =>
@@ -385,8 +385,8 @@ class ALS(@Since("1.4.0") override val uid: String) extends Estimator[ALSModel]
     this
   }
 
-  @Since("1.3.0")
-  override def fit(dataset: DataFrame): ALSModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): ALSModel = {
     import dataset.sqlContext.implicits._
     val r = if ($(ratingCol) != "") col($(ratingCol)).cast(FloatType) else lit(1.0f)
     val ratings = dataset
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
index 3278974954..afed1f32b6 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
@@ -32,7 +32,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{BLAS, Vector, Vectors, VectorUDT}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DoubleType, StructType}
 import org.apache.spark.storage.StorageLevel
@@ -183,7 +183,7 @@ class AFTSurvivalRegression @Since("1.6.0") (@Since("1.6.0") override val uid: S
    * Extract [[featuresCol]], [[labelCol]] and [[censorCol]] from input dataset,
    * and put it in an RDD with strong types.
    */
-  protected[ml] def extractAFTPoints(dataset: DataFrame): RDD[AFTPoint] = {
+  protected[ml] def extractAFTPoints(dataset: Dataset[_]): RDD[AFTPoint] = {
     dataset.select(col($(featuresCol)), col($(labelCol)).cast(DoubleType), col($(censorCol)))
       .rdd.map {
         case Row(features: Vector, label: Double, censor: Double) =>
@@ -191,8 +191,8 @@ class AFTSurvivalRegression @Since("1.6.0") (@Since("1.6.0") override val uid: S
       }
   }
 
-  @Since("1.6.0")
-  override def fit(dataset: DataFrame): AFTSurvivalRegressionModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): AFTSurvivalRegressionModel = {
     validateAndTransformSchema(dataset.schema, fitting = true)
     val instances = extractAFTPoints(dataset)
     val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE
@@ -299,8 +299,8 @@ class AFTSurvivalRegressionModel private[ml] (
     math.exp(BLAS.dot(coefficients, features) + intercept)
   }
 
-  @Since("1.6.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema)
     val predictUDF = udf { features: Vector => predict(features) }
     val predictQuantilesUDF = udf { features: Vector => predictQuantiles(features)}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
index 1289a317ee..c04c416aaf 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/DecisionTreeRegressor.scala
@@ -33,7 +33,7 @@ import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, Strategy => OldStrategy}
 import org.apache.spark.mllib.tree.model.{DecisionTreeModel => OldDecisionTreeModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 
 
@@ -83,7 +83,7 @@ final class DecisionTreeRegressor @Since("1.4.0") (@Since("1.4.0") override val
   /** @group setParam */
   def setVarianceCol(value: String): this.type = set(varianceCol, value)
 
-  override protected def train(dataset: DataFrame): DecisionTreeRegressionModel = {
+  override protected def train(dataset: Dataset[_]): DecisionTreeRegressionModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
     val oldDataset: RDD[LabeledPoint] = extractLabeledPoints(dataset)
@@ -158,15 +158,16 @@ final class DecisionTreeRegressionModel private[ml] (
     rootNode.predictImpl(features).impurityStats.calculate()
   }
 
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     transformImpl(dataset)
   }
 
-  override protected def transformImpl(dataset: DataFrame): DataFrame = {
+  override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val predictUDF = udf { (features: Vector) => predict(features) }
     val predictVarianceUDF = udf { (features: Vector) => predictVariance(features) }
-    var output = dataset
+    var output = dataset.toDF
     if ($(predictionCol).nonEmpty) {
       output = output.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
     }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
index 8eb2984f7b..0b52fe2d13 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -33,7 +33,7 @@ import org.apache.spark.mllib.tree.loss.{AbsoluteError => OldAbsoluteError, Loss
   SquaredError => OldSquaredError}
 import org.apache.spark.mllib.tree.model.{GradientBoostedTreesModel => OldGBTModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 
 /**
@@ -147,7 +147,7 @@ final class GBTRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: Stri
     }
   }
 
-  override protected def train(dataset: DataFrame): GBTRegressionModel = {
+  override protected def train(dataset: Dataset[_]): GBTRegressionModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
     val oldDataset: RDD[LabeledPoint] = extractLabeledPoints(dataset)
@@ -209,7 +209,7 @@ final class GBTRegressionModel private[ml](
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
 
-  override protected def transformImpl(dataset: DataFrame): DataFrame = {
+  override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sqlContext.sparkContext.broadcast(this)
     val predictUDF = udf { (features: Any) =>
       bcastModel.value.predict(features.asInstanceOf[Vector])
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
index 05bf64591b..00cf25dc54 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
@@ -31,7 +31,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{BLAS, Vector}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.{DataType, DoubleType, StructType}
 
@@ -196,7 +196,7 @@ class GeneralizedLinearRegression @Since("2.0.0") (@Since("2.0.0") override val
   def setSolver(value: String): this.type = set(solver, value)
   setDefault(solver -> "irls")
 
-  override protected def train(dataset: DataFrame): GeneralizedLinearRegressionModel = {
+  override protected def train(dataset: Dataset[_]): GeneralizedLinearRegressionModel = {
     val familyObj = Family.fromName($(family))
     val linkObj = if (isDefined(link)) {
       Link.fromName($(link))
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala
index bd0b631d89..7a78ecbdf1 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/IsotonicRegression.scala
@@ -30,7 +30,7 @@ import org.apache.spark.mllib.linalg.{Vector, Vectors, VectorUDT}
 import org.apache.spark.mllib.regression.{IsotonicRegression => MLlibIsotonicRegression}
 import org.apache.spark.mllib.regression.{IsotonicRegressionModel => MLlibIsotonicRegressionModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.{col, lit, udf}
 import org.apache.spark.sql.types.{DoubleType, StructType}
 import org.apache.spark.storage.StorageLevel
@@ -77,7 +77,7 @@ private[regression] trait IsotonicRegressionBase extends Params with HasFeatures
    * Extracts (label, feature, weight) from input dataset.
    */
   protected[ml] def extractWeightedLabeledPoints(
-      dataset: DataFrame): RDD[(Double, Double, Double)] = {
+      dataset: Dataset[_]): RDD[(Double, Double, Double)] = {
     val f = if (dataset.schema($(featuresCol)).dataType.isInstanceOf[VectorUDT]) {
       val idx = $(featureIndex)
       val extract = udf { v: Vector => v(idx) }
@@ -164,8 +164,8 @@ class IsotonicRegression @Since("1.5.0") (@Since("1.5.0") override val uid: Stri
   @Since("1.5.0")
   override def copy(extra: ParamMap): IsotonicRegression = defaultCopy(extra)
 
-  @Since("1.5.0")
-  override def fit(dataset: DataFrame): IsotonicRegressionModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): IsotonicRegressionModel = {
     validateAndTransformSchema(dataset.schema, fitting = true)
     // Extract columns from data.  If dataset is persisted, do not persist oldDataset.
     val instances = extractWeightedLabeledPoints(dataset)
@@ -236,8 +236,8 @@ class IsotonicRegressionModel private[ml] (
     copyValues(new IsotonicRegressionModel(uid, oldModel), extra).setParent(parent)
   }
 
-  @Since("1.5.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     val predict = dataset.schema($(featuresCol)).dataType match {
       case DoubleType =>
         udf { feature: Double => oldModel.predict(feature) }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
index aacff4ea47..71e02730c7 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala
@@ -38,7 +38,7 @@ import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.linalg.BLAS._
 import org.apache.spark.mllib.stat.MultivariateOnlineSummarizer
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.types.DoubleType
 import org.apache.spark.storage.StorageLevel
@@ -158,7 +158,7 @@ class LinearRegression @Since("1.3.0") (@Since("1.3.0") override val uid: String
   def setSolver(value: String): this.type = set(solver, value)
   setDefault(solver -> "auto")
 
-  override protected def train(dataset: DataFrame): LinearRegressionModel = {
+  override protected def train(dataset: Dataset[_]): LinearRegressionModel = {
     // Extract the number of features before deciding optimization solver.
     val numFeatures = dataset.select(col($(featuresCol))).limit(1).rdd.map {
       case Row(features: Vector) => features.size
@@ -417,7 +417,7 @@ class LinearRegressionModel private[ml] (
    * @param dataset Test dataset to evaluate model on.
    */
   @Since("2.0.0")
-  def evaluate(dataset: DataFrame): LinearRegressionSummary = {
+  def evaluate(dataset: Dataset[_]): LinearRegressionSummary = {
     // Handle possible missing or invalid prediction columns
     val (summaryModel, predictionColName) = findSummaryModelAndPredictionCol()
     new LinearRegressionSummary(summaryModel.transform(dataset), predictionColName,
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
index 736cd9f776..bee13c2ebf 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
@@ -32,7 +32,7 @@ import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
 import org.apache.spark.mllib.tree.model.{RandomForestModel => OldRandomForestModel}
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.functions._
 
 
@@ -93,7 +93,7 @@ final class RandomForestRegressor @Since("1.4.0") (@Since("1.4.0") override val
   override def setFeatureSubsetStrategy(value: String): this.type =
     super.setFeatureSubsetStrategy(value)
 
-  override protected def train(dataset: DataFrame): RandomForestRegressionModel = {
+  override protected def train(dataset: Dataset[_]): RandomForestRegressionModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
     val oldDataset: RDD[LabeledPoint] = extractLabeledPoints(dataset)
@@ -164,7 +164,7 @@ final class RandomForestRegressionModel private[ml] (
   @Since("1.4.0")
   override def treeWeights: Array[Double] = _treeWeights
 
-  override protected def transformImpl(dataset: DataFrame): DataFrame = {
+  override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sqlContext.sparkContext.broadcast(this)
     val predictUDF = udf { (features: Any) =>
       bcastModel.value.predict(features.asInstanceOf[Vector])
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
index 4d9d4d472e..de563d4fad 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala
@@ -33,7 +33,7 @@ import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared.HasSeed
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.util.MLUtils
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.StructType
 
 /**
@@ -90,8 +90,8 @@ class CrossValidator @Since("1.2.0") (@Since("1.4.0") override val uid: String)
   @Since("2.0.0")
   def setSeed(value: Long): this.type = set(seed, value)
 
-  @Since("1.4.0")
-  override def fit(dataset: DataFrame): CrossValidatorModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): CrossValidatorModel = {
     val schema = dataset.schema
     transformSchema(schema, logging = true)
     val sqlCtx = dataset.sqlContext
@@ -100,7 +100,7 @@ class CrossValidator @Since("1.2.0") (@Since("1.4.0") override val uid: String)
     val epm = $(estimatorParamMaps)
     val numModels = epm.length
     val metrics = new Array[Double](epm.length)
-    val splits = MLUtils.kFold(dataset.rdd, $(numFolds), $(seed))
+    val splits = MLUtils.kFold(dataset.toDF.rdd, $(numFolds), $(seed))
     splits.zipWithIndex.foreach { case ((training, validation), splitIndex) =>
       val trainingDataset = sqlCtx.createDataFrame(training, schema).cache()
       val validationDataset = sqlCtx.createDataFrame(validation, schema).cache()
@@ -209,8 +209,8 @@ class CrossValidatorModel private[ml] (
     this(uid, bestModel, avgMetrics.asScala.toArray)
   }
 
-  @Since("1.4.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     bestModel.transform(dataset)
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
index 0f2179c2a1..12d6905510 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tuning/TrainValidationSplit.scala
@@ -20,6 +20,7 @@ package org.apache.spark.ml.tuning
 import java.util.{List => JList}
 
 import scala.collection.JavaConverters._
+import scala.language.existentials
 
 import org.apache.hadoop.fs.Path
 import org.json4s.DefaultFormats
@@ -31,7 +32,7 @@ import org.apache.spark.ml.evaluation.Evaluator
 import org.apache.spark.ml.param.{DoubleParam, ParamMap, ParamValidators}
 import org.apache.spark.ml.param.shared.HasSeed
 import org.apache.spark.ml.util._
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.StructType
 
 /**
@@ -89,8 +90,8 @@ class TrainValidationSplit @Since("1.5.0") (@Since("1.5.0") override val uid: St
   @Since("2.0.0")
   def setSeed(value: Long): this.type = set(seed, value)
 
-  @Since("1.5.0")
-  override def fit(dataset: DataFrame): TrainValidationSplitModel = {
+  @Since("2.0.0")
+  override def fit(dataset: Dataset[_]): TrainValidationSplitModel = {
     val schema = dataset.schema
     transformSchema(schema, logging = true)
     val sqlCtx = dataset.sqlContext
@@ -207,8 +208,8 @@ class TrainValidationSplitModel private[ml] (
     this(uid, bestModel, validationMetrics.asScala.toArray)
   }
 
-  @Since("1.5.0")
-  override def transform(dataset: DataFrame): DataFrame = {
+  @Since("2.0.0")
+  override def transform(dataset: Dataset[_]): DataFrame = {
     transformSchema(dataset.schema, logging = true)
     bestModel.transform(dataset)
   }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
index 0f0c3a2df5..5812cdde2c 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
@@ -186,7 +186,7 @@ sealed trait Vector extends Serializable {
  * :: AlphaComponent ::
  *
  * User-defined type for [[Vector]] which allows easy interaction with SQL
- * via [[org.apache.spark.sql.DataFrame]].
+ * via [[org.apache.spark.sql.Dataset]].
  */
 @AlphaComponent
 class VectorUDT extends UserDefinedType[Vector] {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/PipelineSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/PipelineSuite.scala
index f3321fb5a1..a8c4ac6d05 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/PipelineSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/PipelineSuite.scala
@@ -31,7 +31,7 @@ import org.apache.spark.ml.param.{IntParam, ParamMap}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.StructType
 
 class PipelineSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
@@ -51,6 +51,12 @@ class PipelineSuite extends SparkFunSuite with MLlibTestSparkContext with Defaul
     val dataset3 = mock[DataFrame]
     val dataset4 = mock[DataFrame]
 
+    when(dataset0.toDF).thenReturn(dataset0)
+    when(dataset1.toDF).thenReturn(dataset1)
+    when(dataset2.toDF).thenReturn(dataset2)
+    when(dataset3.toDF).thenReturn(dataset3)
+    when(dataset4.toDF).thenReturn(dataset4)
+
     when(estimator0.copy(any[ParamMap])).thenReturn(estimator0)
     when(model0.copy(any[ParamMap])).thenReturn(model0)
     when(transformer1.copy(any[ParamMap])).thenReturn(transformer1)
@@ -213,7 +219,7 @@ class WritableStage(override val uid: String) extends Transformer with MLWritabl
 
   override def write: MLWriter = new DefaultParamsWriter(this)
 
-  override def transform(dataset: DataFrame): DataFrame = dataset
+  override def transform(dataset: Dataset[_]): DataFrame = dataset.toDF
 
   override def transformSchema(schema: StructType): StructType = schema
 }
@@ -234,7 +240,7 @@ class UnWritableStage(override val uid: String) extends Transformer {
 
   override def copy(extra: ParamMap): UnWritableStage = defaultCopy(extra)
 
-  override def transform(dataset: DataFrame): DataFrame = dataset
+  override def transform(dataset: Dataset[_]): DataFrame = dataset.toDF
 
   override def transformSchema(schema: StructType): StructType = schema
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
index 7eefaf2346..48db428130 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
@@ -29,13 +29,13 @@ import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions.lit
 
 class LogisticRegressionSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
   @transient var binaryDataset: DataFrame = _
   private val eps: Double = 1e-5
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
index 06ff049b48..80547fad6a 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala
@@ -26,12 +26,12 @@ import org.apache.spark.mllib.evaluation.MulticlassMetrics
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 
 class MultilayerPerceptronClassifierSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala
index 4727cd436f..80a46fc70c 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/NaiveBayesSuite.scala
@@ -27,11 +27,11 @@ import org.apache.spark.mllib.classification.NaiveBayesSuite._
 import org.apache.spark.mllib.linalg._
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 
 class NaiveBayesSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
index 4131396726..f3e8fd11b2 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
@@ -30,12 +30,12 @@ import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.Metadata
 
 class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
   @transient var rdd: RDD[LabeledPoint] = _
 
   override def beforeAll(): Unit = {
@@ -246,7 +246,7 @@ private class MockLogisticRegression(uid: String) extends LogisticRegression(uid
 
   setMaxIter(1)
 
-  override protected[spark] def train(dataset: DataFrame): LogisticRegressionModel = {
+  override protected[spark] def train(dataset: Dataset[_]): LogisticRegressionModel = {
     val labelSchema = dataset.schema($(labelCol))
     // check for label attribute propagation.
     assert(MetadataUtils.getNumClasses(labelSchema).forall(_ == 2))
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala
index 18f2c994b4..e641d79c17 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/BisectingKMeansSuite.scala
@@ -20,13 +20,13 @@ package org.apache.spark.ml.clustering
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 class BisectingKMeansSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
   final val k = 5
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala
index 8edd44e5f1..1a274aea29 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/GaussianMixtureSuite.scala
@@ -20,14 +20,14 @@ package org.apache.spark.ml.clustering
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 
 class GaussianMixtureSuite extends SparkFunSuite with MLlibTestSparkContext
   with DefaultReadWriteTest {
 
   final val k = 5
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala
index c684bc11cc..2076c745e2 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala
@@ -22,14 +22,14 @@ import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.clustering.{KMeans => MLlibKMeans}
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.{DataFrame, SQLContext}
+import org.apache.spark.sql.{DataFrame, Dataset, SQLContext}
 
 private[clustering] case class TestRow(features: Vector)
 
 class KMeansSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
   final val k = 5
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala
index a1c93891c7..ee8eae8f69 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/LDASuite.scala
@@ -24,7 +24,7 @@ import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
-import org.apache.spark.sql.{DataFrame, Row, SQLContext}
+import org.apache.spark.sql.{DataFrame, Dataset, Row, SQLContext}
 
 
 object LDASuite {
@@ -64,7 +64,7 @@ class LDASuite extends SparkFunSuite with MLlibTestSparkContext with DefaultRead
 
   val k: Int = 5
   val vocabSize: Int = 30
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/NGramSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/NGramSuite.scala
index 58fda29aa1..e4e15f4331 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/NGramSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/NGramSuite.scala
@@ -22,7 +22,7 @@ import scala.beans.BeanInfo
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 
 @BeanInfo
 case class NGramTestData(inputTokens: Array[String], wantedNGrams: Array[String])
@@ -92,7 +92,7 @@ class NGramSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultRe
 
 object NGramSuite extends SparkFunSuite {
 
-  def testNGram(t: NGram, dataset: DataFrame): Unit = {
+  def testNGram(t: NGram, dataset: Dataset[_]): Unit = {
     t.transform(dataset)
       .select("nGrams", "wantedNGrams")
       .collect()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/StopWordsRemoverSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/StopWordsRemoverSuite.scala
index a5b24c1856..3505befdf8 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/StopWordsRemoverSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/StopWordsRemoverSuite.scala
@@ -20,10 +20,10 @@ package org.apache.spark.ml.feature
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 
 object StopWordsRemoverSuite extends SparkFunSuite {
-  def testStopWordsRemover(t: StopWordsRemover, dataset: DataFrame): Unit = {
+  def testStopWordsRemover(t: StopWordsRemover, dataset: Dataset[_]): Unit = {
     t.transform(dataset)
       .select("filtered", "expected")
       .collect()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/StringIndexerSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/StringIndexerSuite.scala
index 2c3255ef33..d0f3cdc841 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/StringIndexerSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/StringIndexerSuite.scala
@@ -115,7 +115,7 @@ class StringIndexerSuite
       .setInputCol("label")
       .setOutputCol("labelIndex")
     val df = sqlContext.range(0L, 10L).toDF()
-    assert(indexerModel.transform(df).eq(df))
+    assert(indexerModel.transform(df).collect().toSet === df.collect().toSet)
   }
 
   test("StringIndexerModel can't overwrite output column") {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/TokenizerSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/TokenizerSuite.scala
index 36e8e5d868..299f6223b2 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/TokenizerSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/TokenizerSuite.scala
@@ -23,7 +23,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.param.ParamsSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.{DataFrame, Row}
+import org.apache.spark.sql.{DataFrame, Dataset, Row}
 
 @BeanInfo
 case class TokenizerTestData(rawText: String, wantedTokens: Array[String])
@@ -106,7 +106,7 @@ class RegexTokenizerSuite
 
 object RegexTokenizerSuite extends SparkFunSuite {
 
-  def testRegexTokenizer(t: RegexTokenizer, dataset: DataFrame): Unit = {
+  def testRegexTokenizer(t: RegexTokenizer, dataset: Dataset[_]): Unit = {
     t.transform(dataset)
       .select("tokens", "wantedTokens")
       .collect()
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
index 2265464b51..4905f3e068 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
@@ -992,6 +992,14 @@ class GeneralizedLinearRegressionSuite
         assert(expected.coefficients === actual.coefficients)
       }
   }
+
+  test("glm accepts Dataset[LabeledPoint]") {
+    val context = sqlContext
+    import context.implicits._
+    new GeneralizedLinearRegression()
+      .setFamily("gaussian")
+      .fit(datasetGaussianIdentity.as[LabeledPoint])
+  }
 }
 
 object GeneralizedLinearRegressionSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tuning/CrossValidatorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tuning/CrossValidatorSuite.scala
index 7af3c6d6ed..3e734aabc5 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tuning/CrossValidatorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tuning/CrossValidatorSuite.scala
@@ -29,13 +29,13 @@ import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInput
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.util.{LinearDataGenerator, MLlibTestSparkContext}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.{StructField, StructType}
 
 class CrossValidatorSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
-  @transient var dataset: DataFrame = _
+  @transient var dataset: Dataset[_] = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
@@ -311,7 +311,7 @@ object CrossValidatorSuite extends SparkFunSuite {
 
   class MyEstimator(override val uid: String) extends Estimator[MyModel] with HasInputCol {
 
-    override def fit(dataset: DataFrame): MyModel = {
+    override def fit(dataset: Dataset[_]): MyModel = {
       throw new UnsupportedOperationException
     }
 
@@ -325,7 +325,7 @@ object CrossValidatorSuite extends SparkFunSuite {
 
   class MyEvaluator extends Evaluator {
 
-    override def evaluate(dataset: DataFrame): Double = {
+    override def evaluate(dataset: Dataset[_]): Double = {
       throw new UnsupportedOperationException
     }
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
index 4030956fab..dbee47c847 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tuning/TrainValidationSplitSuite.scala
@@ -28,7 +28,7 @@ import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInput
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.util.{LinearDataGenerator, MLlibTestSparkContext}
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 import org.apache.spark.sql.types.StructType
 
 class TrainValidationSplitSuite
@@ -158,7 +158,7 @@ object TrainValidationSplitSuite {
 
   class MyEstimator(override val uid: String) extends Estimator[MyModel] with HasInputCol {
 
-    override def fit(dataset: DataFrame): MyModel = {
+    override def fit(dataset: Dataset[_]): MyModel = {
       throw new UnsupportedOperationException
     }
 
@@ -172,7 +172,7 @@ object TrainValidationSplitSuite {
 
   class MyEvaluator extends Evaluator {
 
-    override def evaluate(dataset: DataFrame): Double = {
+    override def evaluate(dataset: Dataset[_]): Double = {
       throw new UnsupportedOperationException
     }
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/util/DefaultReadWriteTest.scala b/mllib/src/test/scala/org/apache/spark/ml/util/DefaultReadWriteTest.scala
index 16280473c6..7ebd7eb144 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/util/DefaultReadWriteTest.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/util/DefaultReadWriteTest.scala
@@ -25,7 +25,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.param._
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.DataFrame
+import org.apache.spark.sql.{DataFrame, Dataset}
 
 trait DefaultReadWriteTest extends TempDirectory { self: Suite =>
 
@@ -98,7 +98,7 @@ trait DefaultReadWriteTest extends TempDirectory { self: Suite =>
   def testEstimatorAndModelReadWrite[
     E <: Estimator[M] with MLWritable, M <: Model[M] with MLWritable](
       estimator: E,
-      dataset: DataFrame,
+      dataset: Dataset[_],
       testParams: Map[String, Any],
       checkModelData: (M, M) => Unit): Unit = {
     // Set some Params to make sure set Params are serialized.
-- 
cgit v1.2.3


From 643b4e2257c56338b192f8554e2fe5523bea4bdf Mon Sep 17 00:00:00 2001
From: Zheng RuiFeng <ruifengz@foxmail.com>
Date: Mon, 11 Apr 2016 09:33:52 -0700
Subject: [SPARK-14510][MLLIB] Add args-checking for LDA and StreamingKMeans

## What changes were proposed in this pull request?
add the checking for LDA and StreamingKMeans

## How was this patch tested?
manual tests

Author: Zheng RuiFeng <ruifengz@foxmail.com>

Closes #12062 from zhengruifeng/initmodel.
---
 .../src/main/scala/org/apache/spark/mllib/clustering/LDA.scala | 10 +++++++---
 .../org/apache/spark/mllib/clustering/StreamingKMeans.scala    | 10 ++++++++++
 2 files changed, 17 insertions(+), 3 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDA.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDA.scala
index 12813fd412..d999b9be8e 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDA.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LDA.scala
@@ -130,7 +130,8 @@ class LDA private (
    */
   @Since("1.5.0")
   def setDocConcentration(docConcentration: Vector): this.type = {
-    require(docConcentration.size > 0, "docConcentration must have > 0 elements")
+    require(docConcentration.size == 1 || docConcentration.size == k,
+      s"Size of docConcentration must be 1 or ${k} but got ${docConcentration.size}")
     this.docConcentration = docConcentration
     this
   }
@@ -260,15 +261,18 @@ class LDA private (
   def getCheckpointInterval: Int = checkpointInterval
 
   /**
-   * Period (in iterations) between checkpoints (default = 10). Checkpointing helps with recovery
+   * Parameter for set checkpoint interval (>= 1) or disable checkpoint (-1). E.g. 10 means that
+   * the cache will get checkpointed every 10 iterations. Checkpointing helps with recovery
    * (when nodes fail). It also helps with eliminating temporary shuffle files on disk, which can be
    * important when LDA is run for many iterations. If the checkpoint directory is not set in
-   * [[org.apache.spark.SparkContext]], this setting is ignored.
+   * [[org.apache.spark.SparkContext]], this setting is ignored. (default = 10)
    *
    * @see [[org.apache.spark.SparkContext#setCheckpointDir]]
    */
   @Since("1.3.0")
   def setCheckpointInterval(checkpointInterval: Int): this.type = {
+    require(checkpointInterval == -1 || checkpointInterval > 0,
+      s"Period between checkpoints must be -1 or positive but got ${checkpointInterval}")
     this.checkpointInterval = checkpointInterval
     this
   }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/StreamingKMeans.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/StreamingKMeans.scala
index 4eb8fc049e..24e1cff0dc 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/StreamingKMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/StreamingKMeans.scala
@@ -218,6 +218,12 @@ class StreamingKMeans @Since("1.2.0") (
    */
   @Since("1.2.0")
   def setInitialCenters(centers: Array[Vector], weights: Array[Double]): this.type = {
+    require(centers.size == weights.size,
+      "Number of initial centers must be equal to number of weights")
+    require(centers.size == k,
+      s"Number of initial centers must be ${k} but got ${centers.size}")
+    require(weights.forall(_ >= 0),
+      s"Weight for each inital center must be nonnegative but got [${weights.mkString(" ")}]")
     model = new StreamingKMeansModel(centers, weights)
     this
   }
@@ -231,6 +237,10 @@ class StreamingKMeans @Since("1.2.0") (
    */
   @Since("1.2.0")
   def setRandomCenters(dim: Int, weight: Double, seed: Long = Utils.random.nextLong): this.type = {
+    require(dim > 0,
+      s"Number of dimensions must be positive but got ${dim}")
+    require(weight >= 0,
+      s"Weight for each center must be nonnegative but got ${weight}")
     val random = new XORShiftRandom(seed)
     val centers = Array.fill(k)(Vectors.dense(Array.fill(dim)(random.nextGaussian())))
     val weights = Array.fill(k)(weight)
-- 
cgit v1.2.3


From 89a41c5b7a3f727b44a7f615a1352ca006d12f73 Mon Sep 17 00:00:00 2001
From: Oliver Pierson <ocp@gatech.edu>
Date: Mon, 11 Apr 2016 12:02:48 -0700
Subject: [SPARK-13600][MLLIB] Use approxQuantile from DataFrame stats in
 QuantileDiscretizer

## What changes were proposed in this pull request?
QuantileDiscretizer can return an unexpected number of buckets in certain cases.  This PR proposes to fix this issue and also refactor QuantileDiscretizer to use approxQuantiles from DataFrame stats functions.
## How was this patch tested?
QuantileDiscretizerSuite unit tests (some existing tests will change or even be removed in this PR)

Author: Oliver Pierson <ocp@gatech.edu>

Closes #11553 from oliverpierson/SPARK-13600.
---
 .../spark/ml/feature/QuantileDiscretizer.scala     | 119 +++++----------------
 .../ml/feature/QuantileDiscretizerSuite.scala      | 115 +++++++-------------
 2 files changed, 65 insertions(+), 169 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala
index efe8b93d82..5c7993af64 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/QuantileDiscretizer.scala
@@ -37,7 +37,7 @@ private[feature] trait QuantileDiscretizerBase extends Params
   with HasInputCol with HasOutputCol with HasSeed {
 
   /**
-   * Maximum number of buckets (quantiles, or categories) into which data points are grouped. Must
+   * Number of buckets (quantiles, or categories) into which data points are grouped. Must
    * be >= 2.
    * default: 2
    * @group param
@@ -49,6 +49,21 @@ private[feature] trait QuantileDiscretizerBase extends Params
 
   /** @group getParam */
   def getNumBuckets: Int = getOrDefault(numBuckets)
+
+  /**
+   * Relative error (see documentation for
+   * [[org.apache.spark.sql.DataFrameStatFunctions.approxQuantile approxQuantile]] for description)
+   * Must be a number in [0, 1].
+   * default: 0.001
+   * @group param
+   */
+  val relativeError = new DoubleParam(this, "relativeError", "The relative target precision " +
+    "for approxQuantile",
+    ParamValidators.inRange(0.0, 1.0))
+  setDefault(relativeError -> 0.001)
+
+  /** @group getParam */
+  def getRelativeError: Double = getOrDefault(relativeError)
 }
 
 /**
@@ -56,8 +71,7 @@ private[feature] trait QuantileDiscretizerBase extends Params
  * `QuantileDiscretizer` takes a column with continuous features and outputs a column with binned
  * categorical features. The bin ranges are chosen by taking a sample of the data and dividing it
  * into roughly equal parts. The lower and upper bin bounds will be -Infinity and +Infinity,
- * covering all real values. This attempts to find numBuckets partitions based on a sample of data,
- * but it may find fewer depending on the data sample values.
+ * covering all real values.
  */
 @Experimental
 final class QuantileDiscretizer(override val uid: String)
@@ -65,6 +79,9 @@ final class QuantileDiscretizer(override val uid: String)
 
   def this() = this(Identifiable.randomUID("quantileDiscretizer"))
 
+  /** @group setParam */
+  def setRelativeError(value: Double): this.type = set(relativeError, value)
+
   /** @group setParam */
   def setNumBuckets(value: Int): this.type = set(numBuckets, value)
 
@@ -89,11 +106,11 @@ final class QuantileDiscretizer(override val uid: String)
 
   @Since("2.0.0")
   override def fit(dataset: Dataset[_]): Bucketizer = {
-    val samples = QuantileDiscretizer
-      .getSampledInput(dataset.select($(inputCol)), $(numBuckets), $(seed))
-      .map { case Row(feature: Double) => feature }
-    val candidates = QuantileDiscretizer.findSplitCandidates(samples, $(numBuckets) - 1)
-    val splits = QuantileDiscretizer.getSplits(candidates)
+    val splits = dataset.stat.approxQuantile($(inputCol),
+      (0.0 to 1.0 by 1.0/$(numBuckets)).toArray, $(relativeError))
+    splits(0) = Double.NegativeInfinity
+    splits(splits.length - 1) = Double.PositiveInfinity
+
     val bucketizer = new Bucketizer(uid).setSplits(splits)
     copyValues(bucketizer.setParent(this))
   }
@@ -104,92 +121,6 @@ final class QuantileDiscretizer(override val uid: String)
 @Since("1.6.0")
 object QuantileDiscretizer extends DefaultParamsReadable[QuantileDiscretizer] with Logging {
 
-  /**
-   * Minimum number of samples required for finding splits, regardless of number of bins.  If
-   * the dataset has fewer rows than this value, the entire dataset will be used.
-   */
-  private[spark] val minSamplesRequired: Int = 10000
-
-  /**
-   * Sampling from the given dataset to collect quantile statistics.
-   */
-  private[feature]
-  def getSampledInput(dataset: Dataset[_], numBins: Int, seed: Long): Array[Row] = {
-    val totalSamples = dataset.count()
-    require(totalSamples > 0,
-      "QuantileDiscretizer requires non-empty input dataset but was given an empty input.")
-    val requiredSamples = math.max(numBins * numBins, minSamplesRequired)
-    val fraction = math.min(requiredSamples.toDouble / totalSamples, 1.0)
-    dataset.toDF.sample(withReplacement = false, fraction, new XORShiftRandom(seed).nextInt())
-      .collect()
-  }
-
-  /**
-   * Compute split points with respect to the sample distribution.
-   */
-  private[feature]
-  def findSplitCandidates(samples: Array[Double], numSplits: Int): Array[Double] = {
-    val valueCountMap = samples.foldLeft(Map.empty[Double, Int]) { (m, x) =>
-      m + ((x, m.getOrElse(x, 0) + 1))
-    }
-    val valueCounts = valueCountMap.toSeq.sortBy(_._1).toArray ++ Array((Double.MaxValue, 1))
-    val possibleSplits = valueCounts.length - 1
-    if (possibleSplits <= numSplits) {
-      valueCounts.dropRight(1).map(_._1)
-    } else {
-      val stride: Double = math.ceil(samples.length.toDouble / (numSplits + 1))
-      val splitsBuilder = mutable.ArrayBuilder.make[Double]
-      var index = 1
-      // currentCount: sum of counts of values that have been visited
-      var currentCount = valueCounts(0)._2
-      // targetCount: target value for `currentCount`. If `currentCount` is closest value to
-      // `targetCount`, then current value is a split threshold. After finding a split threshold,
-      // `targetCount` is added by stride.
-      var targetCount = stride
-      while (index < valueCounts.length) {
-        val previousCount = currentCount
-        currentCount += valueCounts(index)._2
-        val previousGap = math.abs(previousCount - targetCount)
-        val currentGap = math.abs(currentCount - targetCount)
-        // If adding count of current value to currentCount makes the gap between currentCount and
-        // targetCount smaller, previous value is a split threshold.
-        if (previousGap < currentGap) {
-          splitsBuilder += valueCounts(index - 1)._1
-          targetCount += stride
-        }
-        index += 1
-      }
-      splitsBuilder.result()
-    }
-  }
-
-  /**
-   * Adjust split candidates to proper splits by: adding positive/negative infinity to both sides as
-   * needed, and adding a default split value of 0 if no good candidates are found.
-   */
-  private[feature] def getSplits(candidates: Array[Double]): Array[Double] = {
-    val effectiveValues = if (candidates.nonEmpty) {
-      if (candidates.head == Double.NegativeInfinity
-        && candidates.last == Double.PositiveInfinity) {
-        candidates.drop(1).dropRight(1)
-      } else if (candidates.head == Double.NegativeInfinity) {
-        candidates.drop(1)
-      } else if (candidates.last == Double.PositiveInfinity) {
-        candidates.dropRight(1)
-      } else {
-        candidates
-      }
-    } else {
-      candidates
-    }
-
-    if (effectiveValues.isEmpty) {
-      Array(Double.NegativeInfinity, 0, Double.PositiveInfinity)
-    } else {
-      Array(Double.NegativeInfinity) ++ effectiveValues ++ Array(Double.PositiveInfinity)
-    }
-  }
-
   @Since("1.6.0")
   override def load(path: String): QuantileDiscretizer = super.load(path)
 }
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/QuantileDiscretizerSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/QuantileDiscretizerSuite.scala
index 25fabf64d5..8895d630a0 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/QuantileDiscretizerSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/QuantileDiscretizerSuite.scala
@@ -17,78 +17,60 @@
 
 package org.apache.spark.ml.feature
 
-import org.apache.spark.{SparkContext, SparkFunSuite}
-import org.apache.spark.ml.attribute.{Attribute, NominalAttribute}
+import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
-import org.apache.spark.sql.{Row, SQLContext}
+import org.apache.spark.sql.SQLContext
+import org.apache.spark.sql.functions.udf
 
 class QuantileDiscretizerSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
 
-  import org.apache.spark.ml.feature.QuantileDiscretizerSuite._
-
-  test("Test quantile discretizer") {
-    checkDiscretizedData(sc,
-      Array[Double](1, 2, 3, 3, 3, 3, 3, 3, 3),
-      10,
-      Array[Double](1, 2, 3, 3, 3, 3, 3, 3, 3),
-      Array("-Infinity, 1.0", "1.0, 2.0", "2.0, 3.0", "3.0, Infinity"))
-
-    checkDiscretizedData(sc,
-      Array[Double](1, 2, 3, 3, 3, 3, 3, 3, 3),
-      4,
-      Array[Double](1, 2, 3, 3, 3, 3, 3, 3, 3),
-      Array("-Infinity, 1.0", "1.0, 2.0", "2.0, 3.0", "3.0, Infinity"))
-
-    checkDiscretizedData(sc,
-      Array[Double](1, 2, 3, 3, 3, 3, 3, 3, 3),
-      3,
-      Array[Double](0, 1, 2, 2, 2, 2, 2, 2, 2),
-      Array("-Infinity, 2.0", "2.0, 3.0", "3.0, Infinity"))
+  test("Test observed number of buckets and their sizes match expected values") {
+    val sqlCtx = SQLContext.getOrCreate(sc)
+    import sqlCtx.implicits._
 
-    checkDiscretizedData(sc,
-      Array[Double](1, 2, 3, 3, 3, 3, 3, 3, 3),
-      2,
-      Array[Double](0, 1, 1, 1, 1, 1, 1, 1, 1),
-      Array("-Infinity, 2.0", "2.0, Infinity"))
+    val datasetSize = 100000
+    val numBuckets = 5
+    val df = sc.parallelize(1.0 to datasetSize by 1.0).map(Tuple1.apply).toDF("input")
+    val discretizer = new QuantileDiscretizer()
+      .setInputCol("input")
+      .setOutputCol("result")
+      .setNumBuckets(numBuckets)
+    val result = discretizer.fit(df).transform(df)
 
-  }
+    val observedNumBuckets = result.select("result").distinct.count
+    assert(observedNumBuckets === numBuckets,
+      "Observed number of buckets does not equal expected number of buckets.")
 
-  test("Test getting splits") {
-    val splitTestPoints = Array(
-      Array[Double]() -> Array(Double.NegativeInfinity, 0, Double.PositiveInfinity),
-      Array(Double.NegativeInfinity) -> Array(Double.NegativeInfinity, 0, Double.PositiveInfinity),
-      Array(Double.PositiveInfinity) -> Array(Double.NegativeInfinity, 0, Double.PositiveInfinity),
-      Array(Double.NegativeInfinity, Double.PositiveInfinity)
-        -> Array(Double.NegativeInfinity, 0, Double.PositiveInfinity),
-      Array(0.0) -> Array(Double.NegativeInfinity, 0, Double.PositiveInfinity),
-      Array(1.0) -> Array(Double.NegativeInfinity, 1, Double.PositiveInfinity),
-      Array(0.0, 1.0) -> Array(Double.NegativeInfinity, 0, 1, Double.PositiveInfinity)
-    )
-    for ((ori, res) <- splitTestPoints) {
-      assert(QuantileDiscretizer.getSplits(ori) === res, "Returned splits are invalid.")
+    val relativeError = discretizer.getRelativeError
+    val isGoodBucket = udf {
+      (size: Int) => math.abs( size - (datasetSize / numBuckets)) <= (relativeError * datasetSize)
     }
+    val numGoodBuckets = result.groupBy("result").count.filter(isGoodBucket($"count")).count
+    assert(numGoodBuckets === numBuckets,
+      "Bucket sizes are not within expected relative error tolerance.")
   }
 
-  test("Test splits on dataset larger than minSamplesRequired") {
+  test("Test transform method on unseen data") {
     val sqlCtx = SQLContext.getOrCreate(sc)
     import sqlCtx.implicits._
 
-    val datasetSize = QuantileDiscretizer.minSamplesRequired + 1
-    val numBuckets = 5
-    val df = sc.parallelize((1.0 to datasetSize by 1.0).map(Tuple1.apply)).toDF("input")
+    val trainDF = sc.parallelize(1.0 to 100.0 by 1.0).map(Tuple1.apply).toDF("input")
+    val testDF = sc.parallelize(-10.0 to 110.0 by 1.0).map(Tuple1.apply).toDF("input")
     val discretizer = new QuantileDiscretizer()
       .setInputCol("input")
       .setOutputCol("result")
-      .setNumBuckets(numBuckets)
-      .setSeed(1)
+      .setNumBuckets(5)
 
-    val result = discretizer.fit(df).transform(df)
-    val observedNumBuckets = result.select("result").distinct.count
+    val result = discretizer.fit(trainDF).transform(testDF)
+    val firstBucketSize = result.filter(result("result") === 0.0).count
+    val lastBucketSize = result.filter(result("result") === 4.0).count
 
-    assert(observedNumBuckets === numBuckets,
-      "Observed number of buckets does not equal expected number of buckets.")
+    assert(firstBucketSize === 30L,
+      s"Size of first bucket ${firstBucketSize} did not equal expected value of 30.")
+    assert(lastBucketSize === 31L,
+      s"Size of last bucket ${lastBucketSize} did not equal expected value of 31.")
   }
 
   test("read/write") {
@@ -98,34 +80,17 @@ class QuantileDiscretizerSuite
       .setNumBuckets(6)
     testDefaultReadWrite(t)
   }
-}
-
-private object QuantileDiscretizerSuite extends SparkFunSuite {
 
-  def checkDiscretizedData(
-      sc: SparkContext,
-      data: Array[Double],
-      numBucket: Int,
-      expectedResult: Array[Double],
-      expectedAttrs: Array[String]): Unit = {
+  test("Verify resulting model has parent") {
     val sqlCtx = SQLContext.getOrCreate(sc)
     import sqlCtx.implicits._
 
-    val df = sc.parallelize(data.map(Tuple1.apply)).toDF("input")
-    val discretizer = new QuantileDiscretizer().setInputCol("input").setOutputCol("result")
-      .setNumBuckets(numBucket).setSeed(1)
+    val df = sc.parallelize(1 to 100).map(Tuple1.apply).toDF("input")
+    val discretizer = new QuantileDiscretizer()
+      .setInputCol("input")
+      .setOutputCol("result")
+      .setNumBuckets(5)
     val model = discretizer.fit(df)
     assert(model.hasParent)
-    val result = model.transform(df)
-
-    val transformedFeatures = result.select("result").collect()
-      .map { case Row(transformedFeature: Double) => transformedFeature }
-    val transformedAttrs = Attribute.fromStructField(result.schema("result"))
-      .asInstanceOf[NominalAttribute].values.get
-
-    assert(transformedFeatures === expectedResult,
-      "Transformed features do not equal expected features.")
-    assert(transformedAttrs === expectedAttrs,
-      "Transformed attributes do not equal expected attributes.")
   }
 }
-- 
cgit v1.2.3


From 678b96e77bf77a64b8df14b19db5a3bb18febfe3 Mon Sep 17 00:00:00 2001
From: Wenchen Fan <wenchen@databricks.com>
Date: Mon, 11 Apr 2016 22:59:42 -0700
Subject: [SPARK-14535][SQL] Remove buildInternalScan from FileFormat

## What changes were proposed in this pull request?

Now `HadoopFsRelation` with all kinds of file formats can be handled in `FileSourceStrategy`, we can remove the branches for  `HadoopFsRelation` in `FileSourceStrategy` and the `buildInternalScan` API from `FileFormat`.

## How was this patch tested?

existing tests.

Author: Wenchen Fan <wenchen@databricks.com>

Closes #12300 from cloud-fan/remove.
---
 .../spark/ml/source/libsvm/LibSVMRelation.scala    |  34 +-
 .../execution/datasources/DataSourceStrategy.scala | 390 ---------------------
 .../execution/datasources/FileSourceStrategy.scala |  10 +-
 .../execution/datasources/csv/DefaultSource.scala  |  31 --
 .../execution/datasources/json/JSONRelation.scala  |  29 --
 .../datasources/parquet/ParquetRelation.scala      | 110 +-----
 .../execution/datasources/text/DefaultSource.scala |  39 ---
 .../org/apache/spark/sql/internal/SQLConf.scala    |   8 -
 .../org/apache/spark/sql/sources/interfaces.scala  |  10 -
 .../datasources/FileSourceStrategySuite.scala      |  12 -
 .../apache/spark/sql/hive/orc/OrcRelation.scala    |  13 -
 11 files changed, 5 insertions(+), 681 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala b/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
index 2e9b6be9a2..4737b6fe52 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
@@ -178,39 +178,6 @@ class DefaultSource extends FileFormat with DataSourceRegister {
     }
   }
 
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    // TODO: This does not handle cases where column pruning has been performed.
-
-    verifySchema(dataSchema)
-    val dataFiles = inputFiles.filterNot(_.getPath.getName startsWith "_")
-
-    val path = if (dataFiles.length == 1) dataFiles.head.getPath.toUri.toString
-    else if (dataFiles.isEmpty) throw new IOException("No input path specified for libsvm data")
-    else throw new IOException("Multiple input paths are not supported for libsvm data.")
-
-    val numFeatures = options.getOrElse("numFeatures", "-1").toInt
-    val vectorType = options.getOrElse("vectorType", "sparse")
-
-    val sc = sqlContext.sparkContext
-    val baseRdd = MLUtils.loadLibSVMFile(sc, path, numFeatures)
-    val sparse = vectorType == "sparse"
-    baseRdd.map { pt =>
-      val features = if (sparse) pt.features.toSparse else pt.features.toDense
-      Row(pt.label, features)
-    }.mapPartitions { externalRows =>
-      val converter = RowEncoder(dataSchema)
-      externalRows.map(converter.toRow)
-    }
-  }
-
   override def buildReader(
       sqlContext: SQLContext,
       dataSchema: StructType,
@@ -218,6 +185,7 @@ class DefaultSource extends FileFormat with DataSourceRegister {
       requiredSchema: StructType,
       filters: Seq[Filter],
       options: Map[String, String]): (PartitionedFile) => Iterator[InternalRow] = {
+    verifySchema(dataSchema)
     val numFeatures = options("numFeatures").toInt
     assert(numFeatures > 0)
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala
index 8c183317f6..c3885a3be5 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala
@@ -110,133 +110,6 @@ private[sql] object DataSourceStrategy extends Strategy with Logging {
         filters,
         (a, _) => toCatalystRDD(l, a, t.buildScan(a.map(_.name).toArray))) :: Nil
 
-    // Scanning partitioned HadoopFsRelation
-    case PhysicalOperation(projects, filters, l @ LogicalRelation(t: HadoopFsRelation, _, _))
-        if t.partitionSchema.nonEmpty =>
-      // We divide the filter expressions into 3 parts
-      val partitionColumns = AttributeSet(
-        t.partitionSchema.map(c => l.output.find(_.name == c.name).get))
-
-      // Only pruning the partition keys
-      val partitionFilters = filters.filter(_.references.subsetOf(partitionColumns))
-
-      // Only pushes down predicates that do not reference partition keys.
-      val pushedFilters = filters.filter(_.references.intersect(partitionColumns).isEmpty)
-
-      // Predicates with both partition keys and attributes
-      val partitionAndNormalColumnFilters =
-        filters.toSet -- partitionFilters.toSet -- pushedFilters.toSet
-
-      val selectedPartitions = t.location.listFiles(partitionFilters)
-
-      logInfo {
-        val total = t.partitionSpec.partitions.length
-        val selected = selectedPartitions.length
-        val percentPruned = (1 - selected.toDouble / total.toDouble) * 100
-        s"Selected $selected partitions out of $total, pruned $percentPruned% partitions."
-      }
-
-      // need to add projections from "partitionAndNormalColumnAttrs" in if it is not empty
-      val partitionAndNormalColumnAttrs = AttributeSet(partitionAndNormalColumnFilters)
-      val partitionAndNormalColumnProjs = if (partitionAndNormalColumnAttrs.isEmpty) {
-        projects
-      } else {
-        (partitionAndNormalColumnAttrs ++ projects).toSeq
-      }
-
-      // Prune the buckets based on the pushed filters that do not contain partitioning key
-      // since the bucketing key is not allowed to use the columns in partitioning key
-      val bucketSet = getBuckets(pushedFilters, t.bucketSpec)
-      val scan = buildPartitionedTableScan(
-        l,
-        partitionAndNormalColumnProjs,
-        pushedFilters,
-        bucketSet,
-        t.partitionSpec.partitionColumns,
-        selectedPartitions,
-        t.options)
-
-      // Add a Projection to guarantee the original projection:
-      // this is because "partitionAndNormalColumnAttrs" may be different
-      // from the original "projects", in elements or their ordering
-
-      partitionAndNormalColumnFilters.reduceLeftOption(expressions.And).map(cf =>
-        if (projects.isEmpty || projects == partitionAndNormalColumnProjs) {
-          // if the original projection is empty, no need for the additional Project either
-          execution.Filter(cf, scan)
-        } else {
-          execution.Project(projects, execution.Filter(cf, scan))
-        }
-      ).getOrElse(scan) :: Nil
-
-    // TODO: The code for planning bucketed/unbucketed/partitioned/unpartitioned tables contains
-    // a lot of duplication and produces overly complicated RDDs.
-
-    // Scanning non-partitioned HadoopFsRelation
-    case PhysicalOperation(projects, filters, l @ LogicalRelation(t: HadoopFsRelation, _, _)) =>
-      // See buildPartitionedTableScan for the reason that we need to create a shard
-      // broadcast HadoopConf.
-      val sharedHadoopConf = SparkHadoopUtil.get.conf
-      val confBroadcast =
-        t.sqlContext.sparkContext.broadcast(new SerializableConfiguration(sharedHadoopConf))
-
-      t.bucketSpec match {
-        case Some(spec) if t.sqlContext.conf.bucketingEnabled =>
-          val scanBuilder: (Seq[Attribute], Array[Filter]) => RDD[InternalRow] = {
-            (requiredColumns: Seq[Attribute], filters: Array[Filter]) => {
-              val bucketed =
-                t.location
-                  .allFiles()
-                  .filterNot(_.getPath.getName startsWith "_")
-                  .groupBy { f =>
-                    BucketingUtils
-                      .getBucketId(f.getPath.getName)
-                      .getOrElse(sys.error(s"Invalid bucket file ${f.getPath}"))
-                  }
-
-              val bucketedDataMap = bucketed.mapValues { bucketFiles =>
-                t.fileFormat.buildInternalScan(
-                  t.sqlContext,
-                  t.dataSchema,
-                  requiredColumns.map(_.name).toArray,
-                  filters,
-                  None,
-                  bucketFiles,
-                  confBroadcast,
-                  t.options).coalesce(1)
-              }
-
-              val bucketedRDD = new UnionRDD(t.sqlContext.sparkContext,
-                (0 until spec.numBuckets).map { bucketId =>
-                  bucketedDataMap.getOrElse(bucketId, t.sqlContext.emptyResult: RDD[InternalRow])
-                })
-              bucketedRDD
-            }
-          }
-
-          pruneFilterProject(
-            l,
-            projects,
-            filters,
-            scanBuilder) :: Nil
-
-        case _ =>
-          pruneFilterProject(
-            l,
-            projects,
-            filters,
-            (a, f) =>
-              t.fileFormat.buildInternalScan(
-                t.sqlContext,
-                t.dataSchema,
-                a.map(_.name).toArray,
-                f,
-                None,
-                t.location.allFiles(),
-                confBroadcast,
-                t.options)) :: Nil
-      }
-
     case l @ LogicalRelation(baseRelation: TableScan, _, _) =>
       execution.DataSourceScan.create(
         l.output, toCatalystRDD(l, baseRelation.buildScan()), baseRelation) :: Nil
@@ -248,218 +121,6 @@ private[sql] object DataSourceStrategy extends Strategy with Logging {
     case _ => Nil
   }
 
-  private def buildPartitionedTableScan(
-      logicalRelation: LogicalRelation,
-      projections: Seq[NamedExpression],
-      filters: Seq[Expression],
-      buckets: Option[BitSet],
-      partitionColumns: StructType,
-      partitions: Seq[Partition],
-      options: Map[String, String]): SparkPlan = {
-    val relation = logicalRelation.relation.asInstanceOf[HadoopFsRelation]
-
-    // Because we are creating one RDD per partition, we need to have a shared HadoopConf.
-    // Otherwise, the cost of broadcasting HadoopConf in every RDD will be high.
-    val sharedHadoopConf = SparkHadoopUtil.get.conf
-    val confBroadcast =
-      relation.sqlContext.sparkContext.broadcast(new SerializableConfiguration(sharedHadoopConf))
-    val partitionColumnNames = partitionColumns.fieldNames.toSet
-
-    // Now, we create a scan builder, which will be used by pruneFilterProject. This scan builder
-    // will union all partitions and attach partition values if needed.
-    val scanBuilder: (Seq[Attribute], Array[Filter]) => RDD[InternalRow] = {
-      (requiredColumns: Seq[Attribute], filters: Array[Filter]) => {
-
-        relation.bucketSpec match {
-          case Some(spec) if relation.sqlContext.conf.bucketingEnabled =>
-            val requiredDataColumns =
-              requiredColumns.filterNot(c => partitionColumnNames.contains(c.name))
-
-            // Builds RDD[Row]s for each selected partition.
-            val perPartitionRows: Seq[(Int, RDD[InternalRow])] = partitions.flatMap {
-              case Partition(partitionValues, files) =>
-                val bucketed = files.groupBy { f =>
-                  BucketingUtils
-                    .getBucketId(f.getPath.getName)
-                    .getOrElse(sys.error(s"Invalid bucket file ${f.getPath}"))
-                }
-
-                bucketed.map { bucketFiles =>
-                  // Don't scan any partition columns to save I/O.  Here we are being optimistic and
-                  // assuming partition columns data stored in data files are always consistent with
-                  // those partition values encoded in partition directory paths.
-                  val dataRows = relation.fileFormat.buildInternalScan(
-                    relation.sqlContext,
-                    relation.dataSchema,
-                    requiredDataColumns.map(_.name).toArray,
-                    filters,
-                    buckets,
-                    bucketFiles._2,
-                    confBroadcast,
-                    options)
-
-                  // Merges data values with partition values.
-                  bucketFiles._1 -> mergeWithPartitionValues(
-                    requiredColumns,
-                    requiredDataColumns,
-                    partitionColumns,
-                    partitionValues,
-                    dataRows)
-                }
-            }
-
-            val bucketedDataMap: Map[Int, Seq[RDD[InternalRow]]] =
-              perPartitionRows.groupBy(_._1).mapValues(_.map(_._2))
-
-            val bucketed = new UnionRDD(relation.sqlContext.sparkContext,
-              (0 until spec.numBuckets).map { bucketId =>
-                bucketedDataMap.get(bucketId).map(i => i.reduce(_ ++ _).coalesce(1)).getOrElse {
-                  relation.sqlContext.emptyResult: RDD[InternalRow]
-                }
-              })
-            bucketed
-
-          case _ =>
-            val requiredDataColumns =
-              requiredColumns.filterNot(c => partitionColumnNames.contains(c.name))
-
-            // Builds RDD[Row]s for each selected partition.
-            val perPartitionRows = partitions.map {
-              case Partition(partitionValues, files) =>
-                val dataRows = relation.fileFormat.buildInternalScan(
-                  relation.sqlContext,
-                  relation.dataSchema,
-                  requiredDataColumns.map(_.name).toArray,
-                  filters,
-                  buckets,
-                  files,
-                  confBroadcast,
-                  options)
-
-                // Merges data values with partition values.
-                mergeWithPartitionValues(
-                  requiredColumns,
-                  requiredDataColumns,
-                  partitionColumns,
-                  partitionValues,
-                  dataRows)
-            }
-            new UnionRDD(relation.sqlContext.sparkContext, perPartitionRows)
-        }
-      }
-    }
-
-    // Create the scan operator. If needed, add Filter and/or Project on top of the scan.
-    // The added Filter/Project is on top of the unioned RDD. We do not want to create
-    // one Filter/Project for every partition.
-    val sparkPlan = pruneFilterProject(
-      logicalRelation,
-      projections,
-      filters,
-      scanBuilder)
-
-    sparkPlan
-  }
-
-  /**
-   * Creates a ColumnarBatch that contains the values for `requiredColumns`. These columns can
-   * either come from `input` (columns scanned from the data source) or from the partitioning
-   * values (data from `partitionValues`). This is done *once* per physical partition. When
-   * the column is from `input`, it just references the same underlying column. When using
-   * partition columns, the column is populated once.
-   * TODO: there's probably a cleaner way to do this.
-   */
-  private def projectedColumnBatch(
-      input: ColumnarBatch,
-      requiredColumns: Seq[Attribute],
-      dataColumns: Seq[Attribute],
-      partitionColumnSchema: StructType,
-      partitionValues: InternalRow) : ColumnarBatch = {
-    val result = ColumnarBatch.allocate(StructType.fromAttributes(requiredColumns))
-    var resultIdx = 0
-    var inputIdx = 0
-
-    while (resultIdx < requiredColumns.length) {
-      val attr = requiredColumns(resultIdx)
-      if (inputIdx < dataColumns.length && requiredColumns(resultIdx) == dataColumns(inputIdx)) {
-        result.setColumn(resultIdx, input.column(inputIdx))
-        inputIdx += 1
-      } else {
-        require(partitionColumnSchema.fields.count(_.name == attr.name) == 1)
-        var partitionIdx = 0
-        partitionColumnSchema.fields.foreach { f => {
-          if (f.name.equals(attr.name)) {
-            ColumnVectorUtils.populate(result.column(resultIdx), partitionValues, partitionIdx)
-          }
-          partitionIdx += 1
-        }}
-      }
-      resultIdx += 1
-    }
-    result
-  }
-
-  private def mergeWithPartitionValues(
-      requiredColumns: Seq[Attribute],
-      dataColumns: Seq[Attribute],
-      partitionColumnSchema: StructType,
-      partitionValues: InternalRow,
-      dataRows: RDD[InternalRow]): RDD[InternalRow] = {
-    // If output columns contain any partition column(s), we need to merge scanned data
-    // columns and requested partition columns to form the final result.
-    if (requiredColumns != dataColumns) {
-      // Builds `AttributeReference`s for all partition columns so that we can use them to project
-      // required partition columns.  Note that if a partition column appears in `requiredColumns`,
-      // we should use the `AttributeReference` in `requiredColumns`.
-      val partitionColumns = {
-        val requiredColumnMap = requiredColumns.map(a => a.name -> a).toMap
-        partitionColumnSchema.toAttributes.map { a =>
-          requiredColumnMap.getOrElse(a.name, a)
-        }
-      }
-
-      val mapPartitionsFunc = (_: TaskContext, _: Int, iterator: Iterator[Object]) => {
-        // Note that we can't use an `UnsafeRowJoiner` to replace the following `JoinedRow` and
-        // `UnsafeProjection`.  Because the projection may also adjust column order.
-        val mutableJoinedRow = new JoinedRow()
-        val unsafePartitionValues = UnsafeProjection.create(partitionColumnSchema)(partitionValues)
-        val unsafeProjection =
-          UnsafeProjection.create(requiredColumns, dataColumns ++ partitionColumns)
-
-        // If we are returning batches directly, we need to augment them with the partitioning
-        // columns. We want to do this without a row by row operation.
-        var columnBatch: ColumnarBatch = null
-        var mergedBatch: ColumnarBatch = null
-
-        iterator.map { input => {
-          if (input.isInstanceOf[InternalRow]) {
-            unsafeProjection(mutableJoinedRow(
-              input.asInstanceOf[InternalRow], unsafePartitionValues))
-          } else {
-            require(input.isInstanceOf[ColumnarBatch])
-            val inputBatch = input.asInstanceOf[ColumnarBatch]
-            if (inputBatch != mergedBatch) {
-              mergedBatch = inputBatch
-              columnBatch = projectedColumnBatch(inputBatch, requiredColumns,
-                dataColumns, partitionColumnSchema, partitionValues)
-            }
-            columnBatch.setNumRows(inputBatch.numRows())
-            columnBatch
-          }
-        }}
-      }
-
-      // This is an internal RDD whose call site the user should not be concerned with
-      // Since we create many of these (one per partition), the time spent on computing
-      // the call site may add up.
-      Utils.withDummyCallSite(dataRows.sparkContext) {
-        new MapPartitionsRDD(dataRows, mapPartitionsFunc, preservesPartitioning = false)
-      }.asInstanceOf[RDD[InternalRow]]
-    } else {
-      dataRows
-    }
-  }
-
   // Get the bucket ID based on the bucketing values.
   // Restriction: Bucket pruning works iff the bucketing column has one and only one column.
   def getBucketId(bucketColumn: Attribute, numBuckets: Int, value: Any): Int = {
@@ -472,57 +133,6 @@ private[sql] object DataSourceStrategy extends Strategy with Logging {
     bucketIdGeneration(mutableRow).getInt(0)
   }
 
-  // Get the bucket BitSet by reading the filters that only contains bucketing keys.
-  // Note: When the returned BitSet is None, no pruning is possible.
-  // Restriction: Bucket pruning works iff the bucketing column has one and only one column.
-  private def getBuckets(
-      filters: Seq[Expression],
-      bucketSpec: Option[BucketSpec]): Option[BitSet] = {
-
-    if (bucketSpec.isEmpty ||
-      bucketSpec.get.numBuckets == 1 ||
-      bucketSpec.get.bucketColumnNames.length != 1) {
-      // None means all the buckets need to be scanned
-      return None
-    }
-
-    // Just get the first because bucketing pruning only works when the column has one column
-    val bucketColumnName = bucketSpec.get.bucketColumnNames.head
-    val numBuckets = bucketSpec.get.numBuckets
-    val matchedBuckets = new BitSet(numBuckets)
-    matchedBuckets.clear()
-
-    filters.foreach {
-      case expressions.EqualTo(a: Attribute, Literal(v, _)) if a.name == bucketColumnName =>
-        matchedBuckets.set(getBucketId(a, numBuckets, v))
-      case expressions.EqualTo(Literal(v, _), a: Attribute) if a.name == bucketColumnName =>
-        matchedBuckets.set(getBucketId(a, numBuckets, v))
-      case expressions.EqualNullSafe(a: Attribute, Literal(v, _)) if a.name == bucketColumnName =>
-        matchedBuckets.set(getBucketId(a, numBuckets, v))
-      case expressions.EqualNullSafe(Literal(v, _), a: Attribute) if a.name == bucketColumnName =>
-        matchedBuckets.set(getBucketId(a, numBuckets, v))
-      // Because we only convert In to InSet in Optimizer when there are more than certain
-      // items. So it is possible we still get an In expression here that needs to be pushed
-      // down.
-      case expressions.In(a: Attribute, list)
-          if list.forall(_.isInstanceOf[Literal]) && a.name == bucketColumnName =>
-        val hSet = list.map(e => e.eval(EmptyRow))
-        hSet.foreach(e => matchedBuckets.set(getBucketId(a, numBuckets, e)))
-      case expressions.IsNull(a: Attribute) if a.name == bucketColumnName =>
-        matchedBuckets.set(getBucketId(a, numBuckets, null))
-      case _ =>
-    }
-
-    logInfo {
-      val selected = matchedBuckets.cardinality()
-      val percentPruned = (1 - selected.toDouble / numBuckets.toDouble) * 100
-      s"Selected $selected buckets out of $numBuckets, pruned $percentPruned% partitions."
-    }
-
-    // None means all the buckets need to be scanned
-    if (matchedBuckets.cardinality() == 0) None else Some(matchedBuckets)
-  }
-
   // Based on Public API.
   protected def pruneFilterProject(
       relation: LogicalRelation,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala
index aa1f76450c..bcddf72851 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala
@@ -55,15 +55,7 @@ import org.apache.spark.sql.sources._
  */
 private[sql] object FileSourceStrategy extends Strategy with Logging {
   def apply(plan: LogicalPlan): Seq[SparkPlan] = plan match {
-    case PhysicalOperation(projects, filters, l @ LogicalRelation(files: HadoopFsRelation, _, _))
-      if (files.fileFormat.toString == "TestFileFormat" ||
-         files.fileFormat.isInstanceOf[parquet.DefaultSource] ||
-         files.fileFormat.toString == "ORC" ||
-         files.fileFormat.toString == "LibSVM" ||
-         files.fileFormat.isInstanceOf[csv.DefaultSource] ||
-         files.fileFormat.isInstanceOf[text.DefaultSource] ||
-         files.fileFormat.isInstanceOf[json.DefaultSource]) &&
-         files.sqlContext.conf.useFileScan =>
+    case PhysicalOperation(projects, filters, l @ LogicalRelation(files: HadoopFsRelation, _, _)) =>
       // Filters on this relation fall into four categories based on where we can use them to avoid
       // reading unneeded data:
       //  - partition keys only - used to prune directories to read
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala
index 34fcbdf871..06a371b88b 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/DefaultSource.scala
@@ -133,37 +133,6 @@ class DefaultSource extends FileFormat with DataSourceRegister {
     }
   }
 
-  /**
-   * This supports to eliminate unneeded columns before producing an RDD
-   * containing all of its tuples as Row objects. This reads all the tokens of each line
-   * and then drop unneeded tokens without casting and type-checking by mapping
-   * both the indices produced by `requiredColumns` and the ones of tokens.
-   */
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    // TODO: Filter before calling buildInternalScan.
-    val csvFiles = inputFiles.filterNot(_.getPath.getName startsWith "_")
-
-    val csvOptions = new CSVOptions(options)
-    val pathsString = csvFiles.map(_.getPath.toUri.toString)
-    val header = dataSchema.fields.map(_.name)
-    val tokenizedRdd = tokenRdd(sqlContext, csvOptions, header, pathsString)
-    val rows = CSVRelation.parseCsv(tokenizedRdd, dataSchema, requiredColumns, csvOptions)
-
-    val requiredDataSchema = StructType(requiredColumns.map(c => dataSchema.find(_.name == c).get))
-    rows.mapPartitions { iterator =>
-      val unsafeProjection = UnsafeProjection.create(requiredDataSchema)
-      iterator.map(unsafeProjection)
-    }
-  }
-
   private def baseRdd(
       sqlContext: SQLContext,
       options: CSVOptions,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala
index 42cd25a18c..f32fea4183 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala
@@ -93,35 +93,6 @@ class DefaultSource extends FileFormat with DataSourceRegister {
     }
   }
 
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    // TODO: Filter files for all formats before calling buildInternalScan.
-    val jsonFiles = inputFiles.filterNot(_.getPath.getName startsWith "_")
-
-    val parsedOptions: JSONOptions = new JSONOptions(options)
-    val requiredDataSchema = StructType(requiredColumns.map(dataSchema(_)))
-    val columnNameOfCorruptRecord =
-      parsedOptions.columnNameOfCorruptRecord
-        .getOrElse(sqlContext.conf.columnNameOfCorruptRecord)
-    val rows = JacksonParser.parse(
-      createBaseRdd(sqlContext, jsonFiles),
-      requiredDataSchema,
-      columnNameOfCorruptRecord,
-      parsedOptions)
-
-    rows.mapPartitions { iterator =>
-      val unsafeProjection = UnsafeProjection.create(requiredDataSchema)
-      iterator.map(unsafeProjection)
-    }
-  }
-
   override def buildReader(
       sqlContext: SQLContext,
       dataSchema: StructType,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
index bcb2b2de13..dbda094996 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
@@ -251,12 +251,12 @@ private[sql] class DefaultSource
   }
 
   /**
-   * Returns whether the reader will the rows as batch or not.
+   * Returns whether the reader will return the rows as batch or not.
    */
   override def supportBatch(sqlContext: SQLContext, schema: StructType): Boolean = {
     val conf = SQLContext.getActive().get.conf
-    conf.useFileScan && conf.parquetVectorizedReaderEnabled &&
-      conf.wholeStageEnabled && schema.length <= conf.wholeStageMaxNumFields &&
+    conf.parquetVectorizedReaderEnabled && conf.wholeStageEnabled &&
+      schema.length <= conf.wholeStageMaxNumFields &&
       schema.forall(_.dataType.isInstanceOf[AtomicType])
   }
 
@@ -375,110 +375,6 @@ private[sql] class DefaultSource
       }
     }
   }
-
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      allFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    val useMetadataCache = sqlContext.getConf(SQLConf.PARQUET_CACHE_METADATA)
-    val parquetFilterPushDown = sqlContext.conf.parquetFilterPushDown
-    val assumeBinaryIsString = sqlContext.conf.isParquetBinaryAsString
-    val assumeInt96IsTimestamp = sqlContext.conf.isParquetINT96AsTimestamp
-
-    // Parquet row group size. We will use this value as the value for
-    // mapreduce.input.fileinputformat.split.minsize and mapred.min.split.size if the value
-    // of these flags are smaller than the parquet row group size.
-    val parquetBlockSize = ParquetOutputFormat.getLongBlockSize(broadcastedConf.value.value)
-
-    // Create the function to set variable Parquet confs at both driver and executor side.
-    val initLocalJobFuncOpt =
-      ParquetRelation.initializeLocalJobFunc(
-        requiredColumns,
-        filters,
-        dataSchema,
-        parquetBlockSize,
-        useMetadataCache,
-        parquetFilterPushDown,
-        assumeBinaryIsString,
-        assumeInt96IsTimestamp) _
-
-    val inputFiles = splitFiles(allFiles).data.toArray
-
-    // Create the function to set input paths at the driver side.
-    val setInputPaths =
-      ParquetRelation.initializeDriverSideJobFunc(inputFiles, parquetBlockSize) _
-
-    val allPrimitiveTypes = dataSchema.forall(_.dataType.isInstanceOf[AtomicType])
-    val inputFormatCls = if (sqlContext.conf.parquetVectorizedReaderEnabled
-      && allPrimitiveTypes) {
-      classOf[VectorizedParquetInputFormat]
-    } else {
-      classOf[ParquetInputFormat[InternalRow]]
-    }
-
-    Utils.withDummyCallSite(sqlContext.sparkContext) {
-      new SqlNewHadoopRDD(
-        sqlContext = sqlContext,
-        broadcastedConf = broadcastedConf,
-        initDriverSideJobFuncOpt = Some(setInputPaths),
-        initLocalJobFuncOpt = Some(initLocalJobFuncOpt),
-        inputFormatClass = inputFormatCls,
-        valueClass = classOf[InternalRow]) {
-
-        val cacheMetadata = useMetadataCache
-
-        @transient val cachedStatuses = inputFiles.map { f =>
-          // In order to encode the authority of a Path containing special characters such as '/'
-          // (which does happen in some S3N credentials), we need to use the string returned by the
-          // URI of the path to create a new Path.
-          val pathWithEscapedAuthority = escapePathUserInfo(f.getPath)
-          new FileStatus(
-            f.getLen, f.isDirectory, f.getReplication, f.getBlockSize, f.getModificationTime,
-            f.getAccessTime, f.getPermission, f.getOwner, f.getGroup, pathWithEscapedAuthority)
-        }.toSeq
-
-        private def escapePathUserInfo(path: Path): Path = {
-          val uri = path.toUri
-          new Path(new URI(
-            uri.getScheme, uri.getRawUserInfo, uri.getHost, uri.getPort, uri.getPath,
-            uri.getQuery, uri.getFragment))
-        }
-
-        // Overridden so we can inject our own cached files statuses.
-        override def getPartitions: Array[SparkPartition] = {
-          val inputFormat = new ParquetInputFormat[InternalRow] {
-            override def listStatus(jobContext: JobContext): JList[FileStatus] = {
-              if (cacheMetadata) cachedStatuses.asJava else super.listStatus(jobContext)
-            }
-          }
-
-          val jobContext = new JobContextImpl(getConf(isDriverSide = true), jobId)
-          val rawSplits = inputFormat.getSplits(jobContext)
-
-          Array.tabulate[SparkPartition](rawSplits.size) { i =>
-            new SqlNewHadoopPartition(
-              id, i, rawSplits.get(i).asInstanceOf[InputSplit with Writable])
-          }
-        }
-      }
-    }
-  }
-}
-
-/**
- * The ParquetInputFormat that create VectorizedParquetRecordReader.
- */
-final class VectorizedParquetInputFormat extends ParquetInputFormat[InternalRow] {
-  override def createRecordReader(
-    inputSplit: InputSplit,
-    taskAttemptContext: TaskAttemptContext): RecordReader[Void, InternalRow] = {
-    new VectorizedParquetRecordReader().asInstanceOf[RecordReader[Void, InternalRow]]
-  }
 }
 
 // NOTE: This class is instantiated and used on executor side only, no need to be serializable.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala
index 99459ba1d3..28b03ee7c3 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala
@@ -88,45 +88,6 @@ class DefaultSource extends FileFormat with DataSourceRegister {
     }
   }
 
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    verifySchema(dataSchema)
-
-    val job = Job.getInstance(sqlContext.sparkContext.hadoopConfiguration)
-    val conf = job.getConfiguration
-    val paths = inputFiles
-        .filterNot(_.getPath.getName startsWith "_")
-        .map(_.getPath)
-        .sortBy(_.toUri)
-
-    if (paths.nonEmpty) {
-      FileInputFormat.setInputPaths(job, paths: _*)
-    }
-
-    sqlContext.sparkContext.hadoopRDD(
-      conf.asInstanceOf[JobConf], classOf[TextInputFormat], classOf[LongWritable], classOf[Text])
-        .mapPartitions { iter =>
-          val unsafeRow = new UnsafeRow(1)
-          val bufferHolder = new BufferHolder(unsafeRow)
-          val unsafeRowWriter = new UnsafeRowWriter(bufferHolder, 1)
-
-          iter.map { case (_, line) =>
-            // Writes to an UnsafeRow directly
-            bufferHolder.reset()
-            unsafeRowWriter.write(0, line.getBytes, 0, line.getLength)
-            unsafeRow.setTotalSize(bufferHolder.totalSize())
-            unsafeRow
-          }
-        }
-  }
-
   override def buildReader(
       sqlContext: SQLContext,
       dataSchema: StructType,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala b/sql/core/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala
index b58f960897..e74fb00cb2 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala
@@ -145,12 +145,6 @@ object SQLConf {
     .booleanConf
     .createWithDefault(true)
 
-  val USE_FILE_SCAN = SQLConfigBuilder("spark.sql.sources.fileScan")
-    .internal()
-    .doc("Use the new FileScanRDD path for reading HDSF based data sources.")
-    .booleanConf
-    .createWithDefault(true)
-
   val PARQUET_SCHEMA_MERGING_ENABLED = SQLConfigBuilder("spark.sql.parquet.mergeSchema")
     .doc("When true, the Parquet data source merges schemas collected from all data files, " +
          "otherwise the schema is picked from the summary file or a random data file " +
@@ -481,8 +475,6 @@ private[sql] class SQLConf extends Serializable with CatalystConf with Logging {
 
   def useCompression: Boolean = getConf(COMPRESS_CACHED)
 
-  def useFileScan: Boolean = getConf(USE_FILE_SCAN)
-
   def parquetCompressionCodec: String = getConf(PARQUET_COMPRESSION)
 
   def parquetCacheMetadata: Boolean = getConf(PARQUET_CACHE_METADATA)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
index 6acb41dd1f..65b1f61349 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
@@ -458,16 +458,6 @@ trait FileFormat {
       options: Map[String, String],
       dataSchema: StructType): OutputWriterFactory
 
-  def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow]
-
   /**
    * Returns whether this format support returning columnar batch or not.
    *
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategySuite.scala
index 41f536fc37..90d7f53884 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategySuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategySuite.scala
@@ -365,18 +365,6 @@ class TestFileFormat extends FileFormat {
     throw new NotImplementedError("JUST FOR TESTING")
   }
 
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    throw new NotImplementedError("JUST FOR TESTING")
-  }
-
   override def buildReader(
       sqlContext: SQLContext,
       dataSchema: StructType,
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala
index 43f445edcb..e915f3dfe2 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala
@@ -111,19 +111,6 @@ private[sql] class DefaultSource
     }
   }
 
-  override def buildInternalScan(
-      sqlContext: SQLContext,
-      dataSchema: StructType,
-      requiredColumns: Array[String],
-      filters: Array[Filter],
-      bucketSet: Option[BitSet],
-      inputFiles: Seq[FileStatus],
-      broadcastedConf: Broadcast[SerializableConfiguration],
-      options: Map[String, String]): RDD[InternalRow] = {
-    val output = StructType(requiredColumns.map(dataSchema(_))).toAttributes
-    OrcTableScan(sqlContext, output, filters, inputFiles).execute()
-  }
-
   override def buildReader(
       sqlContext: SQLContext,
       dataSchema: StructType,
-- 
cgit v1.2.3


From b0f5497e9520575e5082fa8ce8be5569f43abe74 Mon Sep 17 00:00:00 2001
From: Dongjoon Hyun <dongjoon@apache.org>
Date: Tue, 12 Apr 2016 00:43:28 -0700
Subject: [SPARK-14508][BUILD] Add a new ScalaStyle Rule `OmitBracesInCase`

## What changes were proposed in this pull request?

According to the [Spark Code Style Guide](https://cwiki.apache.org/confluence/display/SPARK/Spark+Code+Style+Guide) and [Scala Style Guide](http://docs.scala-lang.org/style/control-structures.html#curlybraces), we had better enforce the following rule.
  ```
  case: Always omit braces in case clauses.
  ```
This PR makes a new ScalaStyle rule, 'OmitBracesInCase', and enforces it to the code.

## How was this patch tested?

Pass the Jenkins tests (including Scala style checking)

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #12280 from dongjoon-hyun/SPARK-14508.
---
 .../main/scala/org/apache/spark/SparkContext.scala | 12 ++----
 .../src/main/scala/org/apache/spark/SparkEnv.scala |  3 +-
 .../apache/spark/api/python/PythonHadoopUtil.scala |  3 +-
 .../org/apache/spark/api/python/PythonRDD.scala    |  2 +-
 .../org/apache/spark/deploy/SparkHadoopUtil.scala  | 12 ++----
 .../org/apache/spark/deploy/master/Master.scala    | 48 ++++++++--------------
 .../spark/deploy/master/MasterArguments.scala      |  2 +-
 .../deploy/master/ZooKeeperPersistenceEngine.scala |  3 +-
 .../mesos/MesosClusterDispatcherArguments.scala    |  3 +-
 .../spark/deploy/worker/ExecutorRunner.scala       |  6 +--
 .../org/apache/spark/deploy/worker/Worker.scala    | 12 ++----
 .../spark/deploy/worker/WorkerArguments.scala      |  3 +-
 .../executor/CoarseGrainedExecutorBackend.scala    |  3 +-
 .../org/apache/spark/metrics/MetricsSystem.scala   |  3 +-
 .../org/apache/spark/partial/BoundedDouble.scala   |  3 +-
 .../org/apache/spark/rdd/DoubleRDDFunctions.scala  |  4 +-
 .../org/apache/spark/rdd/OrderedRDDFunctions.scala |  3 +-
 .../apache/spark/rdd/ParallelCollectionRDD.scala   |  9 ++--
 .../spark/rdd/PartitionerAwareUnionRDD.scala       |  3 +-
 .../apache/spark/scheduler/InputFormatInfo.scala   |  6 +--
 .../org/apache/spark/scheduler/SplitInfo.scala     |  3 +-
 .../apache/spark/scheduler/TaskResultGetter.scala  |  2 +-
 .../apache/spark/scheduler/TaskSetManager.scala    |  9 ++--
 .../mesos/MesosClusterPersistenceEngine.scala      |  3 +-
 .../cluster/mesos/MesosSchedulerBackendUtil.scala  |  6 +--
 .../cluster/mesos/MesosSchedulerUtils.scala        |  6 +--
 .../org/apache/spark/serializer/Serializer.scala   |  3 +-
 .../storage/ShuffleBlockFetcherIterator.scala      |  6 +--
 .../spark/ui/exec/ExecutorThreadDumpPage.scala     |  3 +-
 .../scala/org/apache/spark/util/EventLoop.scala    |  3 +-
 .../org/apache/spark/util/SizeEstimator.scala      |  3 +-
 .../scala/org/apache/spark/DistributedSuite.scala  |  2 +-
 .../org/apache/spark/SparkContextInfoSuite.scala   |  6 +--
 .../scala/org/apache/spark/UnpersistSuite.scala    |  2 +-
 .../scala/org/apache/spark/rpc/RpcEnvSuite.scala   | 15 +++----
 .../apache/spark/examples/CassandraCQLTest.scala   |  6 +--
 .../org/apache/spark/examples/CassandraTest.scala  |  6 +--
 .../scala/org/apache/spark/examples/LocalALS.scala |  6 +--
 .../spark/examples/ml/OneVsRestExample.scala       |  6 +--
 .../spark/examples/mllib/DecisionTreeRunner.scala  |  6 +--
 .../streaming/kinesis/KinesisRecordProcessor.scala | 25 +++++------
 .../org/apache/spark/ml/r/SparkRWrappers.scala     |  6 +--
 .../spark/mllib/clustering/GaussianMixture.scala   |  3 +-
 .../mllib/clustering/GaussianMixtureModel.scala    |  3 +-
 .../org/apache/spark/mllib/clustering/KMeans.scala |  6 +--
 .../mllib/stat/test/KolmogorovSmirnovTest.scala    |  3 +-
 .../apache/spark/repl/ExecutorClassLoader.scala    | 36 ++++++++--------
 scalastyle-config.xml                              |  5 +++
 .../spark/sql/catalyst/expressions/Cast.scala      |  3 +-
 .../expressions/EquivalentExpressions.scala        |  4 +-
 .../org/apache/spark/sql/RandomDataGenerator.scala | 18 +++-----
 .../org/apache/spark/sql/DataFrameSuite.scala      |  9 ++--
 .../execution/vectorized/ColumnarBatchSuite.scala  |  8 +---
 .../apache/spark/streaming/dstream/DStream.scala   |  3 +-
 .../streaming/dstream/DStreamCheckpointData.scala  |  3 +-
 .../spark/streaming/dstream/FileInputDStream.scala |  3 +-
 .../spark/streaming/dstream/StateDStream.scala     | 26 ++++--------
 .../spark/streaming/BasicOperationsSuite.scala     |  3 +-
 .../apache/spark/streaming/CheckpointSuite.scala   |  6 +--
 .../apache/spark/streaming/MasterFailureTest.scala |  3 +-
 .../spark/deploy/yarn/ApplicationMaster.scala      |  3 +-
 .../apache/spark/deploy/yarn/YarnAllocator.scala   |  3 +-
 .../scheduler/cluster/YarnSchedulerBackend.scala   |  6 +--
 .../deploy/yarn/YarnSparkHadoopUtilSuite.scala     | 24 ++++-------
 64 files changed, 164 insertions(+), 293 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala
index f0d152f05a..966198dd5e 100644
--- a/core/src/main/scala/org/apache/spark/SparkContext.scala
+++ b/core/src/main/scala/org/apache/spark/SparkContext.scala
@@ -2397,9 +2397,8 @@ object SparkContext extends Logging {
         } catch {
           // TODO: Enumerate the exact reasons why it can fail
           // But irrespective of it, it means we cannot proceed !
-          case e: Exception => {
+          case e: Exception =>
             throw new SparkException("YARN mode not available ?", e)
-          }
         }
         val backend = try {
           val clazz =
@@ -2407,9 +2406,8 @@ object SparkContext extends Logging {
           val cons = clazz.getConstructor(classOf[TaskSchedulerImpl], classOf[SparkContext])
           cons.newInstance(scheduler, sc).asInstanceOf[CoarseGrainedSchedulerBackend]
         } catch {
-          case e: Exception => {
+          case e: Exception =>
             throw new SparkException("YARN mode not available ?", e)
-          }
         }
         scheduler.initialize(backend)
         (backend, scheduler)
@@ -2421,9 +2419,8 @@ object SparkContext extends Logging {
           cons.newInstance(sc).asInstanceOf[TaskSchedulerImpl]
 
         } catch {
-          case e: Exception => {
+          case e: Exception =>
             throw new SparkException("YARN mode not available ?", e)
-          }
         }
 
         val backend = try {
@@ -2432,9 +2429,8 @@ object SparkContext extends Logging {
           val cons = clazz.getConstructor(classOf[TaskSchedulerImpl], classOf[SparkContext])
           cons.newInstance(scheduler, sc).asInstanceOf[CoarseGrainedSchedulerBackend]
         } catch {
-          case e: Exception => {
+          case e: Exception =>
             throw new SparkException("YARN mode not available ?", e)
-          }
         }
 
         scheduler.initialize(backend)
diff --git a/core/src/main/scala/org/apache/spark/SparkEnv.scala b/core/src/main/scala/org/apache/spark/SparkEnv.scala
index ab89f4c4e4..3d11db7461 100644
--- a/core/src/main/scala/org/apache/spark/SparkEnv.scala
+++ b/core/src/main/scala/org/apache/spark/SparkEnv.scala
@@ -101,14 +101,13 @@ class SparkEnv (
       // We only need to delete the tmp dir create by driver, because sparkFilesDir is point to the
       // current working dir in executor which we do not need to delete.
       driverTmpDirToDelete match {
-        case Some(path) => {
+        case Some(path) =>
           try {
             Utils.deleteRecursively(new File(path))
           } catch {
             case e: Exception =>
               logWarning(s"Exception while deleting Spark temp dir: $path", e)
           }
-        }
         case None => // We just need to delete tmp dir created by driver, so do nothing on executor
       }
     }
diff --git a/core/src/main/scala/org/apache/spark/api/python/PythonHadoopUtil.scala b/core/src/main/scala/org/apache/spark/api/python/PythonHadoopUtil.scala
index 6f6730690f..6259bead3e 100644
--- a/core/src/main/scala/org/apache/spark/api/python/PythonHadoopUtil.scala
+++ b/core/src/main/scala/org/apache/spark/api/python/PythonHadoopUtil.scala
@@ -134,11 +134,10 @@ private[python] class JavaToWritableConverter extends Converter[Any, Writable] {
           mapWritable.put(convertToWritable(k), convertToWritable(v))
         }
         mapWritable
-      case array: Array[Any] => {
+      case array: Array[Any] =>
         val arrayWriteable = new ArrayWritable(classOf[Writable])
         arrayWriteable.set(array.map(convertToWritable(_)))
         arrayWriteable
-      }
       case other => throw new SparkException(
         s"Data of type ${other.getClass.getName} cannot be used")
     }
diff --git a/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala b/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
index 4bca16a234..ab5b6c8380 100644
--- a/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
+++ b/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
@@ -470,7 +470,7 @@ private[spark] object PythonRDD extends Logging {
           objs.append(obj)
         }
       } catch {
-        case eof: EOFException => {}
+        case eof: EOFException => // No-op
       }
       JavaRDD.fromRDD(sc.sc.parallelize(objs, parallelism))
     } finally {
diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala b/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala
index 41ac308808..cda9d38c6a 100644
--- a/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala
@@ -152,10 +152,9 @@ class SparkHadoopUtil extends Logging {
       val baselineBytesRead = f()
       Some(() => f() - baselineBytesRead)
     } catch {
-      case e @ (_: NoSuchMethodException | _: ClassNotFoundException) => {
+      case e @ (_: NoSuchMethodException | _: ClassNotFoundException) =>
         logDebug("Couldn't find method for retrieving thread-level FileSystem input data", e)
         None
-      }
     }
   }
 
@@ -174,10 +173,9 @@ class SparkHadoopUtil extends Logging {
       val baselineBytesWritten = f()
       Some(() => f() - baselineBytesWritten)
     } catch {
-      case e @ (_: NoSuchMethodException | _: ClassNotFoundException) => {
+      case e @ (_: NoSuchMethodException | _: ClassNotFoundException) =>
         logDebug("Couldn't find method for retrieving thread-level FileSystem output data", e)
         None
-      }
     }
   }
 
@@ -315,7 +313,7 @@ class SparkHadoopUtil extends Logging {
    */
   def substituteHadoopVariables(text: String, hadoopConf: Configuration): String = {
     text match {
-      case HADOOP_CONF_PATTERN(matched) => {
+      case HADOOP_CONF_PATTERN(matched) =>
         logDebug(text + " matched " + HADOOP_CONF_PATTERN)
         val key = matched.substring(13, matched.length() - 1) // remove ${hadoopconf- .. }
         val eval = Option[String](hadoopConf.get(key))
@@ -330,11 +328,9 @@ class SparkHadoopUtil extends Logging {
           // Continue to substitute more variables.
           substituteHadoopVariables(eval.get, hadoopConf)
         }
-      }
-      case _ => {
+      case _ =>
         logDebug(text + " didn't match " + HADOOP_CONF_PATTERN)
         text
-      }
     }
   }
 
diff --git a/core/src/main/scala/org/apache/spark/deploy/master/Master.scala b/core/src/main/scala/org/apache/spark/deploy/master/Master.scala
index 01901bbf85..9bd3fc1033 100644
--- a/core/src/main/scala/org/apache/spark/deploy/master/Master.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/master/Master.scala
@@ -217,7 +217,7 @@ private[deploy] class Master(
   }
 
   override def receive: PartialFunction[Any, Unit] = {
-    case ElectedLeader => {
+    case ElectedLeader =>
       val (storedApps, storedDrivers, storedWorkers) = persistenceEngine.readPersistedData(rpcEnv)
       state = if (storedApps.isEmpty && storedDrivers.isEmpty && storedWorkers.isEmpty) {
         RecoveryState.ALIVE
@@ -233,16 +233,14 @@ private[deploy] class Master(
           }
         }, WORKER_TIMEOUT_MS, TimeUnit.MILLISECONDS)
       }
-    }
 
     case CompleteRecovery => completeRecovery()
 
-    case RevokedLeadership => {
+    case RevokedLeadership =>
       logError("Leadership has been revoked -- master shutting down.")
       System.exit(0)
-    }
 
-    case RegisterApplication(description, driver) => {
+    case RegisterApplication(description, driver) =>
       // TODO Prevent repeated registrations from some driver
       if (state == RecoveryState.STANDBY) {
         // ignore, don't send response
@@ -255,12 +253,11 @@ private[deploy] class Master(
         driver.send(RegisteredApplication(app.id, self))
         schedule()
       }
-    }
 
-    case ExecutorStateChanged(appId, execId, state, message, exitStatus) => {
+    case ExecutorStateChanged(appId, execId, state, message, exitStatus) =>
       val execOption = idToApp.get(appId).flatMap(app => app.executors.get(execId))
       execOption match {
-        case Some(exec) => {
+        case Some(exec) =>
           val appInfo = idToApp(appId)
           val oldState = exec.state
           exec.state = state
@@ -298,22 +295,19 @@ private[deploy] class Master(
               }
             }
           }
-        }
         case None =>
           logWarning(s"Got status update for unknown executor $appId/$execId")
       }
-    }
 
-    case DriverStateChanged(driverId, state, exception) => {
+    case DriverStateChanged(driverId, state, exception) =>
       state match {
         case DriverState.ERROR | DriverState.FINISHED | DriverState.KILLED | DriverState.FAILED =>
           removeDriver(driverId, state, exception)
         case _ =>
           throw new Exception(s"Received unexpected state update for driver $driverId: $state")
       }
-    }
 
-    case Heartbeat(workerId, worker) => {
+    case Heartbeat(workerId, worker) =>
       idToWorker.get(workerId) match {
         case Some(workerInfo) =>
           workerInfo.lastHeartbeat = System.currentTimeMillis()
@@ -327,9 +321,8 @@ private[deploy] class Master(
               " This worker was never registered, so ignoring the heartbeat.")
           }
       }
-    }
 
-    case MasterChangeAcknowledged(appId) => {
+    case MasterChangeAcknowledged(appId) =>
       idToApp.get(appId) match {
         case Some(app) =>
           logInfo("Application has been re-registered: " + appId)
@@ -339,9 +332,8 @@ private[deploy] class Master(
       }
 
       if (canCompleteRecovery) { completeRecovery() }
-    }
 
-    case WorkerSchedulerStateResponse(workerId, executors, driverIds) => {
+    case WorkerSchedulerStateResponse(workerId, executors, driverIds) =>
       idToWorker.get(workerId) match {
         case Some(worker) =>
           logInfo("Worker has been re-registered: " + workerId)
@@ -367,7 +359,6 @@ private[deploy] class Master(
       }
 
       if (canCompleteRecovery) { completeRecovery() }
-    }
 
     case WorkerLatestState(workerId, executors, driverIds) =>
       idToWorker.get(workerId) match {
@@ -397,9 +388,8 @@ private[deploy] class Master(
       logInfo(s"Received unregister request from application $applicationId")
       idToApp.get(applicationId).foreach(finishApplication)
 
-    case CheckForWorkerTimeOut => {
+    case CheckForWorkerTimeOut =>
       timeOutDeadWorkers()
-    }
 
     case AttachCompletedRebuildUI(appId) =>
       // An asyncRebuildSparkUI has completed, so need to attach to master webUi
@@ -408,7 +398,7 @@ private[deploy] class Master(
 
   override def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
     case RegisterWorker(
-        id, workerHost, workerPort, workerRef, cores, memory, workerWebUiUrl) => {
+        id, workerHost, workerPort, workerRef, cores, memory, workerWebUiUrl) =>
       logInfo("Registering worker %s:%d with %d cores, %s RAM".format(
         workerHost, workerPort, cores, Utils.megabytesToString(memory)))
       if (state == RecoveryState.STANDBY) {
@@ -430,9 +420,8 @@ private[deploy] class Master(
             + workerAddress))
         }
       }
-    }
 
-    case RequestSubmitDriver(description) => {
+    case RequestSubmitDriver(description) =>
       if (state != RecoveryState.ALIVE) {
         val msg = s"${Utils.BACKUP_STANDALONE_MASTER_PREFIX}: $state. " +
           "Can only accept driver submissions in ALIVE state."
@@ -451,9 +440,8 @@ private[deploy] class Master(
         context.reply(SubmitDriverResponse(self, true, Some(driver.id),
           s"Driver successfully submitted as ${driver.id}"))
       }
-    }
 
-    case RequestKillDriver(driverId) => {
+    case RequestKillDriver(driverId) =>
       if (state != RecoveryState.ALIVE) {
         val msg = s"${Utils.BACKUP_STANDALONE_MASTER_PREFIX}: $state. " +
           s"Can only kill drivers in ALIVE state."
@@ -484,9 +472,8 @@ private[deploy] class Master(
             context.reply(KillDriverResponse(self, driverId, success = false, msg))
         }
       }
-    }
 
-    case RequestDriverStatus(driverId) => {
+    case RequestDriverStatus(driverId) =>
       if (state != RecoveryState.ALIVE) {
         val msg = s"${Utils.BACKUP_STANDALONE_MASTER_PREFIX}: $state. " +
           "Can only request driver status in ALIVE state."
@@ -501,18 +488,15 @@ private[deploy] class Master(
             context.reply(DriverStatusResponse(found = false, None, None, None, None))
         }
       }
-    }
 
-    case RequestMasterState => {
+    case RequestMasterState =>
       context.reply(MasterStateResponse(
         address.host, address.port, restServerBoundPort,
         workers.toArray, apps.toArray, completedApps.toArray,
         drivers.toArray, completedDrivers.toArray, state))
-    }
 
-    case BoundPortsRequest => {
+    case BoundPortsRequest =>
       context.reply(BoundPortsResponse(address.port, webUi.boundPort, restServerBoundPort))
-    }
 
     case RequestExecutors(appId, requestedTotal) =>
       context.reply(handleRequestExecutors(appId, requestedTotal))
diff --git a/core/src/main/scala/org/apache/spark/deploy/master/MasterArguments.scala b/core/src/main/scala/org/apache/spark/deploy/master/MasterArguments.scala
index 9cd7458ba0..585e0839d0 100644
--- a/core/src/main/scala/org/apache/spark/deploy/master/MasterArguments.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/master/MasterArguments.scala
@@ -78,7 +78,7 @@ private[master] class MasterArguments(args: Array[String], conf: SparkConf) {
     case ("--help") :: tail =>
       printUsageAndExit(0)
 
-    case Nil => {}
+    case Nil => // No-op
 
     case _ =>
       printUsageAndExit(1)
diff --git a/core/src/main/scala/org/apache/spark/deploy/master/ZooKeeperPersistenceEngine.scala b/core/src/main/scala/org/apache/spark/deploy/master/ZooKeeperPersistenceEngine.scala
index 79f77212fe..af850e4871 100644
--- a/core/src/main/scala/org/apache/spark/deploy/master/ZooKeeperPersistenceEngine.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/master/ZooKeeperPersistenceEngine.scala
@@ -70,11 +70,10 @@ private[master] class ZooKeeperPersistenceEngine(conf: SparkConf, val serializer
     try {
       Some(serializer.newInstance().deserialize[T](ByteBuffer.wrap(fileData)))
     } catch {
-      case e: Exception => {
+      case e: Exception =>
         logWarning("Exception while reading persisted file, deleting", e)
         zk.delete().forPath(WORKING_DIR + "/" + filename)
         None
-      }
     }
   }
 }
diff --git a/core/src/main/scala/org/apache/spark/deploy/mesos/MesosClusterDispatcherArguments.scala b/core/src/main/scala/org/apache/spark/deploy/mesos/MesosClusterDispatcherArguments.scala
index b97805a28b..11e13441ee 100644
--- a/core/src/main/scala/org/apache/spark/deploy/mesos/MesosClusterDispatcherArguments.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/mesos/MesosClusterDispatcherArguments.scala
@@ -76,14 +76,13 @@ private[mesos] class MesosClusterDispatcherArguments(args: Array[String], conf:
     case ("--help") :: tail =>
       printUsageAndExit(0)
 
-    case Nil => {
+    case Nil =>
       if (masterUrl == null) {
         // scalastyle:off println
         System.err.println("--master is required")
         // scalastyle:on println
         printUsageAndExit(1)
       }
-    }
 
     case _ =>
       printUsageAndExit(1)
diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala b/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
index f9c92c3bb9..06066248ea 100644
--- a/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
@@ -179,16 +179,14 @@ private[deploy] class ExecutorRunner(
       val message = "Command exited with code " + exitCode
       worker.send(ExecutorStateChanged(appId, execId, state, Some(message), Some(exitCode)))
     } catch {
-      case interrupted: InterruptedException => {
+      case interrupted: InterruptedException =>
         logInfo("Runner thread for executor " + fullId + " interrupted")
         state = ExecutorState.KILLED
         killProcess(None)
-      }
-      case e: Exception => {
+      case e: Exception =>
         logError("Error running executor", e)
         state = ExecutorState.FAILED
         killProcess(Some(e.toString))
-      }
     }
   }
 }
diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala b/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala
index 1b7637a39c..449beb0811 100755
--- a/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/worker/Worker.scala
@@ -480,7 +480,7 @@ private[deploy] class Worker(
           memoryUsed += memory_
           sendToMaster(ExecutorStateChanged(appId, execId, manager.state, None, None))
         } catch {
-          case e: Exception => {
+          case e: Exception =>
             logError(s"Failed to launch executor $appId/$execId for ${appDesc.name}.", e)
             if (executors.contains(appId + "/" + execId)) {
               executors(appId + "/" + execId).kill()
@@ -488,7 +488,6 @@ private[deploy] class Worker(
             }
             sendToMaster(ExecutorStateChanged(appId, execId, ExecutorState.FAILED,
               Some(e.toString), None))
-          }
         }
       }
 
@@ -509,7 +508,7 @@ private[deploy] class Worker(
         }
       }
 
-    case LaunchDriver(driverId, driverDesc) => {
+    case LaunchDriver(driverId, driverDesc) =>
       logInfo(s"Asked to launch driver $driverId")
       val driver = new DriverRunner(
         conf,
@@ -525,9 +524,8 @@ private[deploy] class Worker(
 
       coresUsed += driverDesc.cores
       memoryUsed += driverDesc.mem
-    }
 
-    case KillDriver(driverId) => {
+    case KillDriver(driverId) =>
       logInfo(s"Asked to kill driver $driverId")
       drivers.get(driverId) match {
         case Some(runner) =>
@@ -535,11 +533,9 @@ private[deploy] class Worker(
         case None =>
           logError(s"Asked to kill unknown driver $driverId")
       }
-    }
 
-    case driverStateChanged @ DriverStateChanged(driverId, state, exception) => {
+    case driverStateChanged @ DriverStateChanged(driverId, state, exception) =>
       handleDriverStateChanged(driverStateChanged)
-    }
 
     case ReregisterWithMaster =>
       reregisterWithMaster()
diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/WorkerArguments.scala b/core/src/main/scala/org/apache/spark/deploy/worker/WorkerArguments.scala
index 391eb41190..777020d4d5 100644
--- a/core/src/main/scala/org/apache/spark/deploy/worker/WorkerArguments.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/worker/WorkerArguments.scala
@@ -165,12 +165,11 @@ private[worker] class WorkerArguments(args: Array[String], conf: SparkConf) {
       }
       // scalastyle:on classforname
     } catch {
-      case e: Exception => {
+      case e: Exception =>
         totalMb = 2*1024
         // scalastyle:off println
         System.out.println("Failed to get total physical memory. Using " + totalMb + " MB")
         // scalastyle:on println
-      }
     }
     // Leave out 1 GB for the operating system, but don't return a negative memory size
     math.max(totalMb - 1024, Utils.DEFAULT_DRIVER_MEM_MB)
diff --git a/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala b/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala
index d4ed5845e7..71b4ad160d 100644
--- a/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala
+++ b/core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala
@@ -62,10 +62,9 @@ private[spark] class CoarseGrainedExecutorBackend(
       // This is a very fast action so we can use "ThreadUtils.sameThread"
       case Success(msg) =>
         // Always receive `true`. Just ignore it
-      case Failure(e) => {
+      case Failure(e) =>
         logError(s"Cannot register with driver: $driverUrl", e)
         System.exit(1)
-      }
     }(ThreadUtils.sameThread)
   }
 
diff --git a/core/src/main/scala/org/apache/spark/metrics/MetricsSystem.scala b/core/src/main/scala/org/apache/spark/metrics/MetricsSystem.scala
index 4da1017d28..0fed991049 100644
--- a/core/src/main/scala/org/apache/spark/metrics/MetricsSystem.scala
+++ b/core/src/main/scala/org/apache/spark/metrics/MetricsSystem.scala
@@ -196,10 +196,9 @@ private[spark] class MetricsSystem private (
             sinks += sink.asInstanceOf[Sink]
           }
         } catch {
-          case e: Exception => {
+          case e: Exception =>
             logError("Sink class " + classPath + " cannot be instantiated")
             throw e
-          }
         }
       }
     }
diff --git a/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala b/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala
index c562c70aba..ab6aba6fc7 100644
--- a/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala
+++ b/core/src/main/scala/org/apache/spark/partial/BoundedDouble.scala
@@ -32,12 +32,11 @@ class BoundedDouble(val mean: Double, val confidence: Double, val low: Double, v
    */
   override def equals(that: Any): Boolean =
     that match {
-      case that: BoundedDouble => {
+      case that: BoundedDouble =>
         this.mean == that.mean &&
         this.confidence == that.confidence &&
         this.low == that.low &&
         this.high == that.high
-      }
       case _ => false
     }
 }
diff --git a/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala
index 5e9230e733..368916a39e 100644
--- a/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala
@@ -166,8 +166,8 @@ class DoubleRDDFunctions(self: RDD[Double]) extends Logging with Serializable {
       val counters = new Array[Long](buckets.length - 1)
       while (iter.hasNext) {
         bucketFunction(iter.next()) match {
-          case Some(x: Int) => {counters(x) += 1}
-          case _ => {}
+          case Some(x: Int) => counters(x) += 1
+          case _ => // No-Op
         }
       }
       Iterator(counters)
diff --git a/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala
index 363004e587..a5992022d0 100644
--- a/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala
@@ -86,12 +86,11 @@ class OrderedRDDFunctions[K : Ordering : ClassTag,
     def inRange(k: K): Boolean = ordering.gteq(k, lower) && ordering.lteq(k, upper)
 
     val rddToFilter: RDD[P] = self.partitioner match {
-      case Some(rp: RangePartitioner[K, V]) => {
+      case Some(rp: RangePartitioner[K, V]) =>
         val partitionIndicies = (rp.getPartition(lower), rp.getPartition(upper)) match {
           case (l, u) => Math.min(l, u) to Math.max(l, u)
         }
         PartitionPruningRDD.create(self, partitionIndicies.contains)
-      }
       case _ =>
         self
     }
diff --git a/core/src/main/scala/org/apache/spark/rdd/ParallelCollectionRDD.scala b/core/src/main/scala/org/apache/spark/rdd/ParallelCollectionRDD.scala
index 582fa93afe..462fb39ea2 100644
--- a/core/src/main/scala/org/apache/spark/rdd/ParallelCollectionRDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/ParallelCollectionRDD.scala
@@ -128,7 +128,7 @@ private object ParallelCollectionRDD {
       })
     }
     seq match {
-      case r: Range => {
+      case r: Range =>
         positions(r.length, numSlices).zipWithIndex.map({ case ((start, end), index) =>
           // If the range is inclusive, use inclusive range for the last slice
           if (r.isInclusive && index == numSlices - 1) {
@@ -138,8 +138,7 @@ private object ParallelCollectionRDD {
             new Range(r.start + start * r.step, r.start + end * r.step, r.step)
           }
         }).toSeq.asInstanceOf[Seq[Seq[T]]]
-      }
-      case nr: NumericRange[_] => {
+      case nr: NumericRange[_] =>
         // For ranges of Long, Double, BigInteger, etc
         val slices = new ArrayBuffer[Seq[T]](numSlices)
         var r = nr
@@ -149,14 +148,12 @@ private object ParallelCollectionRDD {
           r = r.drop(sliceSize)
         }
         slices
-      }
-      case _ => {
+      case _ =>
         val array = seq.toArray // To prevent O(n^2) operations for List etc
         positions(array.length, numSlices).map({
           case (start, end) =>
             array.slice(start, end).toSeq
         }).toSeq
-      }
     }
   }
 }
diff --git a/core/src/main/scala/org/apache/spark/rdd/PartitionerAwareUnionRDD.scala b/core/src/main/scala/org/apache/spark/rdd/PartitionerAwareUnionRDD.scala
index 9e3880714a..c3579d761d 100644
--- a/core/src/main/scala/org/apache/spark/rdd/PartitionerAwareUnionRDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/PartitionerAwareUnionRDD.scala
@@ -78,11 +78,10 @@ class PartitionerAwareUnionRDD[T: ClassTag](
     logDebug("Finding preferred location for " + this + ", partition " + s.index)
     val parentPartitions = s.asInstanceOf[PartitionerAwareUnionRDDPartition].parents
     val locations = rdds.zip(parentPartitions).flatMap {
-      case (rdd, part) => {
+      case (rdd, part) =>
         val parentLocations = currPrefLocs(rdd, part)
         logDebug("Location of " + rdd + " partition " + part.index + " = " + parentLocations)
         parentLocations
-      }
     }
     val location = if (locations.isEmpty) {
       None
diff --git a/core/src/main/scala/org/apache/spark/scheduler/InputFormatInfo.scala b/core/src/main/scala/org/apache/spark/scheduler/InputFormatInfo.scala
index 0640f26051..a6b032cc00 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/InputFormatInfo.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/InputFormatInfo.scala
@@ -57,11 +57,10 @@ class InputFormatInfo(val configuration: Configuration, val inputFormatClazz: Cl
   // Since we are not doing canonicalization of path, this can be wrong : like relative vs
   // absolute path .. which is fine, this is best case effort to remove duplicates - right ?
   override def equals(other: Any): Boolean = other match {
-    case that: InputFormatInfo => {
+    case that: InputFormatInfo =>
       // not checking config - that should be fine, right ?
       this.inputFormatClazz == that.inputFormatClazz &&
         this.path == that.path
-    }
     case _ => false
   }
 
@@ -86,10 +85,9 @@ class InputFormatInfo(val configuration: Configuration, val inputFormatClazz: Cl
       }
     }
     catch {
-      case e: ClassNotFoundException => {
+      case e: ClassNotFoundException =>
         throw new IllegalArgumentException("Specified inputformat " + inputFormatClazz +
           " cannot be found ?", e)
-      }
     }
   }
 
diff --git a/core/src/main/scala/org/apache/spark/scheduler/SplitInfo.scala b/core/src/main/scala/org/apache/spark/scheduler/SplitInfo.scala
index 6e9337bb90..bc1431835e 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/SplitInfo.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/SplitInfo.scala
@@ -49,14 +49,13 @@ class SplitInfo(
   // So unless there is identity equality between underlyingSplits, it will always fail even if it
   // is pointing to same block.
   override def equals(other: Any): Boolean = other match {
-    case that: SplitInfo => {
+    case that: SplitInfo =>
       this.hostLocation == that.hostLocation &&
         this.inputFormatClazz == that.inputFormatClazz &&
         this.path == that.path &&
         this.length == that.length &&
         // other split specific checks (like start for FileSplit)
         this.underlyingSplit == that.underlyingSplit
-    }
     case _ => false
   }
 }
diff --git a/core/src/main/scala/org/apache/spark/scheduler/TaskResultGetter.scala b/core/src/main/scala/org/apache/spark/scheduler/TaskResultGetter.scala
index 873f1b56bd..ae7ef46abb 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/TaskResultGetter.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/TaskResultGetter.scala
@@ -133,7 +133,7 @@ private[spark] class TaskResultGetter(sparkEnv: SparkEnv, scheduler: TaskSchedul
               // if we can't deserialize the reason.
               logError(
                 "Could not deserialize TaskEndReason: ClassNotFound with classloader " + loader)
-            case ex: Exception => {}
+            case ex: Exception => // No-op
           }
           scheduler.handleFailedTask(taskSetManager, tid, taskState, reason)
         }
diff --git a/core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala b/core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala
index 15d3515a02..6e08cdd87a 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala
@@ -188,20 +188,18 @@ private[spark] class TaskSetManager(
       loc match {
         case e: ExecutorCacheTaskLocation =>
           pendingTasksForExecutor.getOrElseUpdate(e.executorId, new ArrayBuffer) += index
-        case e: HDFSCacheTaskLocation => {
+        case e: HDFSCacheTaskLocation =>
           val exe = sched.getExecutorsAliveOnHost(loc.host)
           exe match {
-            case Some(set) => {
+            case Some(set) =>
               for (e <- set) {
                 pendingTasksForExecutor.getOrElseUpdate(e, new ArrayBuffer) += index
               }
               logInfo(s"Pending task $index has a cached location at ${e.host} " +
                 ", where there are executors " + set.mkString(","))
-            }
             case None => logDebug(s"Pending task $index has a cached location at ${e.host} " +
                 ", but there are no executors alive there.")
           }
-        }
         case _ =>
       }
       pendingTasksForHost.getOrElseUpdate(loc.host, new ArrayBuffer) += index
@@ -437,7 +435,7 @@ private[spark] class TaskSetManager(
       }
 
       dequeueTask(execId, host, allowedLocality) match {
-        case Some((index, taskLocality, speculative)) => {
+        case Some((index, taskLocality, speculative)) =>
           // Found a task; do some bookkeeping and return a task description
           val task = tasks(index)
           val taskId = sched.newTaskId()
@@ -486,7 +484,6 @@ private[spark] class TaskSetManager(
           sched.dagScheduler.taskStarted(task, info)
           return Some(new TaskDescription(taskId = taskId, attemptNumber = attemptNum, execId,
             taskName, index, serializedTask))
-        }
         case _ =>
       }
     }
diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterPersistenceEngine.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterPersistenceEngine.scala
index 3971e6c382..61ab3e87c5 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterPersistenceEngine.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosClusterPersistenceEngine.scala
@@ -121,11 +121,10 @@ private[spark] class ZookeeperMesosClusterPersistenceEngine(
       Some(Utils.deserialize[T](fileData))
     } catch {
       case e: NoNodeException => None
-      case e: Exception => {
+      case e: Exception =>
         logWarning("Exception while reading persisted file, deleting", e)
         zk.delete().forPath(zkPath)
         None
-      }
     }
   }
 
diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackendUtil.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackendUtil.scala
index 374c79a7e5..1b7ac172de 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackendUtil.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackendUtil.scala
@@ -55,11 +55,10 @@ private[mesos] object MesosSchedulerBackendUtil extends Logging {
             Some(vol.setContainerPath(container_path)
               .setHostPath(host_path)
               .setMode(Volume.Mode.RO))
-          case spec => {
+          case spec =>
             logWarning(s"Unable to parse volume specs: $volumes. "
               + "Expected form: \"[host-dir:]container-dir[:rw|:ro](, ...)\"")
             None
-          }
       }
     }
     .map { _.build() }
@@ -90,11 +89,10 @@ private[mesos] object MesosSchedulerBackendUtil extends Logging {
           Some(portmap.setHostPort(host_port.toInt)
             .setContainerPort(container_port.toInt)
             .setProtocol(protocol))
-        case spec => {
+        case spec =>
           logWarning(s"Unable to parse port mapping specs: $portmaps. "
             + "Expected form: \"host_port:container_port[:udp|:tcp](, ...)\"")
           None
-        }
       }
     }
     .map { _.build() }
diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala
index 233bdc23e6..7295d50682 100644
--- a/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala
+++ b/core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerUtils.scala
@@ -124,11 +124,10 @@ private[mesos] trait MesosSchedulerUtils extends Logging {
               markErr()
             }
           } catch {
-            case e: Exception => {
+            case e: Exception =>
               logError("driver.run() failed", e)
               error = Some(e)
               markErr()
-            }
           }
         }
       }.start()
@@ -184,7 +183,7 @@ private[mesos] trait MesosSchedulerUtils extends Logging {
     var remain = amountToUse
     var requestedResources = new ArrayBuffer[Resource]
     val remainingResources = resources.asScala.map {
-      case r => {
+      case r =>
         if (remain > 0 &&
           r.getType == Value.Type.SCALAR &&
           r.getScalar.getValue > 0.0 &&
@@ -196,7 +195,6 @@ private[mesos] trait MesosSchedulerUtils extends Logging {
         } else {
           r
         }
-      }
     }
 
     // Filter any resource that has depleted.
diff --git a/core/src/main/scala/org/apache/spark/serializer/Serializer.scala b/core/src/main/scala/org/apache/spark/serializer/Serializer.scala
index 5ead40e89e..cb95246d5b 100644
--- a/core/src/main/scala/org/apache/spark/serializer/Serializer.scala
+++ b/core/src/main/scala/org/apache/spark/serializer/Serializer.scala
@@ -188,10 +188,9 @@ abstract class DeserializationStream {
       try {
         (readKey[Any](), readValue[Any]())
       } catch {
-        case eof: EOFException => {
+        case eof: EOFException =>
           finished = true
           null
-        }
       }
     }
 
diff --git a/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala b/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
index 25edb9f1e4..4ec5b4bbb0 100644
--- a/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
+++ b/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
@@ -143,13 +143,12 @@ final class ShuffleBlockFetcherIterator(
     while (iter.hasNext) {
       val result = iter.next()
       result match {
-        case SuccessFetchResult(_, address, _, buf, _) => {
+        case SuccessFetchResult(_, address, _, buf, _) =>
           if (address != blockManager.blockManagerId) {
             shuffleMetrics.incRemoteBytesRead(buf.size)
             shuffleMetrics.incRemoteBlocksFetched(1)
           }
           buf.release()
-        }
         case _ =>
       }
     }
@@ -313,7 +312,7 @@ final class ShuffleBlockFetcherIterator(
     shuffleMetrics.incFetchWaitTime(stopFetchWait - startFetchWait)
 
     result match {
-      case SuccessFetchResult(_, address, size, buf, isNetworkReqDone) => {
+      case SuccessFetchResult(_, address, size, buf, isNetworkReqDone) =>
         if (address != blockManager.blockManagerId) {
           shuffleMetrics.incRemoteBytesRead(buf.size)
           shuffleMetrics.incRemoteBlocksFetched(1)
@@ -323,7 +322,6 @@ final class ShuffleBlockFetcherIterator(
           reqsInFlight -= 1
           logDebug("Number of requests in flight " + reqsInFlight)
         }
-      }
       case _ =>
     }
     // Send fetch requests up to maxBytesInFlight
diff --git a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala
index cc476d61b5..a0ef80d9bd 100644
--- a/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala
+++ b/core/src/main/scala/org/apache/spark/ui/exec/ExecutorThreadDumpPage.scala
@@ -38,7 +38,7 @@ private[ui] class ExecutorThreadDumpPage(parent: ExecutorsTab) extends WebUIPage
 
     val content = maybeThreadDump.map { threadDump =>
       val dumpRows = threadDump.sortWith {
-        case (threadTrace1, threadTrace2) => {
+        case (threadTrace1, threadTrace2) =>
           val v1 = if (threadTrace1.threadName.contains("Executor task launch")) 1 else 0
           val v2 = if (threadTrace2.threadName.contains("Executor task launch")) 1 else 0
           if (v1 == v2) {
@@ -46,7 +46,6 @@ private[ui] class ExecutorThreadDumpPage(parent: ExecutorsTab) extends WebUIPage
           } else {
             v1 > v2
           }
-        }
       }.map { thread =>
         val threadId = thread.threadId
         <tr id={s"thread_${threadId}_tr"} class="accordion-heading"
diff --git a/core/src/main/scala/org/apache/spark/util/EventLoop.scala b/core/src/main/scala/org/apache/spark/util/EventLoop.scala
index 153025cef2..3ea9139e11 100644
--- a/core/src/main/scala/org/apache/spark/util/EventLoop.scala
+++ b/core/src/main/scala/org/apache/spark/util/EventLoop.scala
@@ -47,13 +47,12 @@ private[spark] abstract class EventLoop[E](name: String) extends Logging {
           try {
             onReceive(event)
           } catch {
-            case NonFatal(e) => {
+            case NonFatal(e) =>
               try {
                 onError(e)
               } catch {
                 case NonFatal(e) => logError("Unexpected error in " + name, e)
               }
-            }
           }
         }
       } catch {
diff --git a/core/src/main/scala/org/apache/spark/util/SizeEstimator.scala b/core/src/main/scala/org/apache/spark/util/SizeEstimator.scala
index 3f627a0145..6861a75612 100644
--- a/core/src/main/scala/org/apache/spark/util/SizeEstimator.scala
+++ b/core/src/main/scala/org/apache/spark/util/SizeEstimator.scala
@@ -151,13 +151,12 @@ object SizeEstimator extends Logging {
       // TODO: We could use reflection on the VMOption returned ?
       getVMMethod.invoke(bean, "UseCompressedOops").toString.contains("true")
     } catch {
-      case e: Exception => {
+      case e: Exception =>
         // Guess whether they've enabled UseCompressedOops based on whether maxMemory < 32 GB
         val guess = Runtime.getRuntime.maxMemory < (32L*1024*1024*1024)
         val guessInWords = if (guess) "yes" else "not"
         logWarning("Failed to check whether UseCompressedOops is set; assuming " + guessInWords)
         return guess
-      }
     }
   }
 
diff --git a/core/src/test/scala/org/apache/spark/DistributedSuite.scala b/core/src/test/scala/org/apache/spark/DistributedSuite.scala
index 67d722c1dc..2110d3d770 100644
--- a/core/src/test/scala/org/apache/spark/DistributedSuite.scala
+++ b/core/src/test/scala/org/apache/spark/DistributedSuite.scala
@@ -320,7 +320,7 @@ class DistributedSuite extends SparkFunSuite with Matchers with LocalSparkContex
           Thread.sleep(200)
         }
       } catch {
-        case _: Throwable => { Thread.sleep(10) }
+        case _: Throwable => Thread.sleep(10)
           // Do nothing. We might see exceptions because block manager
           // is racing this thread to remove entries from the driver.
       }
diff --git a/core/src/test/scala/org/apache/spark/SparkContextInfoSuite.scala b/core/src/test/scala/org/apache/spark/SparkContextInfoSuite.scala
index 3706455c3f..8feb3dee05 100644
--- a/core/src/test/scala/org/apache/spark/SparkContextInfoSuite.scala
+++ b/core/src/test/scala/org/apache/spark/SparkContextInfoSuite.scala
@@ -82,20 +82,18 @@ package object testPackage extends Assertions {
     val curCallSite = sc.getCallSite().shortForm // note: 2 lines after definition of "rdd"
 
     val rddCreationLine = rddCreationSite match {
-      case CALL_SITE_REGEX(func, file, line) => {
+      case CALL_SITE_REGEX(func, file, line) =>
         assert(func === "makeRDD")
         assert(file === "SparkContextInfoSuite.scala")
         line.toInt
-      }
       case _ => fail("Did not match expected call site format")
     }
 
     curCallSite match {
-      case CALL_SITE_REGEX(func, file, line) => {
+      case CALL_SITE_REGEX(func, file, line) =>
         assert(func === "getCallSite") // this is correct because we called it from outside of Spark
         assert(file === "SparkContextInfoSuite.scala")
         assert(line.toInt === rddCreationLine.toInt + 2)
-      }
       case _ => fail("Did not match expected call site format")
     }
   }
diff --git a/core/src/test/scala/org/apache/spark/UnpersistSuite.scala b/core/src/test/scala/org/apache/spark/UnpersistSuite.scala
index f7a13ab399..09e21646ee 100644
--- a/core/src/test/scala/org/apache/spark/UnpersistSuite.scala
+++ b/core/src/test/scala/org/apache/spark/UnpersistSuite.scala
@@ -35,7 +35,7 @@ class UnpersistSuite extends SparkFunSuite with LocalSparkContext {
           Thread.sleep(200)
         }
       } catch {
-        case _: Throwable => { Thread.sleep(10) }
+        case _: Throwable => Thread.sleep(10)
           // Do nothing. We might see exceptions because block manager
           // is racing this thread to remove entries from the driver.
       }
diff --git a/core/src/test/scala/org/apache/spark/rpc/RpcEnvSuite.scala b/core/src/test/scala/org/apache/spark/rpc/RpcEnvSuite.scala
index 43e61241b6..cebac2097f 100644
--- a/core/src/test/scala/org/apache/spark/rpc/RpcEnvSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rpc/RpcEnvSuite.scala
@@ -127,9 +127,8 @@ abstract class RpcEnvSuite extends SparkFunSuite with BeforeAndAfterAll {
       override val rpcEnv = env
 
       override def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
-        case msg: String => {
+        case msg: String =>
           context.reply(msg)
-        }
       }
     })
     val reply = rpcEndpointRef.askWithRetry[String]("hello")
@@ -141,9 +140,8 @@ abstract class RpcEnvSuite extends SparkFunSuite with BeforeAndAfterAll {
       override val rpcEnv = env
 
       override def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
-        case msg: String => {
+        case msg: String =>
           context.reply(msg)
-        }
       }
     })
 
@@ -164,10 +162,9 @@ abstract class RpcEnvSuite extends SparkFunSuite with BeforeAndAfterAll {
       override val rpcEnv = env
 
       override def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
-        case msg: String => {
+        case msg: String =>
           Thread.sleep(100)
           context.reply(msg)
-        }
       }
     })
 
@@ -317,10 +314,9 @@ abstract class RpcEnvSuite extends SparkFunSuite with BeforeAndAfterAll {
       override val rpcEnv = env
 
       override def receive: PartialFunction[Any, Unit] = {
-        case m => {
+        case m =>
           self
           callSelfSuccessfully = true
-        }
       }
     })
 
@@ -682,9 +678,8 @@ abstract class RpcEnvSuite extends SparkFunSuite with BeforeAndAfterAll {
         override val rpcEnv = localEnv
 
         override def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
-          case msg: String => {
+          case msg: String =>
             context.reply(msg)
-          }
         }
       })
       val rpcEndpointRef = remoteEnv.setupEndpointRef(localEnv.address, "ask-authentication")
diff --git a/examples/src/main/scala/org/apache/spark/examples/CassandraCQLTest.scala b/examples/src/main/scala/org/apache/spark/examples/CassandraCQLTest.scala
index 973b005f91..ca4eea2356 100644
--- a/examples/src/main/scala/org/apache/spark/examples/CassandraCQLTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/CassandraCQLTest.scala
@@ -106,9 +106,8 @@ object CassandraCQLTest {
 
     println("Count: " + casRdd.count)
     val productSaleRDD = casRdd.map {
-      case (key, value) => {
+      case (key, value) =>
         (ByteBufferUtil.string(value.get("prod_id")), ByteBufferUtil.toInt(value.get("quantity")))
-      }
     }
     val aggregatedRDD = productSaleRDD.reduceByKey(_ + _)
     aggregatedRDD.collect().foreach {
@@ -116,11 +115,10 @@ object CassandraCQLTest {
     }
 
     val casoutputCF = aggregatedRDD.map {
-      case (productId, saleCount) => {
+      case (productId, saleCount) =>
         val outKey = Collections.singletonMap("prod_id", ByteBufferUtil.bytes(productId))
         val outVal = Collections.singletonList(ByteBufferUtil.bytes(saleCount))
         (outKey, outVal)
-      }
     }
 
     casoutputCF.saveAsNewAPIHadoopFile(
diff --git a/examples/src/main/scala/org/apache/spark/examples/CassandraTest.scala b/examples/src/main/scala/org/apache/spark/examples/CassandraTest.scala
index 6a8f73ad00..eff840d36e 100644
--- a/examples/src/main/scala/org/apache/spark/examples/CassandraTest.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/CassandraTest.scala
@@ -90,9 +90,8 @@ object CassandraTest {
 
     // Let us first get all the paragraphs from the retrieved rows
     val paraRdd = casRdd.map {
-      case (key, value) => {
+      case (key, value) =>
         ByteBufferUtil.string(value.get(ByteBufferUtil.bytes("para")).value())
-      }
     }
 
     // Lets get the word count in paras
@@ -103,7 +102,7 @@ object CassandraTest {
     }
 
     counts.map {
-      case (word, count) => {
+      case (word, count) =>
         val colWord = new org.apache.cassandra.thrift.Column()
         colWord.setName(ByteBufferUtil.bytes("word"))
         colWord.setValue(ByteBufferUtil.bytes(word))
@@ -122,7 +121,6 @@ object CassandraTest {
         mutations.get(1).setColumn_or_supercolumn(new ColumnOrSuperColumn())
         mutations.get(1).column_or_supercolumn.setColumn(colCount)
         (outputkey, mutations)
-      }
     }.saveAsNewAPIHadoopFile("casDemo", classOf[ByteBuffer], classOf[List[Mutation]],
       classOf[ColumnFamilyOutputFormat], job.getConfiguration)
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/LocalALS.scala b/examples/src/main/scala/org/apache/spark/examples/LocalALS.scala
index af5f216f28..fa10101955 100644
--- a/examples/src/main/scala/org/apache/spark/examples/LocalALS.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/LocalALS.scala
@@ -104,16 +104,14 @@ object LocalALS {
   def main(args: Array[String]) {
 
     args match {
-      case Array(m, u, f, iters) => {
+      case Array(m, u, f, iters) =>
         M = m.toInt
         U = u.toInt
         F = f.toInt
         ITERATIONS = iters.toInt
-      }
-      case _ => {
+      case _ =>
         System.err.println("Usage: LocalALS <M> <U> <F> <iters>")
         System.exit(1)
-      }
     }
 
     showWarning()
diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/OneVsRestExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/OneVsRestExample.scala
index a0bb5dabf4..0b5d31c0ff 100644
--- a/examples/src/main/scala/org/apache/spark/examples/ml/OneVsRestExample.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/ml/OneVsRestExample.scala
@@ -118,17 +118,15 @@ object OneVsRestExample {
     val inputData = sqlContext.read.format("libsvm").load(params.input)
     // compute the train/test split: if testInput is not provided use part of input.
     val data = params.testInput match {
-      case Some(t) => {
+      case Some(t) =>
         // compute the number of features in the training set.
         val numFeatures = inputData.first().getAs[Vector](1).size
         val testData = sqlContext.read.option("numFeatures", numFeatures.toString)
           .format("libsvm").load(t)
         Array[DataFrame](inputData, testData)
-      }
-      case None => {
+      case None =>
         val f = params.fracTest
         inputData.randomSplit(Array(1 - f, f), seed = 12345)
-      }
     }
     val Array(train, test) = data.map(_.cache())
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala b/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala
index c263f4f595..ee811d3aa1 100644
--- a/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala
@@ -180,7 +180,7 @@ object DecisionTreeRunner {
     }
     // For classification, re-index classes if needed.
     val (examples, classIndexMap, numClasses) = algo match {
-      case Classification => {
+      case Classification =>
         // classCounts: class --> # examples in class
         val classCounts = origExamples.map(_.label).countByValue()
         val sortedClasses = classCounts.keys.toList.sorted
@@ -209,7 +209,6 @@ object DecisionTreeRunner {
           println(s"$c\t$frac\t${classCounts(c)}")
         }
         (examples, classIndexMap, numClasses)
-      }
       case Regression =>
         (origExamples, null, 0)
       case _ =>
@@ -225,7 +224,7 @@ object DecisionTreeRunner {
         case "libsvm" => MLUtils.loadLibSVMFile(sc, testInput, numFeatures)
       }
       algo match {
-        case Classification => {
+        case Classification =>
           // classCounts: class --> # examples in class
           val testExamples = {
             if (classIndexMap.isEmpty) {
@@ -235,7 +234,6 @@ object DecisionTreeRunner {
             }
           }
           Array(examples, testExamples)
-        }
         case Regression =>
           Array(examples, origTestExamples)
       }
diff --git a/external/kinesis-asl/src/main/scala/org/apache/spark/streaming/kinesis/KinesisRecordProcessor.scala b/external/kinesis-asl/src/main/scala/org/apache/spark/streaming/kinesis/KinesisRecordProcessor.scala
index 41c6ab123b..80e0cce055 100644
--- a/external/kinesis-asl/src/main/scala/org/apache/spark/streaming/kinesis/KinesisRecordProcessor.scala
+++ b/external/kinesis-asl/src/main/scala/org/apache/spark/streaming/kinesis/KinesisRecordProcessor.scala
@@ -73,7 +73,7 @@ private[kinesis] class KinesisRecordProcessor[T](receiver: KinesisReceiver[T], w
         logDebug(s"Stored: Worker $workerId stored ${batch.size} records for shardId $shardId")
         receiver.setCheckpointer(shardId, checkpointer)
       } catch {
-        case NonFatal(e) => {
+        case NonFatal(e) =>
           /*
            *  If there is a failure within the batch, the batch will not be checkpointed.
            *  This will potentially cause records since the last checkpoint to be processed
@@ -84,7 +84,6 @@ private[kinesis] class KinesisRecordProcessor[T](receiver: KinesisReceiver[T], w
 
           /* Rethrow the exception to the Kinesis Worker that is managing this RecordProcessor. */
           throw e
-        }
       }
     } else {
       /* RecordProcessor has been stopped. */
@@ -148,29 +147,25 @@ private[kinesis] object KinesisRecordProcessor extends Logging {
       /* If the function failed, either retry or throw the exception */
       case util.Failure(e) => e match {
         /* Retry:  Throttling or other Retryable exception has occurred */
-        case _: ThrottlingException | _: KinesisClientLibDependencyException if numRetriesLeft > 1
-          => {
-               val backOffMillis = Random.nextInt(maxBackOffMillis)
-               Thread.sleep(backOffMillis)
-               logError(s"Retryable Exception:  Random backOffMillis=${backOffMillis}", e)
-               retryRandom(expression, numRetriesLeft - 1, maxBackOffMillis)
-             }
+        case _: ThrottlingException | _: KinesisClientLibDependencyException
+            if numRetriesLeft > 1 =>
+          val backOffMillis = Random.nextInt(maxBackOffMillis)
+          Thread.sleep(backOffMillis)
+          logError(s"Retryable Exception:  Random backOffMillis=${backOffMillis}", e)
+          retryRandom(expression, numRetriesLeft - 1, maxBackOffMillis)
         /* Throw:  Shutdown has been requested by the Kinesis Client Library. */
-        case _: ShutdownException => {
+        case _: ShutdownException =>
           logError(s"ShutdownException:  Caught shutdown exception, skipping checkpoint.", e)
           throw e
-        }
         /* Throw:  Non-retryable exception has occurred with the Kinesis Client Library */
-        case _: InvalidStateException => {
+        case _: InvalidStateException =>
           logError(s"InvalidStateException:  Cannot save checkpoint to the DynamoDB table used" +
               s" by the Amazon Kinesis Client Library.  Table likely doesn't exist.", e)
           throw e
-        }
         /* Throw:  Unexpected exception has occurred */
-        case _ => {
+        case _ =>
           logError(s"Unexpected, non-retryable exception.", e)
           throw e
-        }
       }
     }
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala b/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
index 551e75dc0a..fa143715be 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
@@ -53,7 +53,7 @@ private[r] object SparkRWrappers {
 
   def getModelCoefficients(model: PipelineModel): Array[Double] = {
     model.stages.last match {
-      case m: LinearRegressionModel => {
+      case m: LinearRegressionModel =>
         val coefficientStandardErrorsR = Array(m.summary.coefficientStandardErrors.last) ++
           m.summary.coefficientStandardErrors.dropRight(1)
         val tValuesR = Array(m.summary.tValues.last) ++ m.summary.tValues.dropRight(1)
@@ -64,14 +64,12 @@ private[r] object SparkRWrappers {
         } else {
           m.coefficients.toArray ++ coefficientStandardErrorsR ++ tValuesR ++ pValuesR
         }
-      }
-      case m: LogisticRegressionModel => {
+      case m: LogisticRegressionModel =>
         if (m.getFitIntercept) {
           Array(m.intercept) ++ m.coefficients.toArray
         } else {
           m.coefficients.toArray
         }
-      }
     }
   }
 
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixture.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixture.scala
index 03eb903bb8..f04c87259c 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixture.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixture.scala
@@ -181,13 +181,12 @@ class GaussianMixture private (
     val (weights, gaussians) = initialModel match {
       case Some(gmm) => (gmm.weights, gmm.gaussians)
 
-      case None => {
+      case None =>
         val samples = breezeData.takeSample(withReplacement = true, k * nSamples, seed)
         (Array.fill(k)(1.0 / k), Array.tabulate(k) { i =>
           val slice = samples.view(i * nSamples, (i + 1) * nSamples)
           new MultivariateGaussian(vectorMean(slice), initCovariance(slice))
         })
-      }
     }
 
     var llh = Double.MinValue // current log-likelihood
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixtureModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixtureModel.scala
index 02417b1124..f87613cc72 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixtureModel.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/GaussianMixtureModel.scala
@@ -183,7 +183,7 @@ object GaussianMixtureModel extends Loader[GaussianMixtureModel] {
     val k = (metadata \ "k").extract[Int]
     val classNameV1_0 = SaveLoadV1_0.classNameV1_0
     (loadedClassName, version) match {
-      case (classNameV1_0, "1.0") => {
+      case (classNameV1_0, "1.0") =>
         val model = SaveLoadV1_0.load(sc, path)
         require(model.weights.length == k,
           s"GaussianMixtureModel requires weights of length $k " +
@@ -192,7 +192,6 @@ object GaussianMixtureModel extends Loader[GaussianMixtureModel] {
           s"GaussianMixtureModel requires gaussians of length $k" +
           s"got gaussians of length ${model.gaussians.length}")
         model
-      }
       case _ => throw new Exception(
         s"GaussianMixtureModel.load did not recognize model with (className, format version):" +
         s"($loadedClassName, $version).  Supported:\n" +
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
index 37a21cd879..8ff0b83e8b 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
@@ -253,16 +253,14 @@ class KMeans private (
     }
 
     val centers = initialModel match {
-      case Some(kMeansCenters) => {
+      case Some(kMeansCenters) =>
         Array(kMeansCenters.clusterCenters.map(s => new VectorWithNorm(s)))
-      }
-      case None => {
+      case None =>
         if (initializationMode == KMeans.RANDOM) {
           initRandom(data)
         } else {
           initKMeansParallel(data)
         }
-      }
     }
     val initTimeInSeconds = (System.nanoTime() - initStartTime) / 1e9
     logInfo(s"Initialization with $initializationMode took " + "%.3f".format(initTimeInSeconds) +
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala b/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
index 0ec8975fed..ef284531c9 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
@@ -97,7 +97,7 @@ private[stat] object KolmogorovSmirnovTest extends Logging {
     : KolmogorovSmirnovTestResult = {
     val distObj =
       distName match {
-        case "norm" => {
+        case "norm" =>
           if (params.nonEmpty) {
             // parameters are passed, then can only be 2
             require(params.length == 2, "Normal distribution requires mean and standard " +
@@ -109,7 +109,6 @@ private[stat] object KolmogorovSmirnovTest extends Logging {
               "initialized to standard normal (i.e. N(0, 1))")
             new NormalDistribution(0, 1)
           }
-        }
         case  _ => throw new UnsupportedOperationException(s"$distName not yet supported through" +
           s" convenience method. Current options are:['norm'].")
       }
diff --git a/repl/src/main/scala/org/apache/spark/repl/ExecutorClassLoader.scala b/repl/src/main/scala/org/apache/spark/repl/ExecutorClassLoader.scala
index 928aaa5629..4a15d52b57 100644
--- a/repl/src/main/scala/org/apache/spark/repl/ExecutorClassLoader.scala
+++ b/repl/src/main/scala/org/apache/spark/repl/ExecutorClassLoader.scala
@@ -70,26 +70,24 @@ class ExecutorClassLoader(
   }
 
   override def findClass(name: String): Class[_] = {
-    userClassPathFirst match {
-      case true => findClassLocally(name).getOrElse(parentLoader.loadClass(name))
-      case false => {
-        try {
-          parentLoader.loadClass(name)
-        } catch {
-          case e: ClassNotFoundException => {
-            val classOption = findClassLocally(name)
-            classOption match {
-              case None =>
-                // If this class has a cause, it will break the internal assumption of Janino
-                // (the compiler used for Spark SQL code-gen).
-                // See org.codehaus.janino.ClassLoaderIClassLoader's findIClass, you will see
-                // its behavior will be changed if there is a cause and the compilation
-                // of generated class will fail.
-                throw new ClassNotFoundException(name)
-              case Some(a) => a
-            }
+    if (userClassPathFirst) {
+      findClassLocally(name).getOrElse(parentLoader.loadClass(name))
+    } else {
+      try {
+        parentLoader.loadClass(name)
+      } catch {
+        case e: ClassNotFoundException =>
+          val classOption = findClassLocally(name)
+          classOption match {
+            case None =>
+              // If this class has a cause, it will break the internal assumption of Janino
+              // (the compiler used for Spark SQL code-gen).
+              // See org.codehaus.janino.ClassLoaderIClassLoader's findIClass, you will see
+              // its behavior will be changed if there is a cause and the compilation
+              // of generated class will fail.
+              throw new ClassNotFoundException(name)
+            case Some(a) => a
           }
-        }
       }
     }
   }
diff --git a/scalastyle-config.xml b/scalastyle-config.xml
index 472a8f4084..a14e3e583f 100644
--- a/scalastyle-config.xml
+++ b/scalastyle-config.xml
@@ -228,6 +228,11 @@ This file is divided into 3 sections:
     <customMessage>Use Javadoc style indentation for multiline comments</customMessage>
   </check>
 
+  <check customId="OmitBracesInCase" level="error" class="org.scalastyle.file.RegexChecker" enabled="true">
+    <parameters><parameter name="regex">case[^\n>]*=>\s*\{</parameter></parameters>
+    <customMessage>Omit braces in case clauses.</customMessage>
+  </check>
+
   <!-- ================================================================================ -->
   <!--       rules we'd like to enforce, but haven't cleaned up the codebase yet        -->
   <!-- ================================================================================ -->
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala
index d842ffdc66..0f8876a9e6 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala
@@ -898,7 +898,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
     val result = ctx.freshName("result")
     val tmpRow = ctx.freshName("tmpRow")
 
-    val fieldsEvalCode = fieldsCasts.zipWithIndex.map { case (cast, i) => {
+    val fieldsEvalCode = fieldsCasts.zipWithIndex.map { case (cast, i) =>
       val fromFieldPrim = ctx.freshName("ffp")
       val fromFieldNull = ctx.freshName("ffn")
       val toFieldPrim = ctx.freshName("tfp")
@@ -920,7 +920,6 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
           }
         }
        """
-      }
     }.mkString("\n")
 
     (c, evPrim, evNull) =>
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/EquivalentExpressions.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/EquivalentExpressions.scala
index affd1bdb32..8d8cc152ff 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/EquivalentExpressions.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/EquivalentExpressions.scala
@@ -97,11 +97,11 @@ class EquivalentExpressions {
   def debugString(all: Boolean = false): String = {
     val sb: mutable.StringBuilder = new StringBuilder()
     sb.append("Equivalent expressions:\n")
-    equivalenceMap.foreach { case (k, v) => {
+    equivalenceMap.foreach { case (k, v) =>
       if (all || v.length > 1) {
         sb.append("  " + v.mkString(", ")).append("\n")
       }
-    }}
+    }
     sb.toString()
   }
 }
diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/RandomDataGenerator.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/RandomDataGenerator.scala
index 8207d64798..711e870711 100644
--- a/sql/catalyst/src/test/scala/org/apache/spark/sql/RandomDataGenerator.scala
+++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/RandomDataGenerator.scala
@@ -196,12 +196,11 @@ object RandomDataGenerator {
       case ShortType => randomNumeric[Short](
         rand, _.nextInt().toShort, Seq(Short.MinValue, Short.MaxValue, 0.toShort))
       case NullType => Some(() => null)
-      case ArrayType(elementType, containsNull) => {
+      case ArrayType(elementType, containsNull) =>
         forType(elementType, nullable = containsNull, rand).map {
           elementGenerator => () => Seq.fill(rand.nextInt(MAX_ARR_SIZE))(elementGenerator())
         }
-      }
-      case MapType(keyType, valueType, valueContainsNull) => {
+      case MapType(keyType, valueType, valueContainsNull) =>
         for (
           keyGenerator <- forType(keyType, nullable = false, rand);
           valueGenerator <-
@@ -221,8 +220,7 @@ object RandomDataGenerator {
             keys.zip(values).toMap
           }
         }
-      }
-      case StructType(fields) => {
+      case StructType(fields) =>
         val maybeFieldGenerators: Seq[Option[() => Any]] = fields.map { field =>
           forType(field.dataType, nullable = field.nullable, rand)
         }
@@ -232,8 +230,7 @@ object RandomDataGenerator {
         } else {
           None
         }
-      }
-      case udt: UserDefinedType[_] => {
+      case udt: UserDefinedType[_] =>
         val maybeSqlTypeGenerator = forType(udt.sqlType, nullable, rand)
         // Because random data generator at here returns scala value, we need to
         // convert it to catalyst value to call udt's deserialize.
@@ -253,7 +250,6 @@ object RandomDataGenerator {
         } else {
           None
         }
-      }
       case unsupportedType => None
     }
     // Handle nullability by wrapping the non-null value generator:
@@ -277,7 +273,7 @@ object RandomDataGenerator {
     val fields = mutable.ArrayBuffer.empty[Any]
     schema.fields.foreach { f =>
       f.dataType match {
-        case ArrayType(childType, nullable) => {
+        case ArrayType(childType, nullable) =>
           val data = if (f.nullable && rand.nextFloat() <= PROBABILITY_OF_NULL) {
             null
           } else {
@@ -294,10 +290,8 @@ object RandomDataGenerator {
             arr
           }
           fields += data
-        }
-        case StructType(children) => {
+        case StructType(children) =>
           fields += randomRow(rand, StructType(children))
-        }
         case _ =>
           val generator = RandomDataGenerator.forType(f.dataType, f.nullable, rand)
           assert(generator.isDefined, "Unsupported type")
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala
index 86c6405522..e953a6e8ef 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala
@@ -1153,14 +1153,12 @@ class DataFrameSuite extends QueryTest with SharedSQLContext {
   private def verifyNonExchangingAgg(df: DataFrame) = {
     var atFirstAgg: Boolean = false
     df.queryExecution.executedPlan.foreach {
-      case agg: TungstenAggregate => {
+      case agg: TungstenAggregate =>
         atFirstAgg = !atFirstAgg
-      }
-      case _ => {
+      case _ =>
         if (atFirstAgg) {
           fail("Should not have operators between the two aggregations")
         }
-      }
     }
   }
 
@@ -1170,12 +1168,11 @@ class DataFrameSuite extends QueryTest with SharedSQLContext {
   private def verifyExchangingAgg(df: DataFrame) = {
     var atFirstAgg: Boolean = false
     df.queryExecution.executedPlan.foreach {
-      case agg: TungstenAggregate => {
+      case agg: TungstenAggregate =>
         if (atFirstAgg) {
           fail("Should not have back to back Aggregates")
         }
         atFirstAgg = true
-      }
       case e: ShuffleExchange => atFirstAgg = false
       case _ =>
     }
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/vectorized/ColumnarBatchSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/vectorized/ColumnarBatchSuite.scala
index 8a551cd78c..31b63f2ce1 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/execution/vectorized/ColumnarBatchSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/vectorized/ColumnarBatchSuite.scala
@@ -612,23 +612,20 @@ class ColumnarBatchSuite extends SparkFunSuite {
             val a2 = r2.getList(v._2).toArray
             assert(a1.length == a2.length, "Seed = " + seed)
             childType match {
-              case DoubleType => {
+              case DoubleType =>
                 var i = 0
                 while (i < a1.length) {
                   assert(doubleEquals(a1(i).asInstanceOf[Double], a2(i).asInstanceOf[Double]),
                     "Seed = " + seed)
                   i += 1
                 }
-              }
-              case FloatType => {
+              case FloatType =>
                 var i = 0
                 while (i < a1.length) {
                   assert(doubleEquals(a1(i).asInstanceOf[Float], a2(i).asInstanceOf[Float]),
                     "Seed = " + seed)
                   i += 1
                 }
-              }
-
               case t: DecimalType =>
                 var i = 0
                 while (i < a1.length) {
@@ -640,7 +637,6 @@ class ColumnarBatchSuite extends SparkFunSuite {
                   }
                   i += 1
                 }
-
               case _ => assert(a1 === a2, "Seed = " + seed)
             }
           case StructType(childFields) =>
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala
index c40beeff97..58842f9c2f 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala
@@ -429,13 +429,12 @@ abstract class DStream[T: ClassTag] (
    */
   private[streaming] def generateJob(time: Time): Option[Job] = {
     getOrCompute(time) match {
-      case Some(rdd) => {
+      case Some(rdd) =>
         val jobFunc = () => {
           val emptyFunc = { (iterator: Iterator[T]) => {} }
           context.sparkContext.runJob(rdd, emptyFunc)
         }
         Some(new Job(time, jobFunc))
-      }
       case None => None
     }
   }
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala
index 431c9dbe2c..e73837eb96 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/DStreamCheckpointData.scala
@@ -109,10 +109,9 @@ class DStreamCheckpointData[T: ClassTag](dstream: DStream[T])
   def restore() {
     // Create RDDs from the checkpoint data
     currentCheckpointFiles.foreach {
-      case(time, file) => {
+      case(time, file) =>
         logInfo("Restoring checkpointed RDD for time " + time + " from file '" + file + "'")
         dstream.generatedRDDs += ((time, dstream.context.sparkContext.checkpointFile[T](file)))
-      }
     }
   }
 
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/FileInputDStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/FileInputDStream.scala
index 7fba2e8ec0..36f50e04db 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/FileInputDStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/FileInputDStream.scala
@@ -333,14 +333,13 @@ class FileInputDStream[K, V, F <: NewInputFormat[K, V]](
 
     override def restore() {
       hadoopFiles.toSeq.sortBy(_._1)(Time.ordering).foreach {
-        case (t, f) => {
+        case (t, f) =>
           // Restore the metadata in both files and generatedRDDs
           logInfo("Restoring files for time " + t + " - " +
             f.mkString("[", ", ", "]") )
           batchTimeToSelectedFiles.synchronized { batchTimeToSelectedFiles += ((t, f)) }
           recentlySelectedFiles ++= f
           generatedRDDs += ((t, filesToRDD(f)))
-        }
       }
     }
 
diff --git a/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala b/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala
index 0379957e58..28aed0ca45 100644
--- a/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala
+++ b/streaming/src/main/scala/org/apache/spark/streaming/dstream/StateDStream.scala
@@ -65,14 +65,12 @@ class StateDStream[K: ClassTag, V: ClassTag, S: ClassTag](
     // Try to get the previous state RDD
     getOrCompute(validTime - slideDuration) match {
 
-      case Some(prevStateRDD) => {    // If previous state RDD exists
-
+      case Some(prevStateRDD) =>    // If previous state RDD exists
         // Try to get the parent RDD
         parent.getOrCompute(validTime) match {
-          case Some(parentRDD) => {   // If parent RDD exists, then compute as usual
+          case Some(parentRDD) =>   // If parent RDD exists, then compute as usual
             computeUsingPreviousRDD(parentRDD, prevStateRDD)
-          }
-          case None => {    // If parent RDD does not exist
+          case None =>    // If parent RDD does not exist
 
             // Re-apply the update function to the old state RDD
             val updateFuncLocal = updateFunc
@@ -82,17 +80,14 @@ class StateDStream[K: ClassTag, V: ClassTag, S: ClassTag](
             }
             val stateRDD = prevStateRDD.mapPartitions(finalFunc, preservePartitioning)
             Some(stateRDD)
-          }
         }
-      }
-
-      case None => {    // If previous session RDD does not exist (first input data)
 
+      case None =>    // If previous session RDD does not exist (first input data)
         // Try to get the parent RDD
         parent.getOrCompute(validTime) match {
-          case Some(parentRDD) => {   // If parent RDD exists, then compute as usual
+          case Some(parentRDD) =>   // If parent RDD exists, then compute as usual
             initialRDD match {
-              case None => {
+              case None =>
                 // Define the function for the mapPartition operation on grouped RDD;
                 // first map the grouped tuple to tuples of required type,
                 // and then apply the update function
@@ -105,18 +100,13 @@ class StateDStream[K: ClassTag, V: ClassTag, S: ClassTag](
                 val sessionRDD = groupedRDD.mapPartitions(finalFunc, preservePartitioning)
                 // logDebug("Generating state RDD for time " + validTime + " (first)")
                 Some(sessionRDD)
-              }
-              case Some(initialStateRDD) => {
+              case Some(initialStateRDD) =>
                 computeUsingPreviousRDD(parentRDD, initialStateRDD)
-              }
             }
-          }
-          case None => { // If parent RDD does not exist, then nothing to do!
+          case None => // If parent RDD does not exist, then nothing to do!
             // logDebug("Not generating state RDD (no previous state, no parent)")
             None
-          }
         }
-      }
     }
   }
 }
diff --git a/streaming/src/test/scala/org/apache/spark/streaming/BasicOperationsSuite.scala b/streaming/src/test/scala/org/apache/spark/streaming/BasicOperationsSuite.scala
index bd60059b18..cfcbdc7c38 100644
--- a/streaming/src/test/scala/org/apache/spark/streaming/BasicOperationsSuite.scala
+++ b/streaming/src/test/scala/org/apache/spark/streaming/BasicOperationsSuite.scala
@@ -538,10 +538,9 @@ class BasicOperationsSuite extends TestSuiteBase {
         val stateObj = state.getOrElse(new StateObject)
         values.sum match {
           case 0 => stateObj.expireCounter += 1 // no new values
-          case n => { // has new values, increment and reset expireCounter
+          case n => // has new values, increment and reset expireCounter
             stateObj.counter += n
             stateObj.expireCounter = 0
-          }
         }
         stateObj.expireCounter match {
           case 2 => None // seen twice with no new values, give it the boot
diff --git a/streaming/src/test/scala/org/apache/spark/streaming/CheckpointSuite.scala b/streaming/src/test/scala/org/apache/spark/streaming/CheckpointSuite.scala
index fbb25d4c59..bdbac64b9b 100644
--- a/streaming/src/test/scala/org/apache/spark/streaming/CheckpointSuite.scala
+++ b/streaming/src/test/scala/org/apache/spark/streaming/CheckpointSuite.scala
@@ -267,10 +267,9 @@ class CheckpointSuite extends TestSuiteBase with DStreamCheckpointTester
     assert(!stateStream.checkpointData.currentCheckpointFiles.isEmpty,
       "No checkpointed RDDs in state stream before first failure")
     stateStream.checkpointData.currentCheckpointFiles.foreach {
-      case (time, file) => {
+      case (time, file) =>
         assert(fs.exists(new Path(file)), "Checkpoint file '" + file +"' for time " + time +
             " for state stream before first failure does not exist")
-      }
     }
 
     // Run till a further time such that previous checkpoint files in the stream would be deleted
@@ -297,10 +296,9 @@ class CheckpointSuite extends TestSuiteBase with DStreamCheckpointTester
     assert(!stateStream.checkpointData.currentCheckpointFiles.isEmpty,
       "No checkpointed RDDs in state stream before second failure")
     stateStream.checkpointData.currentCheckpointFiles.foreach {
-      case (time, file) => {
+      case (time, file) =>
         assert(fs.exists(new Path(file)), "Checkpoint file '" + file +"' for time " + time +
           " for state stream before seconds failure does not exist")
-      }
     }
     ssc.stop()
 
diff --git a/streaming/src/test/scala/org/apache/spark/streaming/MasterFailureTest.scala b/streaming/src/test/scala/org/apache/spark/streaming/MasterFailureTest.scala
index 29bee4adf2..60c8e70235 100644
--- a/streaming/src/test/scala/org/apache/spark/streaming/MasterFailureTest.scala
+++ b/streaming/src/test/scala/org/apache/spark/streaming/MasterFailureTest.scala
@@ -382,11 +382,10 @@ class FileGeneratingThread(input: Seq[String], testDir: Path, interval: Long)
                 fs.rename(tempHadoopFile, hadoopFile)
             done = true
           } catch {
-            case ioe: IOException => {
+            case ioe: IOException =>
                   fs = testDir.getFileSystem(new Configuration())
                   logWarning("Attempt " + tries + " at generating file " + hadoopFile + " failed.",
                     ioe)
-            }
           }
         }
         if (!done) {
diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala
index 9e8453429c..d447a59937 100644
--- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala
+++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala
@@ -374,7 +374,7 @@ private[spark] class ApplicationMaster(
             failureCount = 0
           } catch {
             case i: InterruptedException =>
-            case e: Throwable => {
+            case e: Throwable =>
               failureCount += 1
               // this exception was introduced in hadoop 2.4 and this code would not compile
               // with earlier versions if we refer it directly.
@@ -390,7 +390,6 @@ private[spark] class ApplicationMaster(
               } else {
                 logWarning(s"Reporter thread fails $failureCount time(s) in a row.", e)
               }
-            }
           }
           try {
             val numPendingAllocate = allocator.getPendingAllocate.size
diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala
index b0bfe855e9..23742eab62 100644
--- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala
+++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/YarnAllocator.scala
@@ -148,11 +148,10 @@ private[yarn] class YarnAllocator(
         classOf[Array[String]], classOf[Array[String]], classOf[Priority], classOf[Boolean],
         classOf[String]))
     } catch {
-      case e: NoSuchMethodException => {
+      case e: NoSuchMethodException =>
         logWarning(s"Node label expression $expr will be ignored because YARN version on" +
           " classpath does not support it.")
         None
-      }
     }
   }
 
diff --git a/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnSchedulerBackend.scala b/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnSchedulerBackend.scala
index 8720ee57fe..6b3c831e60 100644
--- a/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnSchedulerBackend.scala
+++ b/yarn/src/main/scala/org/apache/spark/scheduler/cluster/YarnSchedulerBackend.scala
@@ -223,17 +223,15 @@ private[spark] abstract class YarnSchedulerBackend(
           val lossReasonRequest = GetExecutorLossReason(executorId)
           val future = am.ask[ExecutorLossReason](lossReasonRequest, askTimeout)
           future onSuccess {
-            case reason: ExecutorLossReason => {
+            case reason: ExecutorLossReason =>
               driverEndpoint.askWithRetry[Boolean](RemoveExecutor(executorId, reason))
-            }
           }
           future onFailure {
-            case NonFatal(e) => {
+            case NonFatal(e) =>
               logWarning(s"Attempted to get executor loss reason" +
                 s" for executor id ${executorId} at RPC address ${executorRpcAddress}," +
                 s" but got no response. Marking as slave lost.", e)
               driverEndpoint.askWithRetry[Boolean](RemoveExecutor(executorId, SlaveLost()))
-            }
             case t => throw t
           }
         case None =>
diff --git a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtilSuite.scala b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtilSuite.scala
index de14e36f4e..fe09808ae5 100644
--- a/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtilSuite.scala
+++ b/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnSparkHadoopUtilSuite.scala
@@ -101,22 +101,18 @@ class YarnSparkHadoopUtilSuite extends SparkFunSuite with Matchers with Logging
     val modifyAcls = acls.get(ApplicationAccessType.MODIFY_APP)
 
     viewAcls match {
-      case Some(vacls) => {
+      case Some(vacls) =>
         val aclSet = vacls.split(',').map(_.trim).toSet
         assert(aclSet.contains(System.getProperty("user.name", "invalid")))
-      }
-      case None => {
+      case None =>
         fail()
-      }
     }
     modifyAcls match {
-      case Some(macls) => {
+      case Some(macls) =>
         val aclSet = macls.split(',').map(_.trim).toSet
         assert(aclSet.contains(System.getProperty("user.name", "invalid")))
-      }
-      case None => {
+      case None =>
         fail()
-      }
     }
   }
 
@@ -135,26 +131,22 @@ class YarnSparkHadoopUtilSuite extends SparkFunSuite with Matchers with Logging
     val modifyAcls = acls.get(ApplicationAccessType.MODIFY_APP)
 
     viewAcls match {
-      case Some(vacls) => {
+      case Some(vacls) =>
         val aclSet = vacls.split(',').map(_.trim).toSet
         assert(aclSet.contains("user1"))
         assert(aclSet.contains("user2"))
         assert(aclSet.contains(System.getProperty("user.name", "invalid")))
-      }
-      case None => {
+      case None =>
         fail()
-      }
     }
     modifyAcls match {
-      case Some(macls) => {
+      case Some(macls) =>
         val aclSet = macls.split(',').map(_.trim).toSet
         assert(aclSet.contains("user3"))
         assert(aclSet.contains("user4"))
         assert(aclSet.contains(System.getProperty("user.name", "invalid")))
-      }
-      case None => {
+      case None =>
         fail()
-      }
     }
 
   }
-- 
cgit v1.2.3


From da60b34d2f6eba19633e4f1b46504ce92cd6c179 Mon Sep 17 00:00:00 2001
From: Yong Tang <yong.tang.github@outlook.com>
Date: Tue, 12 Apr 2016 16:53:26 +0200
Subject: [SPARK-3724][ML] RandomForest: More options for feature subset size.

## What changes were proposed in this pull request?

This PR tries to support more options for feature subset size in RandomForest implementation. Previously, RandomForest only support "auto", "all", "sort", "log2", "onethird". This PR tries to support any given value to allow model search.

In this PR, `featureSubsetStrategy` could be passed with:
a) a real number in the range of `(0.0-1.0]` that represents the fraction of the number of features in each subset,
b)  an integer number (`>0`) that represents the number of features in each subset.

## How was this patch tested?

Two tests `JavaRandomForestClassifierSuite` and `JavaRandomForestRegressorSuite` have been updated to check the additional options for params in this PR.
An additional test has been added to `org.apache.spark.mllib.tree.RandomForestSuite` to cover the cases in this PR.

Author: Yong Tang <yong.tang.github@outlook.com>

Closes #11989 from yongtang/SPARK-3724.
---
 .../spark/ml/tree/impl/DecisionTreeMetadata.scala  |  5 +++
 .../org/apache/spark/ml/tree/treeParams.scala      |  8 ++++-
 .../org/apache/spark/mllib/tree/RandomForest.scala | 11 +++++--
 .../JavaRandomForestClassifierSuite.java           | 19 ++++++++++++
 .../regression/JavaRandomForestRegressorSuite.java | 19 ++++++++++++
 .../spark/ml/tree/impl/RandomForestSuite.scala     | 36 ++++++++++++++++++++++
 6 files changed, 95 insertions(+), 3 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala
index df8eb5d1f9..c7cde1563f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/impl/DecisionTreeMetadata.scala
@@ -183,11 +183,16 @@ private[spark] object DecisionTreeMetadata extends Logging {
         }
       case _ => featureSubsetStrategy
     }
+
+    val isIntRegex = "^([1-9]\\d*)$".r
+    val isFractionRegex = "^(0?\\.\\d*[1-9]\\d*|1\\.0+)$".r
     val numFeaturesPerNode: Int = _featureSubsetStrategy match {
       case "all" => numFeatures
       case "sqrt" => math.sqrt(numFeatures).ceil.toInt
       case "log2" => math.max(1, (math.log(numFeatures) / math.log(2)).ceil.toInt)
       case "onethird" => (numFeatures / 3.0).ceil.toInt
+      case isIntRegex(number) => if (BigInt(number) > numFeatures) numFeatures else number.toInt
+      case isFractionRegex(fraction) => (fraction.toDouble * numFeatures).ceil.toInt
     }
 
     new DecisionTreeMetadata(numFeatures, numExamples, numClasses, numBins.max,
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
index 78e6d3bfac..0767dc17e5 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
@@ -329,6 +329,8 @@ private[ml] trait HasFeatureSubsetStrategy extends Params {
    *  - "onethird": use 1/3 of the features
    *  - "sqrt": use sqrt(number of features)
    *  - "log2": use log2(number of features)
+   *  - "n": when n is in the range (0, 1.0], use n * number of features. When n
+   *         is in the range (1, number of features), use n features.
    * (default = "auto")
    *
    * These various settings are based on the following references:
@@ -346,7 +348,8 @@ private[ml] trait HasFeatureSubsetStrategy extends Params {
     "The number of features to consider for splits at each tree node." +
       s" Supported options: ${RandomForestParams.supportedFeatureSubsetStrategies.mkString(", ")}",
     (value: String) =>
-      RandomForestParams.supportedFeatureSubsetStrategies.contains(value.toLowerCase))
+      RandomForestParams.supportedFeatureSubsetStrategies.contains(value.toLowerCase)
+      || value.matches(RandomForestParams.supportedFeatureSubsetStrategiesRegex))
 
   setDefault(featureSubsetStrategy -> "auto")
 
@@ -393,6 +396,9 @@ private[spark] object RandomForestParams {
   // These options should be lowercase.
   final val supportedFeatureSubsetStrategies: Array[String] =
     Array("auto", "all", "onethird", "sqrt", "log2").map(_.toLowerCase)
+
+  // The regex to capture "(0.0-1.0]", and "n" for integer 0 < n <= (number of features)
+  final val supportedFeatureSubsetStrategiesRegex = "^(?:[1-9]\\d*|0?\\.\\d*[1-9]\\d*|1\\.0+)$"
 }
 
 private[ml] trait RandomForestClassifierParams
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala
index 1841fa4a95..26755849ad 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala
@@ -55,10 +55,15 @@ import org.apache.spark.util.Utils
  * @param numTrees If 1, then no bootstrapping is used.  If > 1, then bootstrapping is done.
  * @param featureSubsetStrategy Number of features to consider for splits at each node.
  *                              Supported values: "auto", "all", "sqrt", "log2", "onethird".
+ *                              Supported numerical values: "(0.0-1.0]", "[1-n]".
  *                              If "auto" is set, this parameter is set based on numTrees:
  *                                if numTrees == 1, set to "all";
  *                                if numTrees > 1 (forest) set to "sqrt" for classification and
  *                                  to "onethird" for regression.
+ *                              If a real value "n" in the range (0, 1.0] is set,
+ *                                use n * number of features.
+ *                              If an integer value "n" in the range (1, num features) is set,
+ *                                use n features.
  * @param seed Random seed for bootstrapping and choosing feature subsets.
  */
 private class RandomForest (
@@ -70,9 +75,11 @@ private class RandomForest (
 
   strategy.assertValid()
   require(numTrees > 0, s"RandomForest requires numTrees > 0, but was given numTrees = $numTrees.")
-  require(RandomForest.supportedFeatureSubsetStrategies.contains(featureSubsetStrategy),
+  require(RandomForest.supportedFeatureSubsetStrategies.contains(featureSubsetStrategy)
+    || featureSubsetStrategy.matches(NewRFParams.supportedFeatureSubsetStrategiesRegex),
     s"RandomForest given invalid featureSubsetStrategy: $featureSubsetStrategy." +
-    s" Supported values: ${RandomForest.supportedFeatureSubsetStrategies.mkString(", ")}.")
+    s" Supported values: ${RandomForest.supportedFeatureSubsetStrategies.mkString(", ")}," +
+    s" (0.0-1.0], [1-n].")
 
   /**
    * Method to train a decision tree model over an RDD
diff --git a/mllib/src/test/java/org/apache/spark/ml/classification/JavaRandomForestClassifierSuite.java b/mllib/src/test/java/org/apache/spark/ml/classification/JavaRandomForestClassifierSuite.java
index 75061464e5..5aec52ac72 100644
--- a/mllib/src/test/java/org/apache/spark/ml/classification/JavaRandomForestClassifierSuite.java
+++ b/mllib/src/test/java/org/apache/spark/ml/classification/JavaRandomForestClassifierSuite.java
@@ -22,6 +22,7 @@ import java.util.HashMap;
 import java.util.Map;
 
 import org.junit.After;
+import org.junit.Assert;
 import org.junit.Before;
 import org.junit.Test;
 
@@ -80,6 +81,24 @@ public class JavaRandomForestClassifierSuite implements Serializable {
     for (String featureSubsetStrategy: RandomForestClassifier.supportedFeatureSubsetStrategies()) {
       rf.setFeatureSubsetStrategy(featureSubsetStrategy);
     }
+    String realStrategies[] = {".1", ".10", "0.10", "0.1", "0.9", "1.0"};
+    for (String strategy: realStrategies) {
+      rf.setFeatureSubsetStrategy(strategy);
+    }
+    String integerStrategies[] = {"1", "10", "100", "1000", "10000"};
+    for (String strategy: integerStrategies) {
+      rf.setFeatureSubsetStrategy(strategy);
+    }
+    String invalidStrategies[] = {"-.1", "-.10", "-0.10", ".0", "0.0", "1.1", "0"};
+    for (String strategy: invalidStrategies) {
+      try {
+        rf.setFeatureSubsetStrategy(strategy);
+        Assert.fail("Expected exception to be thrown for invalid strategies");
+      } catch (Exception e) {
+        Assert.assertTrue(e instanceof IllegalArgumentException);
+      }
+    }
+
     RandomForestClassificationModel model = rf.fit(dataFrame);
 
     model.transform(dataFrame);
diff --git a/mllib/src/test/java/org/apache/spark/ml/regression/JavaRandomForestRegressorSuite.java b/mllib/src/test/java/org/apache/spark/ml/regression/JavaRandomForestRegressorSuite.java
index b6f793f6de..a8736669f7 100644
--- a/mllib/src/test/java/org/apache/spark/ml/regression/JavaRandomForestRegressorSuite.java
+++ b/mllib/src/test/java/org/apache/spark/ml/regression/JavaRandomForestRegressorSuite.java
@@ -22,6 +22,7 @@ import java.util.HashMap;
 import java.util.Map;
 
 import org.junit.After;
+import org.junit.Assert;
 import org.junit.Before;
 import org.junit.Test;
 
@@ -80,6 +81,24 @@ public class JavaRandomForestRegressorSuite implements Serializable {
     for (String featureSubsetStrategy: RandomForestRegressor.supportedFeatureSubsetStrategies()) {
       rf.setFeatureSubsetStrategy(featureSubsetStrategy);
     }
+    String realStrategies[] = {".1", ".10", "0.10", "0.1", "0.9", "1.0"};
+    for (String strategy: realStrategies) {
+      rf.setFeatureSubsetStrategy(strategy);
+    }
+    String integerStrategies[] = {"1", "10", "100", "1000", "10000"};
+    for (String strategy: integerStrategies) {
+      rf.setFeatureSubsetStrategy(strategy);
+    }
+    String invalidStrategies[] = {"-.1", "-.10", "-0.10", ".0", "0.0", "1.1", "0"};
+    for (String strategy: invalidStrategies) {
+      try {
+        rf.setFeatureSubsetStrategy(strategy);
+        Assert.fail("Expected exception to be thrown for invalid strategies");
+      } catch (Exception e) {
+        Assert.assertTrue(e instanceof IllegalArgumentException);
+      }
+    }
+
     RandomForestRegressionModel model = rf.fit(dataFrame);
 
     model.transform(dataFrame);
diff --git a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
index cd402b1e1f..6db9ce150d 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/tree/impl/RandomForestSuite.scala
@@ -426,12 +426,48 @@ class RandomForestSuite extends SparkFunSuite with MLlibTestSparkContext {
       (math.log(numFeatures) / math.log(2)).ceil.toInt)
     checkFeatureSubsetStrategy(numTrees = 1, "onethird", (numFeatures / 3.0).ceil.toInt)
 
+    val realStrategies = Array(".1", ".10", "0.10", "0.1", "0.9", "1.0")
+    for (strategy <- realStrategies) {
+      val expected = (strategy.toDouble * numFeatures).ceil.toInt
+      checkFeatureSubsetStrategy(numTrees = 1, strategy, expected)
+    }
+
+    val integerStrategies = Array("1", "10", "100", "1000", "10000")
+    for (strategy <- integerStrategies) {
+      val expected = if (strategy.toInt < numFeatures) strategy.toInt else numFeatures
+      checkFeatureSubsetStrategy(numTrees = 1, strategy, expected)
+    }
+
+    val invalidStrategies = Array("-.1", "-.10", "-0.10", ".0", "0.0", "1.1", "0")
+    for (invalidStrategy <- invalidStrategies) {
+      intercept[MatchError]{
+        val metadata =
+          DecisionTreeMetadata.buildMetadata(rdd, strategy, numTrees = 1, invalidStrategy)
+      }
+    }
+
     checkFeatureSubsetStrategy(numTrees = 2, "all", numFeatures)
     checkFeatureSubsetStrategy(numTrees = 2, "auto", math.sqrt(numFeatures).ceil.toInt)
     checkFeatureSubsetStrategy(numTrees = 2, "sqrt", math.sqrt(numFeatures).ceil.toInt)
     checkFeatureSubsetStrategy(numTrees = 2, "log2",
       (math.log(numFeatures) / math.log(2)).ceil.toInt)
     checkFeatureSubsetStrategy(numTrees = 2, "onethird", (numFeatures / 3.0).ceil.toInt)
+
+    for (strategy <- realStrategies) {
+      val expected = (strategy.toDouble * numFeatures).ceil.toInt
+      checkFeatureSubsetStrategy(numTrees = 2, strategy, expected)
+    }
+
+    for (strategy <- integerStrategies) {
+      val expected = if (strategy.toInt < numFeatures) strategy.toInt else numFeatures
+      checkFeatureSubsetStrategy(numTrees = 2, strategy, expected)
+    }
+    for (invalidStrategy <- invalidStrategies) {
+      intercept[MatchError]{
+        val metadata =
+          DecisionTreeMetadata.buildMetadata(rdd, strategy, numTrees = 2, invalidStrategy)
+      }
+    }
   }
 
   test("Binary classification with continuous features: subsampling features") {
-- 
cgit v1.2.3


From 75e05a5a964c9585dd09a2ef6178881929bab1f1 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Tue, 12 Apr 2016 10:51:07 -0700
Subject: [SPARK-12566][SPARK-14324][ML] GLM model family, link function
 support in SparkR:::glm

* SparkR glm supports families and link functions which match R's signature for family.
* SparkR glm API refactor. The comparative standard of the new API is R glm, so I only expose the arguments that R glm supports: ```formula, family, data, epsilon and maxit```.
* This PR is focus on glm() and predict(), summary statistics will be done in a separate PR after this get in.
* This PR depends on #12287 which make GLMs support link prediction at Scala side. After that merged, I will add more tests for predict() to this PR.

Unit tests.

cc mengxr jkbradley hhbyyh

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12294 from yanboliang/spark-12566.
---
 R/pkg/R/mllib.R                                    | 139 +++++++++------------
 R/pkg/inst/tests/testthat/test_mllib.R             |  95 +++++---------
 .../ml/r/GeneralizedLinearRegressionWrapper.scala  |  79 ++++++++++++
 .../org/apache/spark/ml/r/SparkRWrappers.scala     | 115 -----------------
 4 files changed, 169 insertions(+), 259 deletions(-)
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/r/GeneralizedLinearRegressionWrapper.scala
 delete mode 100644 mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/R/pkg/R/mllib.R b/R/pkg/R/mllib.R
index f3152cc232..31bca16580 100644
--- a/R/pkg/R/mllib.R
+++ b/R/pkg/R/mllib.R
@@ -17,10 +17,10 @@
 
 # mllib.R: Provides methods for MLlib integration
 
-#' @title S4 class that represents a PipelineModel
-#' @param model A Java object reference to the backing Scala PipelineModel
+#' @title S4 class that represents a generalized linear model
+#' @param jobj a Java object reference to the backing Scala GeneralizedLinearRegressionWrapper
 #' @export
-setClass("PipelineModel", representation(model = "jobj"))
+setClass("GeneralizedLinearRegressionModel", representation(jobj = "jobj"))
 
 #' @title S4 class that represents a NaiveBayesModel
 #' @param jobj a Java object reference to the backing Scala NaiveBayesWrapper
@@ -39,21 +39,18 @@ setClass("KMeansModel", representation(jobj = "jobj"))
 
 #' Fits a generalized linear model
 #'
-#' Fits a generalized linear model, similarly to R's glm(). Also see the glmnet package.
+#' Fits a generalized linear model, similarly to R's glm().
 #'
 #' @param formula A symbolic description of the model to be fitted. Currently only a few formula
 #'                operators are supported, including '~', '.', ':', '+', and '-'.
-#' @param data DataFrame for training
-#' @param family Error distribution. "gaussian" -> linear regression, "binomial" -> logistic reg.
-#' @param lambda Regularization parameter
-#' @param alpha Elastic-net mixing parameter (see glmnet's documentation for details)
-#' @param standardize Whether to standardize features before training
-#' @param solver The solver algorithm used for optimization, this can be "l-bfgs", "normal" and
-#'               "auto". "l-bfgs" denotes Limited-memory BFGS which is a limited-memory
-#'               quasi-Newton optimization method. "normal" denotes using Normal Equation as an
-#'               analytical solution to the linear regression problem. The default value is "auto"
-#'               which means that the solver algorithm is selected automatically.
-#' @return a fitted MLlib model
+#' @param data DataFrame for training.
+#' @param family A description of the error distribution and link function to be used in the model.
+#'               This can be a character string naming a family function, a family function or
+#'               the result of a call to a family function. Refer R family at
+#'               \url{https://stat.ethz.ch/R-manual/R-devel/library/stats/html/family.html}.
+#' @param epsilon Positive convergence tolerance of iterations.
+#' @param maxit Integer giving the maximal number of IRLS iterations.
+#' @return a fitted generalized linear model
 #' @rdname glm
 #' @export
 #' @examples
@@ -64,25 +61,59 @@ setClass("KMeansModel", representation(jobj = "jobj"))
 #' df <- createDataFrame(sqlContext, iris)
 #' model <- glm(Sepal_Length ~ Sepal_Width, df, family="gaussian")
 #' summary(model)
-#'}
+#' }
 setMethod("glm", signature(formula = "formula", family = "ANY", data = "DataFrame"),
-          function(formula, family = c("gaussian", "binomial"), data, lambda = 0, alpha = 0,
-            standardize = TRUE, solver = "auto") {
-            family <- match.arg(family)
+          function(formula, family = gaussian, data, epsilon = 1e-06, maxit = 25) {
+            if (is.character(family)) {
+              family <- get(family, mode = "function", envir = parent.frame())
+            }
+            if (is.function(family)) {
+              family <- family()
+            }
+            if (is.null(family$family)) {
+              print(family)
+              stop("'family' not recognized")
+            }
+
             formula <- paste(deparse(formula), collapse = "")
-            model <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                 "fitRModelFormula", formula, data@sdf, family, lambda,
-                                 alpha, standardize, solver)
-            return(new("PipelineModel", model = model))
+
+            jobj <- callJStatic("org.apache.spark.ml.r.GeneralizedLinearRegressionWrapper",
+                                "fit", formula, data@sdf, family$family, family$link,
+                                epsilon, as.integer(maxit))
+            return(new("GeneralizedLinearRegressionModel", jobj = jobj))
           })
 
-#' Make predictions from a model
+#' Get the summary of a generalized linear model
 #'
-#' Makes predictions from a model produced by glm(), similarly to R's predict().
+#' Returns the summary of a model produced by glm(), similarly to R's summary().
 #'
-#' @param object A fitted MLlib model
+#' @param object A fitted generalized linear model
+#' @return coefficients the model's coefficients, intercept
+#' @rdname summary
+#' @export
+#' @examples
+#' \dontrun{
+#' model <- glm(y ~ x, trainingData)
+#' summary(model)
+#' }
+setMethod("summary", signature(object = "GeneralizedLinearRegressionModel"),
+          function(object, ...) {
+            jobj <- object@jobj
+            features <- callJMethod(jobj, "rFeatures")
+            coefficients <- callJMethod(jobj, "rCoefficients")
+            coefficients <- as.matrix(unlist(coefficients))
+            colnames(coefficients) <- c("Estimate")
+            rownames(coefficients) <- unlist(features)
+            return(list(coefficients = coefficients))
+          })
+
+#' Make predictions from a generalized linear model
+#'
+#' Makes predictions from a generalized linear model produced by glm(), similarly to R's predict().
+#'
+#' @param object A fitted generalized linear model
 #' @param newData DataFrame for testing
-#' @return DataFrame containing predicted values
+#' @return DataFrame containing predicted labels in a column named "prediction"
 #' @rdname predict
 #' @export
 #' @examples
@@ -90,10 +121,10 @@ setMethod("glm", signature(formula = "formula", family = "ANY", data = "DataFram
 #' model <- glm(y ~ x, trainingData)
 #' predicted <- predict(model, testData)
 #' showDF(predicted)
-#'}
-setMethod("predict", signature(object = "PipelineModel"),
+#' }
+setMethod("predict", signature(object = "GeneralizedLinearRegressionModel"),
           function(object, newData) {
-            return(dataFrame(callJMethod(object@model, "transform", newData@sdf)))
+            return(dataFrame(callJMethod(object@jobj, "transform", newData@sdf)))
           })
 
 #' Make predictions from a naive Bayes model
@@ -116,54 +147,6 @@ setMethod("predict", signature(object = "NaiveBayesModel"),
             return(dataFrame(callJMethod(object@jobj, "transform", newData@sdf)))
           })
 
-#' Get the summary of a model
-#'
-#' Returns the summary of a model produced by glm(), similarly to R's summary().
-#'
-#' @param object A fitted MLlib model
-#' @return a list with 'devianceResiduals' and 'coefficients' components for gaussian family
-#'         or a list with 'coefficients' component for binomial family. \cr
-#'         For gaussian family: the 'devianceResiduals' gives the min/max deviance residuals
-#'         of the estimation, the 'coefficients' gives the estimated coefficients and their
-#'         estimated standard errors, t values and p-values. (It only available when model
-#'         fitted by normal solver.) \cr
-#'         For binomial family: the 'coefficients' gives the estimated coefficients.
-#'         See summary.glm for more information. \cr
-#' @rdname summary
-#' @export
-#' @examples
-#' \dontrun{
-#' model <- glm(y ~ x, trainingData)
-#' summary(model)
-#'}
-setMethod("summary", signature(object = "PipelineModel"),
-          function(object, ...) {
-            modelName <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                     "getModelName", object@model)
-            features <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                    "getModelFeatures", object@model)
-            coefficients <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                        "getModelCoefficients", object@model)
-            if (modelName == "LinearRegressionModel") {
-              devianceResiduals <- callJStatic("org.apache.spark.ml.api.r.SparkRWrappers",
-                                               "getModelDevianceResiduals", object@model)
-              devianceResiduals <- matrix(devianceResiduals, nrow = 1)
-              colnames(devianceResiduals) <- c("Min", "Max")
-              rownames(devianceResiduals) <- rep("", times = 1)
-              coefficients <- matrix(coefficients, ncol = 4)
-              colnames(coefficients) <- c("Estimate", "Std. Error", "t value", "Pr(>|t|)")
-              rownames(coefficients) <- unlist(features)
-              return(list(devianceResiduals = devianceResiduals, coefficients = coefficients))
-            } else if (modelName == "LogisticRegressionModel") {
-              coefficients <- as.matrix(unlist(coefficients))
-              colnames(coefficients) <- c("Estimate")
-              rownames(coefficients) <- unlist(features)
-              return(list(coefficients = coefficients))
-            } else {
-              stop(paste("Unsupported model", modelName, sep = " "))
-            }
-          })
-
 #' Get the summary of a naive Bayes model
 #'
 #' Returns the summary of a naive Bayes model produced by naiveBayes(), similarly to R's summary().
diff --git a/R/pkg/inst/tests/testthat/test_mllib.R b/R/pkg/inst/tests/testthat/test_mllib.R
index fdb591756e..a9dbd2bdc4 100644
--- a/R/pkg/inst/tests/testthat/test_mllib.R
+++ b/R/pkg/inst/tests/testthat/test_mllib.R
@@ -25,20 +25,21 @@ sc <- sparkR.init()
 
 sqlContext <- sparkRSQL.init(sc)
 
-test_that("glm and predict", {
+test_that("formula of glm", {
   training <- suppressWarnings(createDataFrame(sqlContext, iris))
-  test <- select(training, "Sepal_Length")
-  model <- glm(Sepal_Width ~ Sepal_Length, training, family = "gaussian")
-  prediction <- predict(model, test)
-  expect_equal(typeof(take(select(prediction, "prediction"), 1)$prediction), "double")
+  # dot minus and intercept vs native glm
+  model <- glm(Sepal_Width ~ . - Species + 0, data = training)
+  vals <- collect(select(predict(model, training), "prediction"))
+  rVals <- predict(glm(Sepal.Width ~ . - Species + 0, data = iris), iris)
+  expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
 
-  # Test stats::predict is working
-  x <- rnorm(15)
-  y <- x + rnorm(15)
-  expect_equal(length(predict(lm(y ~ x))), 15)
-})
+  # feature interaction vs native glm
+  model <- glm(Sepal_Width ~ Species:Sepal_Length, data = training)
+  vals <- collect(select(predict(model, training), "prediction"))
+  rVals <- predict(glm(Sepal.Width ~ Species:Sepal.Length, data = iris), iris)
+  expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
 
-test_that("glm should work with long formula", {
+  # glm should work with long formula
   training <- suppressWarnings(createDataFrame(sqlContext, iris))
   training$LongLongLongLongLongName <- training$Sepal_Width
   training$VeryLongLongLongLonLongName <- training$Sepal_Length
@@ -50,68 +51,30 @@ test_that("glm should work with long formula", {
   expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
 })
 
-test_that("predictions match with native glm", {
+test_that("glm and predict", {
   training <- suppressWarnings(createDataFrame(sqlContext, iris))
+  # gaussian family
   model <- glm(Sepal_Width ~ Sepal_Length + Species, data = training)
-  vals <- collect(select(predict(model, training), "prediction"))
+  prediction <- predict(model, training)
+  expect_equal(typeof(take(select(prediction, "prediction"), 1)$prediction), "double")
+  vals <- collect(select(prediction, "prediction"))
   rVals <- predict(glm(Sepal.Width ~ Sepal.Length + Species, data = iris), iris)
   expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
-})
-
-test_that("dot minus and intercept vs native glm", {
-  training <- suppressWarnings(createDataFrame(sqlContext, iris))
-  model <- glm(Sepal_Width ~ . - Species + 0, data = training)
-  vals <- collect(select(predict(model, training), "prediction"))
-  rVals <- predict(glm(Sepal.Width ~ . - Species + 0, data = iris), iris)
-  expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
-})
 
-test_that("feature interaction vs native glm", {
-  training <- suppressWarnings(createDataFrame(sqlContext, iris))
-  model <- glm(Sepal_Width ~ Species:Sepal_Length, data = training)
-  vals <- collect(select(predict(model, training), "prediction"))
-  rVals <- predict(glm(Sepal.Width ~ Species:Sepal.Length, data = iris), iris)
+  # poisson family
+  model <- glm(Sepal_Width ~ Sepal_Length + Species, data = training,
+               family = poisson(link = identity))
+  prediction <- predict(model, training)
+  expect_equal(typeof(take(select(prediction, "prediction"), 1)$prediction), "double")
+  vals <- collect(select(prediction, "prediction"))
+  rVals <- suppressWarnings(predict(glm(Sepal.Width ~ Sepal.Length + Species,
+           data = iris, family = poisson(link = identity)), iris))
   expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
-})
 
-test_that("summary coefficients match with native glm", {
-  training <- suppressWarnings(createDataFrame(sqlContext, iris))
-  stats <- summary(glm(Sepal_Width ~ Sepal_Length + Species, data = training, solver = "normal"))
-  coefs <- unlist(stats$coefficients)
-  devianceResiduals <- unlist(stats$devianceResiduals)
-
-  rStats <- summary(glm(Sepal.Width ~ Sepal.Length + Species, data = iris))
-  rCoefs <- unlist(rStats$coefficients)
-  rDevianceResiduals <- c(-0.95096, 0.72918)
-
-  expect_true(all(abs(rCoefs - coefs) < 1e-5))
-  expect_true(all(abs(rDevianceResiduals - devianceResiduals) < 1e-5))
-  expect_true(all(
-    rownames(stats$coefficients) ==
-    c("(Intercept)", "Sepal_Length", "Species_versicolor", "Species_virginica")))
-})
-
-test_that("summary coefficients match with native glm of family 'binomial'", {
-  df <- suppressWarnings(createDataFrame(sqlContext, iris))
-  training <- filter(df, df$Species != "setosa")
-  stats <- summary(glm(Species ~ Sepal_Length + Sepal_Width, data = training,
-    family = "binomial"))
-  coefs <- as.vector(stats$coefficients[, 1])
-
-  rTraining <- iris[iris$Species %in% c("versicolor", "virginica"), ]
-  rCoefs <- as.vector(coef(glm(Species ~ Sepal.Length + Sepal.Width, data = rTraining,
-    family = binomial(link = "logit"))))
-
-  expect_true(all(abs(rCoefs - coefs) < 1e-4))
-  expect_true(all(
-    rownames(stats$coefficients) ==
-    c("(Intercept)", "Sepal_Length", "Sepal_Width")))
-})
-
-test_that("summary works on base GLM models", {
-  baseModel <- stats::glm(Sepal.Width ~ Sepal.Length + Species, data = iris)
-  baseSummary <- summary(baseModel)
-  expect_true(abs(baseSummary$deviance - 12.19313) < 1e-4)
+  # Test stats::predict is working
+  x <- rnorm(15)
+  y <- x + rnorm(15)
+  expect_equal(length(predict(lm(y ~ x))), 15)
 })
 
 test_that("kmeans", {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/GeneralizedLinearRegressionWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/GeneralizedLinearRegressionWrapper.scala
new file mode 100644
index 0000000000..475a308385
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/GeneralizedLinearRegressionWrapper.scala
@@ -0,0 +1,79 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.r
+
+import org.apache.spark.ml.{Pipeline, PipelineModel}
+import org.apache.spark.ml.attribute.AttributeGroup
+import org.apache.spark.ml.feature.RFormula
+import org.apache.spark.ml.regression._
+import org.apache.spark.sql._
+
+private[r] class GeneralizedLinearRegressionWrapper private (
+    pipeline: PipelineModel,
+    val features: Array[String]) {
+
+  private val glm: GeneralizedLinearRegressionModel =
+    pipeline.stages(1).asInstanceOf[GeneralizedLinearRegressionModel]
+
+  lazy val rCoefficients: Array[Double] = if (glm.getFitIntercept) {
+    Array(glm.intercept) ++ glm.coefficients.toArray
+  } else {
+    glm.coefficients.toArray
+  }
+
+  lazy val rFeatures: Array[String] = if (glm.getFitIntercept) {
+    Array("(Intercept)") ++ features
+  } else {
+    features
+  }
+
+  def transform(dataset: DataFrame): DataFrame = {
+    pipeline.transform(dataset).drop(glm.getFeaturesCol)
+  }
+}
+
+private[r] object GeneralizedLinearRegressionWrapper {
+
+  def fit(
+      formula: String,
+      data: DataFrame,
+      family: String,
+      link: String,
+      epsilon: Double,
+      maxit: Int): GeneralizedLinearRegressionWrapper = {
+    val rFormula = new RFormula()
+      .setFormula(formula)
+    val rFormulaModel = rFormula.fit(data)
+    // get labels and feature names from output schema
+    val schema = rFormulaModel.transform(data).schema
+    val featureAttrs = AttributeGroup.fromStructField(schema(rFormula.getFeaturesCol))
+      .attributes.get
+    val features = featureAttrs.map(_.name.get)
+    // assemble and fit the pipeline
+    val glm = new GeneralizedLinearRegression()
+      .setFamily(family)
+      .setLink(link)
+      .setFitIntercept(rFormula.hasIntercept)
+      .setTol(epsilon)
+      .setMaxIter(maxit)
+    val pipeline = new Pipeline()
+      .setStages(Array(rFormulaModel, glm))
+      .fit(data)
+    new GeneralizedLinearRegressionWrapper(pipeline, features)
+  }
+}
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala b/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
deleted file mode 100644
index fa143715be..0000000000
--- a/mllib/src/main/scala/org/apache/spark/ml/r/SparkRWrappers.scala
+++ /dev/null
@@ -1,115 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.ml.api.r
-
-import org.apache.spark.ml.{Pipeline, PipelineModel}
-import org.apache.spark.ml.attribute._
-import org.apache.spark.ml.classification.{LogisticRegression, LogisticRegressionModel}
-import org.apache.spark.ml.feature.RFormula
-import org.apache.spark.ml.regression.{LinearRegression, LinearRegressionModel}
-import org.apache.spark.sql.DataFrame
-
-private[r] object SparkRWrappers {
-  def fitRModelFormula(
-      value: String,
-      df: DataFrame,
-      family: String,
-      lambda: Double,
-      alpha: Double,
-      standardize: Boolean,
-      solver: String): PipelineModel = {
-    val formula = new RFormula().setFormula(value)
-    val estimator = family match {
-      case "gaussian" => new LinearRegression()
-        .setRegParam(lambda)
-        .setElasticNetParam(alpha)
-        .setFitIntercept(formula.hasIntercept)
-        .setStandardization(standardize)
-        .setSolver(solver)
-      case "binomial" => new LogisticRegression()
-        .setRegParam(lambda)
-        .setElasticNetParam(alpha)
-        .setFitIntercept(formula.hasIntercept)
-        .setStandardization(standardize)
-    }
-    val pipeline = new Pipeline().setStages(Array(formula, estimator))
-    pipeline.fit(df)
-  }
-
-  def getModelCoefficients(model: PipelineModel): Array[Double] = {
-    model.stages.last match {
-      case m: LinearRegressionModel =>
-        val coefficientStandardErrorsR = Array(m.summary.coefficientStandardErrors.last) ++
-          m.summary.coefficientStandardErrors.dropRight(1)
-        val tValuesR = Array(m.summary.tValues.last) ++ m.summary.tValues.dropRight(1)
-        val pValuesR = Array(m.summary.pValues.last) ++ m.summary.pValues.dropRight(1)
-        if (m.getFitIntercept) {
-          Array(m.intercept) ++ m.coefficients.toArray ++ coefficientStandardErrorsR ++
-            tValuesR ++ pValuesR
-        } else {
-          m.coefficients.toArray ++ coefficientStandardErrorsR ++ tValuesR ++ pValuesR
-        }
-      case m: LogisticRegressionModel =>
-        if (m.getFitIntercept) {
-          Array(m.intercept) ++ m.coefficients.toArray
-        } else {
-          m.coefficients.toArray
-        }
-    }
-  }
-
-  def getModelDevianceResiduals(model: PipelineModel): Array[Double] = {
-    model.stages.last match {
-      case m: LinearRegressionModel =>
-        m.summary.devianceResiduals
-      case m: LogisticRegressionModel =>
-        throw new UnsupportedOperationException(
-          "No deviance residuals available for LogisticRegressionModel")
-    }
-  }
-
-  def getModelFeatures(model: PipelineModel): Array[String] = {
-    model.stages.last match {
-      case m: LinearRegressionModel =>
-        val attrs = AttributeGroup.fromStructField(
-          m.summary.predictions.schema(m.summary.featuresCol))
-        if (m.getFitIntercept) {
-          Array("(Intercept)") ++ attrs.attributes.get.map(_.name.get)
-        } else {
-          attrs.attributes.get.map(_.name.get)
-        }
-      case m: LogisticRegressionModel =>
-        val attrs = AttributeGroup.fromStructField(
-          m.summary.predictions.schema(m.summary.featuresCol))
-        if (m.getFitIntercept) {
-          Array("(Intercept)") ++ attrs.attributes.get.map(_.name.get)
-        } else {
-          attrs.attributes.get.map(_.name.get)
-        }
-    }
-  }
-
-  def getModelName(model: PipelineModel): String = {
-    model.stages.last match {
-      case m: LinearRegressionModel =>
-        "LinearRegressionModel"
-      case m: LogisticRegressionModel =>
-        "LogisticRegressionModel"
-    }
-  }
-}
-- 
cgit v1.2.3


From 101663f1ae222a919fc40510aa4f2bad22d1be6f Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Tue, 12 Apr 2016 11:27:16 -0700
Subject: [SPARK-13322][ML] AFTSurvivalRegression supports feature
 standardization

## What changes were proposed in this pull request?
AFTSurvivalRegression should support feature standardization, it will improve the convergence rate.
Test the convergence rate on the [Ovarian](https://stat.ethz.ch/R-manual/R-devel/library/survival/html/ovarian.html) data which is standard data comes with Survival library in R,
* without standardization(before this PR) -> 74 iterations.
* with standardization(after this PR) -> 38 iterations.

But after this fix, with or without ```standardization``` will converge to the same solution. It means that ```standardization = false``` will run the same code route as ```standardization = true```. Because if the features are not standardized at all, it will result convergency issue when the features have very different scales. This behavior is the same as ML [```LinearRegression``` and ```LogisticRegression```](https://issues.apache.org/jira/browse/SPARK-8522). See more discussion about this topic at #11247.
cc mengxr
## How was this patch tested?
unit test.

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #11365 from yanboliang/spark-13322.
---
 .../ml/regression/AFTSurvivalRegression.scala      | 105 ++++++++++++++-------
 .../ml/regression/AFTSurvivalRegressionSuite.scala |  22 +++++
 2 files changed, 93 insertions(+), 34 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
index afed1f32b6..89ba6ab5d2 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala
@@ -31,6 +31,7 @@ import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.linalg.{BLAS, Vector, Vectors, VectorUDT}
+import org.apache.spark.mllib.stat.MultivariateOnlineSummarizer
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
@@ -198,10 +199,20 @@ class AFTSurvivalRegression @Since("1.6.0") (@Since("1.6.0") override val uid: S
     val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE
     if (handlePersistence) instances.persist(StorageLevel.MEMORY_AND_DISK)
 
-    val costFun = new AFTCostFun(instances, $(fitIntercept))
+    val featuresSummarizer = {
+      val seqOp = (c: MultivariateOnlineSummarizer, v: AFTPoint) => c.add(v.features)
+      val combOp = (c1: MultivariateOnlineSummarizer, c2: MultivariateOnlineSummarizer) => {
+        c1.merge(c2)
+      }
+      instances.treeAggregate(new MultivariateOnlineSummarizer)(seqOp, combOp)
+    }
+
+    val featuresStd = featuresSummarizer.variance.toArray.map(math.sqrt)
+
+    val costFun = new AFTCostFun(instances, $(fitIntercept), featuresStd)
     val optimizer = new BreezeLBFGS[BDV[Double]]($(maxIter), 10, $(tol))
 
-    val numFeatures = dataset.select($(featuresCol)).take(1)(0).getAs[Vector](0).size
+    val numFeatures = featuresStd.size
     /*
        The parameters vector has three parts:
        the first element: Double, log(sigma), the log of scale parameter
@@ -230,7 +241,13 @@ class AFTSurvivalRegression @Since("1.6.0") (@Since("1.6.0") override val uid: S
 
     if (handlePersistence) instances.unpersist()
 
-    val coefficients = Vectors.dense(parameters.slice(2, parameters.length))
+    val rawCoefficients = parameters.slice(2, parameters.length)
+    var i = 0
+    while (i < numFeatures) {
+      rawCoefficients(i) *= { if (featuresStd(i) != 0.0) 1.0 / featuresStd(i) else 0.0 }
+      i += 1
+    }
+    val coefficients = Vectors.dense(rawCoefficients)
     val intercept = parameters(1)
     val scale = math.exp(parameters(0))
     val model = new AFTSurvivalRegressionModel(uid, coefficients, intercept, scale)
@@ -434,29 +451,36 @@ object AFTSurvivalRegressionModel extends MLReadable[AFTSurvivalRegressionModel]
  * @param parameters including three part: The log of scale parameter, the intercept and
  *                regression coefficients corresponding to the features.
  * @param fitIntercept Whether to fit an intercept term.
+ * @param featuresStd The standard deviation values of the features.
  */
-private class AFTAggregator(parameters: BDV[Double], fitIntercept: Boolean)
-  extends Serializable {
+private class AFTAggregator(
+    parameters: BDV[Double],
+    fitIntercept: Boolean,
+    featuresStd: Array[Double]) extends Serializable {
 
   // the regression coefficients to the covariates
   private val coefficients = parameters.slice(2, parameters.length)
-  private val intercept = parameters.valueAt(1)
+  private val intercept = parameters(1)
   // sigma is the scale parameter of the AFT model
   private val sigma = math.exp(parameters(0))
 
   private var totalCnt: Long = 0L
   private var lossSum = 0.0
-  private var gradientCoefficientSum = BDV.zeros[Double](coefficients.length)
-  private var gradientInterceptSum = 0.0
-  private var gradientLogSigmaSum = 0.0
+  // Here we optimize loss function over log(sigma), intercept and coefficients
+  private val gradientSumArray = Array.ofDim[Double](parameters.length)
 
   def count: Long = totalCnt
+  def loss: Double = {
+    require(totalCnt > 0.0, s"The number of instances should be " +
+      s"greater than 0.0, but got $totalCnt.")
+    lossSum / totalCnt
+  }
+  def gradient: BDV[Double] = {
+    require(totalCnt > 0.0, s"The number of instances should be " +
+      s"greater than 0.0, but got $totalCnt.")
+    new BDV(gradientSumArray.map(_ / totalCnt.toDouble))
+  }
 
-  def loss: Double = if (totalCnt == 0) 1.0 else lossSum / totalCnt
-
-  // Here we optimize loss function over coefficients, intercept and log(sigma)
-  def gradient: BDV[Double] = BDV.vertcat(BDV(Array(gradientLogSigmaSum / totalCnt.toDouble)),
-    BDV(Array(gradientInterceptSum/totalCnt.toDouble)), gradientCoefficientSum/totalCnt.toDouble)
 
   /**
    * Add a new training data to this AFTAggregator, and update the loss and gradient
@@ -466,25 +490,32 @@ private class AFTAggregator(parameters: BDV[Double], fitIntercept: Boolean)
    * @return This AFTAggregator object.
    */
   def add(data: AFTPoint): this.type = {
-
-    val interceptFlag = if (fitIntercept) 1.0 else 0.0
-
-    val xi = data.features.toBreeze
+    val xi = data.features
     val ti = data.label
     val delta = data.censor
-    val epsilon = (math.log(ti) - coefficients.dot(xi) - intercept * interceptFlag ) / sigma
 
-    lossSum += math.log(sigma) * delta
-    lossSum += (math.exp(epsilon) - delta * epsilon)
+    val margin = {
+      var sum = 0.0
+      xi.foreachActive { (index, value) =>
+        if (featuresStd(index) != 0.0 && value != 0.0) {
+          sum += coefficients(index) * (value / featuresStd(index))
+        }
+      }
+      sum + intercept
+    }
+    val epsilon = (math.log(ti) - margin) / sigma
+
+    lossSum += delta * math.log(sigma) - delta * epsilon + math.exp(epsilon)
 
-    // Sanity check (should never occur):
-    assert(!lossSum.isInfinity,
-      s"AFTAggregator loss sum is infinity. Error for unknown reason.")
+    val multiplier = (delta - math.exp(epsilon)) / sigma
 
-    val deltaMinusExpEps = delta - math.exp(epsilon)
-    gradientCoefficientSum += xi * deltaMinusExpEps / sigma
-    gradientInterceptSum += interceptFlag * deltaMinusExpEps / sigma
-    gradientLogSigmaSum += delta + deltaMinusExpEps * epsilon
+    gradientSumArray(0) += delta + multiplier * sigma * epsilon
+    gradientSumArray(1) += { if (fitIntercept) multiplier else 0.0 }
+    xi.foreachActive { (index, value) =>
+      if (featuresStd(index) != 0.0 && value != 0.0) {
+        gradientSumArray(index + 2) += multiplier * (value / featuresStd(index))
+      }
+    }
 
     totalCnt += 1
     this
@@ -503,9 +534,12 @@ private class AFTAggregator(parameters: BDV[Double], fitIntercept: Boolean)
       totalCnt += other.totalCnt
       lossSum += other.lossSum
 
-      gradientCoefficientSum += other.gradientCoefficientSum
-      gradientInterceptSum += other.gradientInterceptSum
-      gradientLogSigmaSum += other.gradientLogSigmaSum
+      var i = 0
+      val len = this.gradientSumArray.length
+      while (i < len) {
+        this.gradientSumArray(i) += other.gradientSumArray(i)
+        i += 1
+      }
     }
     this
   }
@@ -516,12 +550,15 @@ private class AFTAggregator(parameters: BDV[Double], fitIntercept: Boolean)
  * It returns the loss and gradient at a particular point (parameters).
  * It's used in Breeze's convex optimization routines.
  */
-private class AFTCostFun(data: RDD[AFTPoint], fitIntercept: Boolean)
-  extends DiffFunction[BDV[Double]] {
+private class AFTCostFun(
+    data: RDD[AFTPoint],
+    fitIntercept: Boolean,
+    featuresStd: Array[Double]) extends DiffFunction[BDV[Double]] {
 
   override def calculate(parameters: BDV[Double]): (Double, BDV[Double]) = {
 
-    val aftAggregator = data.treeAggregate(new AFTAggregator(parameters, fitIntercept))(
+    val aftAggregator = data.treeAggregate(
+      new AFTAggregator(parameters, fitIntercept, featuresStd))(
       seqOp = (c, v) => (c, v) match {
         case (aggregator, instance) => aggregator.add(instance)
       },
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala
index f4844cc671..76891ad562 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/AFTSurvivalRegressionSuite.scala
@@ -33,6 +33,7 @@ class AFTSurvivalRegressionSuite
 
   @transient var datasetUnivariate: DataFrame = _
   @transient var datasetMultivariate: DataFrame = _
+  @transient var datasetUnivariateScaled: DataFrame = _
 
   override def beforeAll(): Unit = {
     super.beforeAll()
@@ -42,6 +43,11 @@ class AFTSurvivalRegressionSuite
     datasetMultivariate = sqlContext.createDataFrame(
       sc.parallelize(generateAFTInput(
         2, Array(0.9, -1.3), Array(0.7, 1.2), 1000, 42, 1.5, 2.5, 2.0)))
+    datasetUnivariateScaled = sqlContext.createDataFrame(
+      sc.parallelize(generateAFTInput(
+        1, Array(5.5), Array(0.8), 1000, 42, 1.0, 2.0, 2.0)).map { x =>
+          AFTPoint(Vectors.dense(x.features(0) * 1.0E3), x.label, x.censor)
+      })
   }
 
   /**
@@ -356,6 +362,22 @@ class AFTSurvivalRegressionSuite
       }
   }
 
+  test("numerical stability of standardization") {
+    val trainer = new AFTSurvivalRegression()
+    val model1 = trainer.fit(datasetUnivariate)
+    val model2 = trainer.fit(datasetUnivariateScaled)
+
+    /**
+     * During training we standardize the dataset first, so no matter how we multiple
+     * a scaling factor into the dataset, the convergence rate should be the same,
+     * and the coefficients should equal to the original coefficients multiple by
+     * the scaling factor. It will have no effect on the intercept and scale.
+     */
+    assert(model1.coefficients(0) ~== model2.coefficients(0) * 1.0E3 absTol 0.01)
+    assert(model1.intercept ~== model2.intercept absTol 0.01)
+    assert(model1.scale ~== model2.scale absTol 0.01)
+  }
+
   test("read/write") {
     def checkModelData(
         model: AFTSurvivalRegressionModel,
-- 
cgit v1.2.3


From 1995c2e6482bf4af5a4be087bfc156311c1bec19 Mon Sep 17 00:00:00 2001
From: Xiangrui Meng <meng@databricks.com>
Date: Tue, 12 Apr 2016 11:30:09 -0700
Subject: [SPARK-14563][ML] use a random table name instead of __THIS__ in
 SQLTransformer

## What changes were proposed in this pull request?

Use a random table name instead of `__THIS__` in SQLTransformer, and add a test for `transformSchema`. The problems of using `__THIS__` are:

* It doesn't work under HiveContext (in Spark 1.6)
* Race conditions

## How was this patch tested?

* Manual test with HiveContext.
* Added a unit test for `transformSchema` to improve coverage.

cc: yhuai

Author: Xiangrui Meng <meng@databricks.com>

Closes #12330 from mengxr/SPARK-14563.
---
 .../scala/org/apache/spark/ml/feature/SQLTransformer.scala     | 10 ++++++----
 .../org/apache/spark/ml/feature/SQLTransformerSuite.scala      | 10 ++++++++++
 2 files changed, 16 insertions(+), 4 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala
index 95fe942c6b..2002d15745 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/SQLTransformer.scala
@@ -68,8 +68,7 @@ class SQLTransformer @Since("1.6.0") (override val uid: String) extends Transfor
     val tableName = Identifiable.randomUID(uid)
     dataset.registerTempTable(tableName)
     val realStatement = $(statement).replace(tableIdentifier, tableName)
-    val outputDF = dataset.sqlContext.sql(realStatement)
-    outputDF
+    dataset.sqlContext.sql(realStatement)
   }
 
   @Since("1.6.0")
@@ -78,8 +77,11 @@ class SQLTransformer @Since("1.6.0") (override val uid: String) extends Transfor
     val sqlContext = SQLContext.getOrCreate(sc)
     val dummyRDD = sc.parallelize(Seq(Row.empty))
     val dummyDF = sqlContext.createDataFrame(dummyRDD, schema)
-    dummyDF.registerTempTable(tableIdentifier)
-    val outputSchema = sqlContext.sql($(statement)).schema
+    val tableName = Identifiable.randomUID(uid)
+    val realStatement = $(statement).replace(tableIdentifier, tableName)
+    dummyDF.registerTempTable(tableName)
+    val outputSchema = sqlContext.sql(realStatement).schema
+    sqlContext.dropTempTable(tableName)
     outputSchema
   }
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/feature/SQLTransformerSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/feature/SQLTransformerSuite.scala
index 553e0b8702..e213e17d0d 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/feature/SQLTransformerSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/feature/SQLTransformerSuite.scala
@@ -21,6 +21,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.param.ParamsSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.sql.types.{LongType, StructField, StructType}
 
 class SQLTransformerSuite
   extends SparkFunSuite with MLlibTestSparkContext with DefaultReadWriteTest {
@@ -49,4 +50,13 @@ class SQLTransformerSuite
       .setStatement("select * from __THIS__")
     testDefaultReadWrite(t)
   }
+
+  test("transformSchema") {
+    val df = sqlContext.range(10)
+    val outputSchema = new SQLTransformer()
+      .setStatement("SELECT id + 1 AS id1 FROM __THIS__")
+      .transformSchema(df.schema)
+    val expected = StructType(Seq(StructField("id1", LongType, nullable = false)))
+    assert(outputSchema === expected)
+  }
 }
-- 
cgit v1.2.3


From 111a62474a2fb7f4e7f19fcfb8efaae37aa40400 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Tue, 12 Apr 2016 11:34:40 -0700
Subject: [SPARK-14147][ML][SPARKR] SparkR predict should not output feature
 column

## What changes were proposed in this pull request?
SparkR does not support type of vector which is the default type of feature column in ML. R predict also does not output intermediate feature column. So SparkR ```predict``` should not output feature column. In this PR, I only fix this issue for ```naiveBayes``` and ```survreg```. ```kmeans``` has the right code route already and  ```glm``` will be fixed at SparkRWrapper refactor(#12294).

## How was this patch tested?
No new tests.

cc mengxr shivaram

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #11958 from yanboliang/spark-14147.
---
 .../scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala    | 2 +-
 mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala    | 4 +++-
 2 files changed, 4 insertions(+), 2 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala
index 2ae411555f..7835468626 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/AFTSurvivalRegressionWrapper.scala
@@ -44,7 +44,7 @@ private[r] class AFTSurvivalRegressionWrapper private (
   }
 
   def transform(dataset: Dataset[_]): DataFrame = {
-    pipeline.transform(dataset)
+    pipeline.transform(dataset).drop(aftModel.getFeaturesCol)
   }
 }
 
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala
index 2cd709d2ee..b17207e99b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/NaiveBayesWrapper.scala
@@ -37,7 +37,9 @@ private[r] class NaiveBayesWrapper private (
   lazy val tables: Array[Double] = naiveBayesModel.theta.toArray.map(math.exp)
 
   def transform(dataset: Dataset[_]): DataFrame = {
-    pipeline.transform(dataset).drop(PREDICTED_LABEL_INDEX_COL)
+    pipeline.transform(dataset)
+      .drop(PREDICTED_LABEL_INDEX_COL)
+      .drop(naiveBayesModel.getFeaturesCol)
   }
 }
 
-- 
cgit v1.2.3


From 587cd554af24601d332e9ce5c74e98b62d0fd830 Mon Sep 17 00:00:00 2001
From: hyukjinkwon <gurwls223@gmail.com>
Date: Wed, 13 Apr 2016 10:20:03 +0800
Subject: [MINOR][SQL] Remove some unused imports in datasources.

## What changes were proposed in this pull request?

It looks several recent commits for datasources (maybe while removing old `HadoopFsRelation` interface) missed removing some unused imports.

This PR removes some unused imports in datasources.

## How was this patch tested?

`sbt scalastyle` and some unit tests for them.

Author: hyukjinkwon <gurwls223@gmail.com>

Closes #12326 from HyukjinKwon/minor-imports.
---
 .../org/apache/spark/ml/source/libsvm/LibSVMRelation.scala   |  3 ---
 .../spark/sql/execution/datasources/DataSourceStrategy.scala |  8 +-------
 .../execution/datasources/InsertIntoHadoopFsRelation.scala   |  1 -
 .../spark/sql/execution/datasources/SqlNewHadoopRDD.scala    |  2 --
 .../spark/sql/execution/datasources/WriterContainer.scala    |  2 +-
 .../spark/sql/execution/datasources/json/JSONRelation.scala  |  4 +---
 .../spark/sql/execution/datasources/json/JacksonParser.scala |  6 +++---
 .../sql/execution/datasources/parquet/ParquetRelation.scala  | 12 +++---------
 .../spark/sql/execution/datasources/text/DefaultSource.scala |  7 +------
 .../scala/org/apache/spark/sql/hive/orc/OrcRelation.scala    |  2 --
 10 files changed, 10 insertions(+), 37 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala b/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
index 4737b6fe52..2f1f2523fd 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/source/libsvm/LibSVMRelation.scala
@@ -26,11 +26,9 @@ import org.apache.hadoop.mapreduce.{Job, RecordWriter, TaskAttemptContext}
 import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat
 
 import org.apache.spark.annotation.Since
-import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.mllib.linalg.{Vector, Vectors, VectorUDT}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.util.MLUtils
-import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, DataFrameReader, Row, SQLContext}
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.encoders.RowEncoder
@@ -40,7 +38,6 @@ import org.apache.spark.sql.execution.datasources.{CaseInsensitiveMap, HadoopFil
 import org.apache.spark.sql.sources._
 import org.apache.spark.sql.types._
 import org.apache.spark.util.SerializableConfiguration
-import org.apache.spark.util.collection.BitSet
 
 private[libsvm] class LibSVMOutputWriter(
     path: String,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala
index c3885a3be5..ac3c52e901 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala
@@ -19,10 +19,8 @@ package org.apache.spark.sql.execution.datasources
 
 import scala.collection.mutable.ArrayBuffer
 
-import org.apache.spark.TaskContext
-import org.apache.spark.deploy.SparkHadoopUtil
 import org.apache.spark.internal.Logging
-import org.apache.spark.rdd.{MapPartitionsRDD, RDD, UnionRDD}
+import org.apache.spark.rdd.RDD
 import org.apache.spark.sql._
 import org.apache.spark.sql.catalyst.{CatalystTypeConverters, InternalRow}
 import org.apache.spark.sql.catalyst.CatalystTypeConverters.convertToScala
@@ -35,14 +33,10 @@ import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
 import org.apache.spark.sql.catalyst.plans.physical.HashPartitioning
 import org.apache.spark.sql.catalyst.rules.Rule
 import org.apache.spark.sql.execution.DataSourceScan.{INPUT_PATHS, PUSHED_FILTERS}
-import org.apache.spark.sql.execution.SparkPlan
 import org.apache.spark.sql.execution.command.ExecutedCommand
-import org.apache.spark.sql.execution.vectorized.{ColumnarBatch, ColumnVectorUtils}
 import org.apache.spark.sql.sources._
 import org.apache.spark.sql.types._
 import org.apache.spark.unsafe.types.UTF8String
-import org.apache.spark.util.{SerializableConfiguration, Utils}
-import org.apache.spark.util.collection.BitSet
 
 /**
  * Replaces generic operations with specific variants that are designed to work with Spark
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InsertIntoHadoopFsRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InsertIntoHadoopFsRelation.scala
index e31380e17d..889c0204f8 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InsertIntoHadoopFsRelation.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InsertIntoHadoopFsRelation.scala
@@ -32,7 +32,6 @@ import org.apache.spark.sql.execution.SQLExecution
 import org.apache.spark.sql.execution.command.RunnableCommand
 import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.sql.sources._
-import org.apache.spark.util.Utils
 
 /**
  * A command for writing data to a [[HadoopFsRelation]].  Supports both overwriting and appending.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala
index 6ddb218a22..4d6864d8ba 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SqlNewHadoopRDD.scala
@@ -34,8 +34,6 @@ import org.apache.spark.deploy.SparkHadoopUtil
 import org.apache.spark.executor.DataReadMethod
 import org.apache.spark.internal.Logging
 import org.apache.spark.sql.SQLContext
-import org.apache.spark.sql.execution.datasources.parquet.VectorizedParquetRecordReader
-import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.storage.StorageLevel
 import org.apache.spark.util.{SerializableConfiguration, ShutdownHookManager}
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/WriterContainer.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/WriterContainer.scala
index b9a3162aba..815d1d01ef 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/WriterContainer.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/WriterContainer.scala
@@ -33,7 +33,7 @@ import org.apache.spark.sql.catalyst.plans.physical.HashPartitioning
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.execution.UnsafeKVExternalSorter
 import org.apache.spark.sql.internal.SQLConf
-import org.apache.spark.sql.sources.{HadoopFsRelation, OutputWriter, OutputWriterFactory}
+import org.apache.spark.sql.sources.{OutputWriter, OutputWriterFactory}
 import org.apache.spark.sql.types.{IntegerType, StringType, StructField, StructType}
 import org.apache.spark.util.{SerializableConfiguration, Utils}
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala
index f32fea4183..7364a1dc06 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JSONRelation.scala
@@ -28,18 +28,16 @@ import org.apache.hadoop.mapreduce.{Job, RecordWriter, TaskAttemptContext}
 import org.apache.hadoop.mapreduce.lib.input.FileInputFormat
 import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat
 
-import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.internal.Logging
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{AnalysisException, Row, SQLContext}
 import org.apache.spark.sql.catalyst.InternalRow
-import org.apache.spark.sql.catalyst.expressions.{JoinedRow, UnsafeProjection}
+import org.apache.spark.sql.catalyst.expressions.JoinedRow
 import org.apache.spark.sql.catalyst.expressions.codegen.GenerateUnsafeProjection
 import org.apache.spark.sql.execution.datasources._
 import org.apache.spark.sql.sources._
 import org.apache.spark.sql.types.StructType
 import org.apache.spark.util.SerializableConfiguration
-import org.apache.spark.util.collection.BitSet
 
 class DefaultSource extends FileFormat with DataSourceRegister {
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JacksonParser.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JacksonParser.scala
index 8bc53bae6c..aeee2600a1 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JacksonParser.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/json/JacksonParser.scala
@@ -54,9 +54,9 @@ object JacksonParser extends Logging {
    * with an array.
    */
   def convertRootField(
-        factory: JsonFactory,
-        parser: JsonParser,
-        schema: DataType): Any = {
+      factory: JsonFactory,
+      parser: JsonParser,
+      schema: DataType): Any = {
     import com.fasterxml.jackson.core.JsonToken._
     (parser.getCurrentToken, schema) match {
       case (START_ARRAY, st: StructType) =>
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
index dbda094996..b91e892f8f 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
@@ -18,7 +18,6 @@
 package org.apache.spark.sql.execution.datasources.parquet
 
 import java.net.URI
-import java.util.{List => JList}
 import java.util.logging.{Logger => JLogger}
 
 import scala.collection.JavaConverters._
@@ -27,23 +26,19 @@ import scala.util.{Failure, Try}
 
 import org.apache.hadoop.conf.Configuration
 import org.apache.hadoop.fs.{FileStatus, Path}
-import org.apache.hadoop.io.Writable
 import org.apache.hadoop.mapreduce._
 import org.apache.hadoop.mapreduce.lib.input.{FileInputFormat, FileSplit}
-import org.apache.hadoop.mapreduce.task.{JobContextImpl, TaskAttemptContextImpl}
+import org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl
 import org.apache.parquet.{Log => ApacheParquetLog}
 import org.apache.parquet.filter2.compat.FilterCompat
 import org.apache.parquet.filter2.predicate.FilterApi
 import org.apache.parquet.hadoop._
-import org.apache.parquet.hadoop.metadata.CompressionCodecName
 import org.apache.parquet.hadoop.util.ContextUtil
 import org.apache.parquet.schema.MessageType
 import org.slf4j.bridge.SLF4JBridgeHandler
 
-import org.apache.spark.{Partition => SparkPartition, SparkException}
-import org.apache.spark.broadcast.Broadcast
+import org.apache.spark.SparkException
 import org.apache.spark.internal.Logging
-import org.apache.spark.rdd.{RDD, SqlNewHadoopPartition, SqlNewHadoopRDD}
 import org.apache.spark.sql._
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions.JoinedRow
@@ -53,8 +48,7 @@ import org.apache.spark.sql.execution.datasources._
 import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.sql.sources._
 import org.apache.spark.sql.types.{AtomicType, DataType, StructType}
-import org.apache.spark.util.{SerializableConfiguration, Utils}
-import org.apache.spark.util.collection.BitSet
+import org.apache.spark.util.SerializableConfiguration
 
 private[sql] class DefaultSource
   extends FileFormat
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala
index 28b03ee7c3..94ecb7a286 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/text/DefaultSource.scala
@@ -19,14 +19,10 @@ package org.apache.spark.sql.execution.datasources.text
 
 import org.apache.hadoop.conf.Configuration
 import org.apache.hadoop.fs.{FileStatus, Path}
-import org.apache.hadoop.io.{LongWritable, NullWritable, Text}
-import org.apache.hadoop.mapred.{JobConf, TextInputFormat}
+import org.apache.hadoop.io.{NullWritable, Text}
 import org.apache.hadoop.mapreduce.{Job, RecordWriter, TaskAttemptContext}
-import org.apache.hadoop.mapreduce.lib.input.FileInputFormat
 import org.apache.hadoop.mapreduce.lib.output.TextOutputFormat
 
-import org.apache.spark.broadcast.Broadcast
-import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{AnalysisException, Row, SQLContext}
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions.UnsafeRow
@@ -35,7 +31,6 @@ import org.apache.spark.sql.execution.datasources.{CompressionCodecs, HadoopFile
 import org.apache.spark.sql.sources._
 import org.apache.spark.sql.types.{StringType, StructType}
 import org.apache.spark.util.SerializableConfiguration
-import org.apache.spark.util.collection.BitSet
 
 /**
  * A data source for reading text files.
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala
index e915f3dfe2..21591ec093 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/orc/OrcRelation.scala
@@ -33,7 +33,6 @@ import org.apache.hadoop.mapreduce._
 import org.apache.hadoop.mapreduce.lib.input.{FileInputFormat, FileSplit}
 import org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl
 
-import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.internal.Logging
 import org.apache.spark.rdd.{HadoopRDD, RDD}
 import org.apache.spark.sql.{Row, SQLContext}
@@ -45,7 +44,6 @@ import org.apache.spark.sql.hive.{HiveInspectors, HiveMetastoreTypes, HiveShim}
 import org.apache.spark.sql.sources.{Filter, _}
 import org.apache.spark.sql.types.StructType
 import org.apache.spark.util.SerializableConfiguration
-import org.apache.spark.util.collection.BitSet
 
 private[sql] class DefaultSource
   extends FileFormat with DataSourceRegister with Serializable {
-- 
cgit v1.2.3


From 323e7390a5c123c48cc7d6d9be44bee3a7eecd99 Mon Sep 17 00:00:00 2001
From: Xiangrui Meng <meng@databricks.com>
Date: Wed, 13 Apr 2016 09:17:46 -0700
Subject: Revert "[SPARK-14154][MLLIB] Simplify the implementation for
 Kolmogorov–Smirnov test"
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This reverts commit d2a819a6363190b946986ebf6f8001d520098c3b.
---
 .../mllib/stat/test/KolmogorovSmirnovTest.scala    | 77 ++++++++++++++++++++--
 1 file changed, 73 insertions(+), 4 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala b/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
index ef284531c9..9748fbf2c9 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/stat/test/KolmogorovSmirnovTest.scala
@@ -64,10 +64,11 @@ private[stat] object KolmogorovSmirnovTest extends Logging {
    */
   def testOneSample(data: RDD[Double], cdf: Double => Double): KolmogorovSmirnovTestResult = {
     val n = data.count().toDouble
-    val ksStat = data.sortBy(x => x).zipWithIndex().map { case (v, i) =>
-      val f = cdf(v)
-      math.max(f - i / n, (i + 1) / n - f)
-    }.max()
+    val localData = data.sortBy(x => x).mapPartitions { part =>
+      val partDiffs = oneSampleDifferences(part, n, cdf) // local distances
+      searchOneSampleCandidates(partDiffs) // candidates: local extrema
+    }.collect()
+    val ksStat = searchOneSampleStatistic(localData, n) // result: global extreme
     evalOneSampleP(ksStat, n.toLong)
   }
 
@@ -83,6 +84,74 @@ private[stat] object KolmogorovSmirnovTest extends Logging {
     testOneSample(data, cdf)
   }
 
+  /**
+   * Calculate unadjusted distances between the empirical CDF and the theoretical CDF in a
+   * partition
+   * @param partData `Iterator[Double]` 1 partition of a sorted RDD
+   * @param n `Double` the total size of the RDD
+   * @param cdf `Double => Double` a function the calculates the theoretical CDF of a value
+   * @return `Iterator[(Double, Double)] `Unadjusted (ie. off by a constant) potential extrema
+   *        in a partition. The first element corresponds to the (empirical CDF - 1/N) - CDF,
+   *        the second element corresponds to empirical CDF - CDF.  We can then search the resulting
+   *        iterator for the minimum of the first and the maximum of the second element, and provide
+   *        this as a partition's candidate extrema
+   */
+  private def oneSampleDifferences(partData: Iterator[Double], n: Double, cdf: Double => Double)
+    : Iterator[(Double, Double)] = {
+    // zip data with index (within that partition)
+    // calculate local (unadjusted) empirical CDF and subtract CDF
+    partData.zipWithIndex.map { case (v, ix) =>
+      // dp and dl are later adjusted by constant, when global info is available
+      val dp = (ix + 1) / n
+      val dl = ix / n
+      val cdfVal = cdf(v)
+      (dl - cdfVal, dp - cdfVal)
+    }
+  }
+
+  /**
+   * Search the unadjusted differences in a partition and return the
+   * two extrema (furthest below and furthest above CDF), along with a count of elements in that
+   * partition
+   * @param partDiffs `Iterator[(Double, Double)]` the unadjusted differences between empirical CDF
+   *                 and CDFin a partition, which come as a tuple of
+   *                 (empirical CDF - 1/N - CDF, empirical CDF - CDF)
+   * @return `Iterator[(Double, Double, Double)]` the local extrema and a count of elements
+   */
+  private def searchOneSampleCandidates(partDiffs: Iterator[(Double, Double)])
+    : Iterator[(Double, Double, Double)] = {
+    val initAcc = (Double.MaxValue, Double.MinValue, 0.0)
+    val pResults = partDiffs.foldLeft(initAcc) { case ((pMin, pMax, pCt), (dl, dp)) =>
+      (math.min(pMin, dl), math.max(pMax, dp), pCt + 1)
+    }
+    val results = if (pResults == initAcc) Array[(Double, Double, Double)]() else Array(pResults)
+    results.iterator
+  }
+
+  /**
+   * Find the global maximum distance between empirical CDF and CDF (i.e. the KS statistic) after
+   * adjusting local extrema estimates from individual partitions with the amount of elements in
+   * preceding partitions
+   * @param localData `Array[(Double, Double, Double)]` A local array containing the collected
+   *                 results of `searchOneSampleCandidates` across all partitions
+   * @param n `Double`The size of the RDD
+   * @return The one-sample Kolmogorov Smirnov Statistic
+   */
+  private def searchOneSampleStatistic(localData: Array[(Double, Double, Double)], n: Double)
+    : Double = {
+    val initAcc = (Double.MinValue, 0.0)
+    // adjust differences based on the number of elements preceding it, which should provide
+    // the correct distance between empirical CDF and CDF
+    val results = localData.foldLeft(initAcc) { case ((prevMax, prevCt), (minCand, maxCand, ct)) =>
+      val adjConst = prevCt / n
+      val dist1 = math.abs(minCand + adjConst)
+      val dist2 = math.abs(maxCand + adjConst)
+      val maxVal = Array(prevMax, dist1, dist2).max
+      (maxVal, prevCt + ct)
+    }
+    results._1
+  }
+
   /**
    * A convenience function that allows running the KS test for 1 set of sample data against
    * a named distribution
-- 
cgit v1.2.3


From 1018a1c1eb33eefbfb9025fac7a1cdafc5cbf8f8 Mon Sep 17 00:00:00 2001
From: Timothy Hunter <timhunter@databricks.com>
Date: Wed, 13 Apr 2016 11:06:42 -0700
Subject: [SPARK-14568][ML] Instrumentation framework for logistic regression

## What changes were proposed in this pull request?

This adds extra logging information about a `LogisticRegression` estimator when being fit on a dataset. With this PR, you see the following extra lines when running the example in the documentation:

```
16/04/13 07:19:00 INFO Instrumentation: Instrumentation(LogisticRegression-logreg_55dd3c09f164-1230977381-1): training: numPartitions=1 storageLevel=StorageLevel(disk=true, memory=true, offheap=false, deserialized=true, replication=1)
16/04/13 07:19:00 INFO Instrumentation: Instrumentation(LogisticRegression-logreg_55dd3c09f164-1230977381-1): {"regParam":0.3,"elasticNetParam":0.8,"maxIter":10}
...
16/04/12 11:48:07 INFO Instrumentation: Instrumentation(LogisticRegression-logreg_a89eb23cb386-358781145):numClasses=2
16/04/12 11:48:07 INFO Instrumentation: Instrumentation(LogisticRegression-logreg_a89eb23cb386-358781145):numFeatures=692
...
16/04/13 07:19:01 INFO Instrumentation: Instrumentation(LogisticRegression-logreg_55dd3c09f164-1230977381-1): training finished
```

## How was this patch tested?

This PR was manually tested.

Author: Timothy Hunter <timhunter@databricks.com>

Closes #12331 from thunterdb/1604-instrumentation.
---
 .../ml/classification/LogisticRegression.scala     |  11 +-
 .../org/apache/spark/ml/util/Instrumentation.scala | 117 +++++++++++++++++++++
 2 files changed, 127 insertions(+), 1 deletion(-)
 create mode 100644 mllib/src/main/scala/org/apache/spark/ml/util/Instrumentation.scala

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
index 4a3fe5c663..c2b440059b 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
@@ -273,6 +273,10 @@ class LogisticRegression @Since("1.2.0") (
 
     if (handlePersistence) instances.persist(StorageLevel.MEMORY_AND_DISK)
 
+    val instr = Instrumentation.create(this, instances)
+    instr.logParams(regParam, elasticNetParam, standardization, threshold,
+      maxIter, tol, fitIntercept)
+
     val (summarizer, labelSummarizer) = {
       val seqOp = (c: (MultivariateOnlineSummarizer, MultiClassSummarizer),
         instance: Instance) =>
@@ -291,6 +295,9 @@ class LogisticRegression @Since("1.2.0") (
     val numClasses = histogram.length
     val numFeatures = summarizer.mean.size
 
+    instr.logNumClasses(numClasses)
+    instr.logNumFeatures(numFeatures)
+
     val (coefficients, intercept, objectiveHistory) = {
       if (numInvalid != 0) {
         val msg = s"Classification labels should be in {0 to ${numClasses - 1} " +
@@ -444,7 +451,9 @@ class LogisticRegression @Since("1.2.0") (
       $(labelCol),
       $(featuresCol),
       objectiveHistory)
-    model.setSummary(logRegSummary)
+    val m = model.setSummary(logRegSummary)
+    instr.logSuccess(m)
+    m
   }
 
   @Since("1.4.0")
diff --git a/mllib/src/main/scala/org/apache/spark/ml/util/Instrumentation.scala b/mllib/src/main/scala/org/apache/spark/ml/util/Instrumentation.scala
new file mode 100644
index 0000000000..7e57cefc44
--- /dev/null
+++ b/mllib/src/main/scala/org/apache/spark/ml/util/Instrumentation.scala
@@ -0,0 +1,117 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.util
+
+import java.util.concurrent.atomic.AtomicLong
+
+import org.json4s._
+import org.json4s.JsonDSL._
+import org.json4s.jackson.JsonMethods._
+
+import org.apache.spark.internal.Logging
+import org.apache.spark.ml.{Estimator, Model}
+import org.apache.spark.ml.param.Param
+import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.Dataset
+
+/**
+ * A small wrapper that defines a training session for an estimator, and some methods to log
+ * useful information during this session.
+ *
+ * A new instance is expected to be created within fit().
+ *
+ * @param estimator the estimator that is being fit
+ * @param dataset the training dataset
+ * @tparam E the type of the estimator
+ */
+private[ml] class Instrumentation[E <: Estimator[_]] private (
+    estimator: E, dataset: RDD[_]) extends Logging {
+
+  private val id = Instrumentation.counter.incrementAndGet()
+  private val prefix = {
+    val className = estimator.getClass.getSimpleName
+    s"$className-${estimator.uid}-${dataset.hashCode()}-$id: "
+  }
+
+  init()
+
+  private def init(): Unit = {
+    log(s"training: numPartitions=${dataset.partitions.length}" +
+      s" storageLevel=${dataset.getStorageLevel}")
+  }
+
+  /**
+   * Logs a message with a prefix that uniquely identifies the training session.
+   */
+  def log(msg: String): Unit = {
+    logInfo(prefix + msg)
+  }
+
+  /**
+   * Logs the value of the given parameters for the estimator being used in this session.
+   */
+  def logParams(params: Param[_]*): Unit = {
+    val pairs: Seq[(String, JValue)] = for {
+      p <- params
+      value <- estimator.get(p)
+    } yield {
+      val cast = p.asInstanceOf[Param[Any]]
+      p.name -> parse(cast.jsonEncode(value))
+    }
+    log(compact(render(map2jvalue(pairs.toMap))))
+  }
+
+  def logNumFeatures(num: Long): Unit = {
+    log(compact(render("numFeatures" -> num)))
+  }
+
+  def logNumClasses(num: Long): Unit = {
+    log(compact(render("numClasses" -> num)))
+  }
+
+  /**
+   * Logs the successful completion of the training session and the value of the learned model.
+   */
+  def logSuccess(model: Model[_]): Unit = {
+    log(s"training finished")
+  }
+}
+
+/**
+ * Some common methods for logging information about a training session.
+ */
+private[ml] object Instrumentation {
+  private val counter = new AtomicLong(0)
+
+  /**
+   * Creates an instrumentation object for a training session.
+   */
+  def create[E <: Estimator[_]](
+      estimator: E, dataset: Dataset[_]): Instrumentation[E] = {
+    create[E](estimator, dataset.rdd)
+  }
+
+  /**
+   * Creates an instrumentation object for a training session.
+   */
+  def create[E <: Estimator[_]](
+      estimator: E, dataset: RDD[_]): Instrumentation[E] = {
+    new Instrumentation[E](estimator, dataset)
+  }
+
+}
-- 
cgit v1.2.3


From f9d578eaa107d8e8503c1563a2b3990c85104298 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Wed, 13 Apr 2016 11:31:10 -0700
Subject: [SPARK-13783][ML] Model export/import for spark.ml: GBTs

## What changes were proposed in this pull request?
* Added save/load for ```GBTClassifier/GBTClassificationModel/GBTRegressor/GBTRegressionModel```.
* Meanwhile, I modified ```EnsembleModelReadWrite.saveImpl/loadImpl``` to support save/load ```treeWeights```.

## How was this patch tested?
Adds standard unit tests for GBT save/load.

cc jkbradley GayathriMurali

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12230 from yanboliang/spark-13783.
---
 .../spark/ml/classification/GBTClassifier.scala    | 110 ++++++++++++--------
 .../ml/classification/RandomForestClassifier.scala |   2 +-
 .../apache/spark/ml/regression/GBTRegressor.scala  | 114 +++++++++++++--------
 .../ml/regression/RandomForestRegressor.scala      |   2 +-
 .../org/apache/spark/ml/tree/treeModels.scala      |  25 +++--
 .../org/apache/spark/ml/tree/treeParams.scala      |  73 ++++++++++++-
 .../ml/classification/GBTClassifierSuite.scala     |  37 +++----
 .../spark/ml/regression/GBTRegressorSuite.scala    |  36 +++----
 8 files changed, 262 insertions(+), 137 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
index 46e8b89d01..39a698af15 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -18,19 +18,21 @@
 package org.apache.spark.ml.classification
 
 import com.github.fommil.netlib.BLAS.{getInstance => blas}
+import org.json4s.{DefaultFormats, JObject}
+import org.json4s.JsonDSL._
 
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{PredictionModel, Predictor}
-import org.apache.spark.ml.param.{Param, ParamMap}
+import org.apache.spark.ml.param.ParamMap
 import org.apache.spark.ml.regression.DecisionTreeRegressionModel
-import org.apache.spark.ml.tree.{GBTParams, TreeClassifierParams, TreeEnsembleModel}
+import org.apache.spark.ml.tree._
 import org.apache.spark.ml.tree.impl.GradientBoostedTrees
-import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
+import org.apache.spark.ml.util._
+import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
-import org.apache.spark.mllib.tree.loss.{LogLoss => OldLogLoss, Loss => OldLoss}
 import org.apache.spark.mllib.tree.model.{GradientBoostedTreesModel => OldGBTModel}
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Dataset}
@@ -58,7 +60,7 @@ import org.apache.spark.sql.functions._
 final class GBTClassifier @Since("1.4.0") (
     @Since("1.4.0") override val uid: String)
   extends Predictor[Vector, GBTClassifier, GBTClassificationModel]
-  with GBTParams with TreeClassifierParams with Logging {
+  with GBTClassifierParams with DefaultParamsWritable with Logging {
 
   @Since("1.4.0")
   def this() = this(Identifiable.randomUID("gbtc"))
@@ -115,40 +117,12 @@ final class GBTClassifier @Since("1.4.0") (
   @Since("1.4.0")
   override def setStepSize(value: Double): this.type = super.setStepSize(value)
 
-  // Parameters for GBTClassifier:
-
-  /**
-   * Loss function which GBT tries to minimize. (case-insensitive)
-   * Supported: "logistic"
-   * (default = logistic)
-   * @group param
-   */
-  @Since("1.4.0")
-  val lossType: Param[String] = new Param[String](this, "lossType", "Loss function which GBT" +
-    " tries to minimize (case-insensitive). Supported options:" +
-    s" ${GBTClassifier.supportedLossTypes.mkString(", ")}",
-    (value: String) => GBTClassifier.supportedLossTypes.contains(value.toLowerCase))
-
-  setDefault(lossType -> "logistic")
+  // Parameters from GBTClassifierParams:
 
   /** @group setParam */
   @Since("1.4.0")
   def setLossType(value: String): this.type = set(lossType, value)
 
-  /** @group getParam */
-  @Since("1.4.0")
-  def getLossType: String = $(lossType).toLowerCase
-
-  /** (private[ml]) Convert new loss to old loss. */
-  override private[ml] def getOldLossType: OldLoss = {
-    getLossType match {
-      case "logistic" => OldLogLoss
-      case _ =>
-        // Should never happen because of check in setter method.
-        throw new RuntimeException(s"GBTClassifier was given bad loss type: $getLossType")
-    }
-  }
-
   override protected def train(dataset: Dataset[_]): GBTClassificationModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
@@ -175,11 +149,14 @@ final class GBTClassifier @Since("1.4.0") (
 
 @Since("1.4.0")
 @Experimental
-object GBTClassifier {
-  // The losses below should be lowercase.
+object GBTClassifier extends DefaultParamsReadable[GBTClassifier] {
+
   /** Accessor for supported loss settings: logistic */
   @Since("1.4.0")
-  final val supportedLossTypes: Array[String] = Array("logistic").map(_.toLowerCase)
+  final val supportedLossTypes: Array[String] = GBTClassifierParams.supportedLossTypes
+
+  @Since("2.0.0")
+  override def load(path: String): GBTClassifier = super.load(path)
 }
 
 /**
@@ -199,7 +176,8 @@ final class GBTClassificationModel private[ml](
     private val _treeWeights: Array[Double],
     @Since("1.6.0") override val numFeatures: Int)
   extends PredictionModel[Vector, GBTClassificationModel]
-  with TreeEnsembleModel[DecisionTreeRegressionModel] with Serializable {
+  with GBTClassifierParams with TreeEnsembleModel[DecisionTreeRegressionModel]
+  with MLWritable with Serializable {
 
   require(_trees.nonEmpty, "GBTClassificationModel requires at least 1 tree.")
   require(_trees.length == _treeWeights.length, "GBTClassificationModel given trees, treeWeights" +
@@ -267,12 +245,62 @@ final class GBTClassificationModel private[ml](
   private[ml] def toOld: OldGBTModel = {
     new OldGBTModel(OldAlgo.Classification, _trees.map(_.toOld), _treeWeights)
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new GBTClassificationModel.GBTClassificationModelWriter(this)
 }
 
-private[ml] object GBTClassificationModel {
+@Since("2.0.0")
+object GBTClassificationModel extends MLReadable[GBTClassificationModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[GBTClassificationModel] = new GBTClassificationModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): GBTClassificationModel = super.load(path)
+
+  private[GBTClassificationModel]
+  class GBTClassificationModelWriter(instance: GBTClassificationModel) extends MLWriter {
+
+    override protected def saveImpl(path: String): Unit = {
+
+      val extraMetadata: JObject = Map(
+        "numFeatures" -> instance.numFeatures,
+        "numTrees" -> instance.getNumTrees)
+      EnsembleModelReadWrite.saveImpl(instance, path, sqlContext, extraMetadata)
+    }
+  }
+
+  private class GBTClassificationModelReader extends MLReader[GBTClassificationModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[GBTClassificationModel].getName
+    private val treeClassName = classOf[DecisionTreeRegressionModel].getName
+
+    override def load(path: String): GBTClassificationModel = {
+      implicit val format = DefaultFormats
+      val (metadata: Metadata, treesData: Array[(Metadata, Node)], treeWeights: Array[Double]) =
+        EnsembleModelReadWrite.loadImpl(path, sqlContext, className, treeClassName)
+      val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
+      val numTrees = (metadata.metadata \ "numTrees").extract[Int]
+
+      val trees: Array[DecisionTreeRegressionModel] = treesData.map {
+        case (treeMetadata, root) =>
+          val tree =
+            new DecisionTreeRegressionModel(treeMetadata.uid, root, numFeatures)
+          DefaultParamsReader.getAndSetParams(tree, treeMetadata)
+          tree
+      }
+      require(numTrees == trees.length, s"GBTClassificationModel.load expected $numTrees" +
+        s" trees based on metadata but found ${trees.length} trees.")
+      val model = new GBTClassificationModel(metadata.uid, trees, treeWeights, numFeatures)
+      DefaultParamsReader.getAndSetParams(model, metadata)
+      model
+    }
+  }
 
-  /** (private[ml]) Convert a model from the old API */
-  def fromOld(
+  /** Convert a model from the old API */
+  private[ml] def fromOld(
       oldModel: OldGBTModel,
       parent: GBTClassifier,
       categoricalFeatures: Map[Int, Int],
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
index 9d80b8eb68..dfa711b243 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
@@ -294,7 +294,7 @@ object RandomForestClassificationModel extends MLReadable[RandomForestClassifica
 
     override def load(path: String): RandomForestClassificationModel = {
       implicit val format = DefaultFormats
-      val (metadata: Metadata, treesData: Array[(Metadata, Node)]) =
+      val (metadata: Metadata, treesData: Array[(Metadata, Node)], _) =
         EnsembleModelReadWrite.loadImpl(path, sqlContext, className, treeClassName)
       val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
       val numClasses = (metadata.metadata \ "numClasses").extract[Int]
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
index 0b52fe2d13..741724d7a1 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -18,19 +18,20 @@
 package org.apache.spark.ml.regression
 
 import com.github.fommil.netlib.BLAS.{getInstance => blas}
+import org.json4s.{DefaultFormats, JObject}
+import org.json4s.JsonDSL._
 
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.internal.Logging
 import org.apache.spark.ml.{PredictionModel, Predictor}
-import org.apache.spark.ml.param.{Param, ParamMap}
-import org.apache.spark.ml.tree.{GBTParams, TreeEnsembleModel, TreeRegressorParams}
+import org.apache.spark.ml.param.ParamMap
+import org.apache.spark.ml.tree._
 import org.apache.spark.ml.tree.impl.GradientBoostedTrees
-import org.apache.spark.ml.util.{Identifiable, MetadataUtils}
+import org.apache.spark.ml.util._
+import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
-import org.apache.spark.mllib.tree.loss.{AbsoluteError => OldAbsoluteError, Loss => OldLoss,
-  SquaredError => OldSquaredError}
 import org.apache.spark.mllib.tree.model.{GradientBoostedTreesModel => OldGBTModel}
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Dataset}
@@ -58,7 +59,7 @@ import org.apache.spark.sql.functions._
 @Experimental
 final class GBTRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: String)
   extends Predictor[Vector, GBTRegressor, GBTRegressionModel]
-  with GBTParams with TreeRegressorParams with Logging {
+  with GBTRegressorParams with DefaultParamsWritable with Logging {
 
   @Since("1.4.0")
   def this() = this(Identifiable.randomUID("gbtr"))
@@ -112,41 +113,12 @@ final class GBTRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: Stri
   @Since("1.4.0")
   override def setStepSize(value: Double): this.type = super.setStepSize(value)
 
-  // Parameters for GBTRegressor:
-
-  /**
-   * Loss function which GBT tries to minimize. (case-insensitive)
-   * Supported: "squared" (L2) and "absolute" (L1)
-   * (default = squared)
-   * @group param
-   */
-  @Since("1.4.0")
-  val lossType: Param[String] = new Param[String](this, "lossType", "Loss function which GBT" +
-    " tries to minimize (case-insensitive). Supported options:" +
-    s" ${GBTRegressor.supportedLossTypes.mkString(", ")}",
-    (value: String) => GBTRegressor.supportedLossTypes.contains(value.toLowerCase))
-
-  setDefault(lossType -> "squared")
+  // Parameters from GBTRegressorParams:
 
   /** @group setParam */
   @Since("1.4.0")
   def setLossType(value: String): this.type = set(lossType, value)
 
-  /** @group getParam */
-  @Since("1.4.0")
-  def getLossType: String = $(lossType).toLowerCase
-
-  /** (private[ml]) Convert new loss to old loss. */
-  override private[ml] def getOldLossType: OldLoss = {
-    getLossType match {
-      case "squared" => OldSquaredError
-      case "absolute" => OldAbsoluteError
-      case _ =>
-        // Should never happen because of check in setter method.
-        throw new RuntimeException(s"GBTRegressorParams was given bad loss type: $getLossType")
-    }
-  }
-
   override protected def train(dataset: Dataset[_]): GBTRegressionModel = {
     val categoricalFeatures: Map[Int, Int] =
       MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
@@ -164,11 +136,14 @@ final class GBTRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: Stri
 
 @Since("1.4.0")
 @Experimental
-object GBTRegressor {
-  // The losses below should be lowercase.
+object GBTRegressor extends DefaultParamsReadable[GBTRegressor] {
+
   /** Accessor for supported loss settings: squared (L2), absolute (L1) */
   @Since("1.4.0")
-  final val supportedLossTypes: Array[String] = Array("squared", "absolute").map(_.toLowerCase)
+  final val supportedLossTypes: Array[String] = GBTRegressorParams.supportedLossTypes
+
+  @Since("2.0.0")
+  override def load(path: String): GBTRegressor = super.load(path)
 }
 
 /**
@@ -188,7 +163,8 @@ final class GBTRegressionModel private[ml](
     private val _treeWeights: Array[Double],
     override val numFeatures: Int)
   extends PredictionModel[Vector, GBTRegressionModel]
-  with TreeEnsembleModel[DecisionTreeRegressionModel] with Serializable {
+  with GBTRegressorParams with TreeEnsembleModel[DecisionTreeRegressionModel]
+  with MLWritable with Serializable {
 
   require(_trees.nonEmpty, "GBTRegressionModel requires at least 1 tree.")
   require(_trees.length == _treeWeights.length, "GBTRegressionModel given trees, treeWeights of" +
@@ -255,12 +231,64 @@ final class GBTRegressionModel private[ml](
   private[ml] def toOld: OldGBTModel = {
     new OldGBTModel(OldAlgo.Regression, _trees.map(_.toOld), _treeWeights)
   }
+
+  @Since("2.0.0")
+  override def write: MLWriter = new GBTRegressionModel.GBTRegressionModelWriter(this)
 }
 
-private[ml] object GBTRegressionModel {
+@Since("2.0.0")
+object GBTRegressionModel extends MLReadable[GBTRegressionModel] {
+
+  @Since("2.0.0")
+  override def read: MLReader[GBTRegressionModel] = new GBTRegressionModelReader
+
+  @Since("2.0.0")
+  override def load(path: String): GBTRegressionModel = super.load(path)
+
+  private[GBTRegressionModel]
+  class GBTRegressionModelWriter(instance: GBTRegressionModel) extends MLWriter {
+
+    override protected def saveImpl(path: String): Unit = {
+      val extraMetadata: JObject = Map(
+        "numFeatures" -> instance.numFeatures,
+        "numTrees" -> instance.getNumTrees)
+      EnsembleModelReadWrite.saveImpl(instance, path, sqlContext, extraMetadata)
+    }
+  }
+
+  private class GBTRegressionModelReader extends MLReader[GBTRegressionModel] {
+
+    /** Checked against metadata when loading model */
+    private val className = classOf[GBTRegressionModel].getName
+    private val treeClassName = classOf[DecisionTreeRegressionModel].getName
+
+    override def load(path: String): GBTRegressionModel = {
+      implicit val format = DefaultFormats
+      val (metadata: Metadata, treesData: Array[(Metadata, Node)], treeWeights: Array[Double]) =
+        EnsembleModelReadWrite.loadImpl(path, sqlContext, className, treeClassName)
+
+      val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
+      val numTrees = (metadata.metadata \ "numTrees").extract[Int]
+
+      val trees: Array[DecisionTreeRegressionModel] = treesData.map {
+        case (treeMetadata, root) =>
+          val tree =
+            new DecisionTreeRegressionModel(treeMetadata.uid, root, numFeatures)
+          DefaultParamsReader.getAndSetParams(tree, treeMetadata)
+          tree
+      }
+
+      require(numTrees == trees.length, s"GBTRegressionModel.load expected $numTrees" +
+        s" trees based on metadata but found ${trees.length} trees.")
+
+      val model = new GBTRegressionModel(metadata.uid, trees, treeWeights, numFeatures)
+      DefaultParamsReader.getAndSetParams(model, metadata)
+      model
+    }
+  }
 
-  /** (private[ml]) Convert a model from the old API */
-  def fromOld(
+  /** Convert a model from the old API */
+  private[ml] def fromOld(
       oldModel: OldGBTModel,
       parent: GBTRegressor,
       categoricalFeatures: Map[Int, Int],
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
index bee13c2ebf..4c4ff278d4 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
@@ -249,7 +249,7 @@ object RandomForestRegressionModel extends MLReadable[RandomForestRegressionMode
 
     override def load(path: String): RandomForestRegressionModel = {
       implicit val format = DefaultFormats
-      val (metadata: Metadata, treesData: Array[(Metadata, Node)]) =
+      val (metadata: Metadata, treesData: Array[(Metadata, Node)], treeWeights: Array[Double]) =
         EnsembleModelReadWrite.loadImpl(path, sqlContext, className, treeClassName)
       val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
       val numTrees = (metadata.metadata \ "numTrees").extract[Int]
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
index c4ab673d9a..f38e1ec7c0 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala
@@ -396,12 +396,14 @@ private[ml] object EnsembleModelReadWrite {
       sql: SQLContext,
       extraMetadata: JObject): Unit = {
     DefaultParamsWriter.saveMetadata(instance, path, sql.sparkContext, Some(extraMetadata))
-    val treesMetadataJson: Array[(Int, String)] = instance.trees.zipWithIndex.map {
+    val treesMetadataWeights: Array[(Int, String, Double)] = instance.trees.zipWithIndex.map {
       case (tree, treeID) =>
-        treeID -> DefaultParamsWriter.getMetadataToSave(tree.asInstanceOf[Params], sql.sparkContext)
+        (treeID,
+          DefaultParamsWriter.getMetadataToSave(tree.asInstanceOf[Params], sql.sparkContext),
+          instance.treeWeights(treeID))
     }
     val treesMetadataPath = new Path(path, "treesMetadata").toString
-    sql.createDataFrame(treesMetadataJson).toDF("treeID", "metadata")
+    sql.createDataFrame(treesMetadataWeights).toDF("treeID", "metadata", "weights")
       .write.parquet(treesMetadataPath)
     val dataPath = new Path(path, "data").toString
     val nodeDataRDD = sql.sparkContext.parallelize(instance.trees.zipWithIndex).flatMap {
@@ -424,7 +426,7 @@ private[ml] object EnsembleModelReadWrite {
       path: String,
       sql: SQLContext,
       className: String,
-      treeClassName: String): (Metadata, Array[(Metadata, Node)]) = {
+      treeClassName: String): (Metadata, Array[(Metadata, Node)], Array[Double]) = {
     import sql.implicits._
     implicit val format = DefaultFormats
     val metadata = DefaultParamsReader.loadMetadata(path, sql.sparkContext, className)
@@ -436,12 +438,15 @@ private[ml] object EnsembleModelReadWrite {
     }
 
     val treesMetadataPath = new Path(path, "treesMetadata").toString
-    val treesMetadataRDD: RDD[(Int, Metadata)] = sql.read.parquet(treesMetadataPath)
-      .select("treeID", "metadata").as[(Int, String)].rdd.map {
-      case (treeID: Int, json: String) =>
-        treeID -> DefaultParamsReader.parseMetadata(json, treeClassName)
+    val treesMetadataRDD: RDD[(Int, (Metadata, Double))] = sql.read.parquet(treesMetadataPath)
+      .select("treeID", "metadata", "weights").as[(Int, String, Double)].rdd.map {
+      case (treeID: Int, json: String, weights: Double) =>
+        treeID -> (DefaultParamsReader.parseMetadata(json, treeClassName), weights)
     }
-    val treesMetadata: Array[Metadata] = treesMetadataRDD.sortByKey().values.collect()
+
+    val treesMetadataWeights = treesMetadataRDD.sortByKey().values.collect()
+    val treesMetadata = treesMetadataWeights.map(_._1)
+    val treesWeights = treesMetadataWeights.map(_._2)
 
     val dataPath = new Path(path, "data").toString
     val nodeData: Dataset[EnsembleNodeData] =
@@ -452,7 +457,7 @@ private[ml] object EnsembleModelReadWrite {
           treeID -> DecisionTreeModelReadWrite.buildTreeFromNodes(nodeData.toArray, impurityType)
       }
     val rootNodes: Array[Node] = rootNodesRDD.sortByKey().values.collect()
-    (metadata, treesMetadata.zip(rootNodes))
+    (metadata, treesMetadata.zip(rootNodes), treesWeights)
   }
 
   /**
diff --git a/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala b/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
index 0767dc17e5..b6783911ad 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala
@@ -23,7 +23,7 @@ import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util.SchemaUtils
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, BoostingStrategy => OldBoostingStrategy, Strategy => OldStrategy}
 import org.apache.spark.mllib.tree.impurity.{Entropy => OldEntropy, Gini => OldGini, Impurity => OldImpurity, Variance => OldVariance}
-import org.apache.spark.mllib.tree.loss.{Loss => OldLoss}
+import org.apache.spark.mllib.tree.loss.{AbsoluteError => OldAbsoluteError, LogLoss => OldLogLoss, Loss => OldLoss, SquaredError => OldSquaredError}
 import org.apache.spark.sql.types.{DataType, DoubleType, StructType}
 
 /**
@@ -462,3 +462,74 @@ private[ml] trait GBTParams extends TreeEnsembleParams with HasMaxIter with HasS
   /** Get old Gradient Boosting Loss type */
   private[ml] def getOldLossType: OldLoss
 }
+
+private[ml] object GBTClassifierParams {
+  // The losses below should be lowercase.
+  /** Accessor for supported loss settings: logistic */
+  final val supportedLossTypes: Array[String] = Array("logistic").map(_.toLowerCase)
+}
+
+private[ml] trait GBTClassifierParams extends GBTParams with TreeClassifierParams {
+
+  /**
+   * Loss function which GBT tries to minimize. (case-insensitive)
+   * Supported: "logistic"
+   * (default = logistic)
+   * @group param
+   */
+  val lossType: Param[String] = new Param[String](this, "lossType", "Loss function which GBT" +
+    " tries to minimize (case-insensitive). Supported options:" +
+    s" ${GBTClassifierParams.supportedLossTypes.mkString(", ")}",
+    (value: String) => GBTClassifierParams.supportedLossTypes.contains(value.toLowerCase))
+
+  setDefault(lossType -> "logistic")
+
+  /** @group getParam */
+  def getLossType: String = $(lossType).toLowerCase
+
+  /** (private[ml]) Convert new loss to old loss. */
+  override private[ml] def getOldLossType: OldLoss = {
+    getLossType match {
+      case "logistic" => OldLogLoss
+      case _ =>
+        // Should never happen because of check in setter method.
+        throw new RuntimeException(s"GBTClassifier was given bad loss type: $getLossType")
+    }
+  }
+}
+
+private[ml] object GBTRegressorParams {
+  // The losses below should be lowercase.
+  /** Accessor for supported loss settings: squared (L2), absolute (L1) */
+  final val supportedLossTypes: Array[String] = Array("squared", "absolute").map(_.toLowerCase)
+}
+
+private[ml] trait GBTRegressorParams extends GBTParams with TreeRegressorParams {
+
+  /**
+   * Loss function which GBT tries to minimize. (case-insensitive)
+   * Supported: "squared" (L2) and "absolute" (L1)
+   * (default = squared)
+   * @group param
+   */
+  val lossType: Param[String] = new Param[String](this, "lossType", "Loss function which GBT" +
+    " tries to minimize (case-insensitive). Supported options:" +
+    s" ${GBTRegressorParams.supportedLossTypes.mkString(", ")}",
+    (value: String) => GBTRegressorParams.supportedLossTypes.contains(value.toLowerCase))
+
+  setDefault(lossType -> "squared")
+
+  /** @group getParam */
+  def getLossType: String = $(lossType).toLowerCase
+
+  /** (private[ml]) Convert new loss to old loss. */
+  override private[ml] def getOldLossType: OldLoss = {
+    getLossType match {
+      case "squared" => OldSquaredError
+      case "absolute" => OldAbsoluteError
+      case _ =>
+        // Should never happen because of check in setter method.
+        throw new RuntimeException(s"GBTRegressorParams was given bad loss type: $getLossType")
+    }
+  }
+}
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
index 76d8c9372e..7e6aec6b1b 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
@@ -22,7 +22,7 @@ import org.apache.spark.ml.param.ParamsSuite
 import org.apache.spark.ml.regression.DecisionTreeRegressionModel
 import org.apache.spark.ml.tree.LeafNode
 import org.apache.spark.ml.tree.impl.TreeTests
-import org.apache.spark.ml.util.MLTestingUtils
+import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.{EnsembleTestHelper, GradientBoostedTrees => OldGBT}
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
@@ -34,7 +34,8 @@ import org.apache.spark.util.Utils
 /**
  * Test suite for [[GBTClassifier]].
  */
-class GBTClassifierSuite extends SparkFunSuite with MLlibTestSparkContext {
+class GBTClassifierSuite extends SparkFunSuite with MLlibTestSparkContext
+  with DefaultReadWriteTest {
 
   import GBTClassifierSuite.compareAPIs
 
@@ -156,27 +157,23 @@ class GBTClassifierSuite extends SparkFunSuite with MLlibTestSparkContext {
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
 
-  // TODO: Reinstate test once save/load are implemented  SPARK-6725
-  /*
   test("model save/load") {
-    val tempDir = Utils.createTempDir()
-    val path = tempDir.toURI.toString
-
-    val trees = Range(0, 3).map(_ => OldDecisionTreeSuite.createModel(OldAlgo.Regression)).toArray
-    val treeWeights = Array(0.1, 0.3, 1.1)
-    val oldModel = new OldGBTModel(OldAlgo.Classification, trees, treeWeights)
-    val newModel = GBTClassificationModel.fromOld(oldModel)
-
-    // Save model, load it back, and compare.
-    try {
-      newModel.save(sc, path)
-      val sameNewModel = GBTClassificationModel.load(sc, path)
-      TreeTests.checkEqual(newModel, sameNewModel)
-    } finally {
-      Utils.deleteRecursively(tempDir)
+    def checkModelData(
+        model: GBTClassificationModel,
+        model2: GBTClassificationModel): Unit = {
+      TreeTests.checkEqual(model, model2)
+      assert(model.numFeatures === model2.numFeatures)
     }
+
+    val gbt = new GBTClassifier()
+    val rdd = TreeTests.getTreeReadWriteData(sc)
+
+    val allParamSettings = TreeTests.allParamSettings ++ Map("lossType" -> "logistic")
+
+    val continuousData: DataFrame =
+      TreeTests.setMetadata(rdd, Map.empty[Int, Int], numClasses = 2)
+    testEstimatorAndModelReadWrite(gbt, continuousData, allParamSettings, checkModelData)
   }
-  */
 }
 
 private object GBTClassifierSuite extends SparkFunSuite {
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
index 3c11631f98..216377959e 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GBTRegressorSuite.scala
@@ -19,7 +19,7 @@ package org.apache.spark.ml.regression
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.tree.impl.TreeTests
-import org.apache.spark.ml.util.MLTestingUtils
+import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.{EnsembleTestHelper, GradientBoostedTrees => OldGBT}
@@ -32,7 +32,8 @@ import org.apache.spark.util.Utils
 /**
  * Test suite for [[GBTRegressor]].
  */
-class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
+class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext
+  with DefaultReadWriteTest {
 
   import GBTRegressorSuite.compareAPIs
 
@@ -164,27 +165,22 @@ class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
   // Tests of model save/load
   /////////////////////////////////////////////////////////////////////////////
 
-  // TODO: Reinstate test once save/load are implemented  SPARK-6725
-  /*
   test("model save/load") {
-    val tempDir = Utils.createTempDir()
-    val path = tempDir.toURI.toString
-
-    val trees = Range(0, 3).map(_ => OldDecisionTreeSuite.createModel(OldAlgo.Regression)).toArray
-    val treeWeights = Array(0.1, 0.3, 1.1)
-    val oldModel = new OldGBTModel(OldAlgo.Regression, trees, treeWeights)
-    val newModel = GBTRegressionModel.fromOld(oldModel)
-
-    // Save model, load it back, and compare.
-    try {
-      newModel.save(sc, path)
-      val sameNewModel = GBTRegressionModel.load(sc, path)
-      TreeTests.checkEqual(newModel, sameNewModel)
-    } finally {
-      Utils.deleteRecursively(tempDir)
+    def checkModelData(
+        model: GBTRegressionModel,
+        model2: GBTRegressionModel): Unit = {
+      TreeTests.checkEqual(model, model2)
+      assert(model.numFeatures === model2.numFeatures)
     }
+
+    val gbt = new GBTRegressor()
+    val rdd = TreeTests.getTreeReadWriteData(sc)
+
+    val allParamSettings = TreeTests.allParamSettings ++ Map("lossType" -> "squared")
+    val continuousData: DataFrame =
+      TreeTests.setMetadata(rdd, Map.empty[Int, Int], numClasses = 0)
+    testEstimatorAndModelReadWrite(gbt, continuousData, allParamSettings, checkModelData)
   }
-  */
 }
 
 private object GBTRegressorSuite extends SparkFunSuite {
-- 
cgit v1.2.3


From b0adb9f543fbac16ea14c64eef6ba032a9919039 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Wed, 13 Apr 2016 13:18:02 -0700
Subject: [SPARK-10386][MLLIB] PrefixSpanModel supports save/load

```PrefixSpanModel``` supports ```save/load```. It's similar with #9267.

cc jkbradley

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #10664 from yanboliang/spark-10386.
---
 .../org/apache/spark/mllib/fpm/PrefixSpan.scala    | 96 +++++++++++++++++++++-
 .../spark/mllib/fpm/JavaPrefixSpanSuite.java       | 37 +++++++++
 .../apache/spark/mllib/fpm/PrefixSpanSuite.scala   | 31 +++++++
 3 files changed, 163 insertions(+), 1 deletion(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/fpm/PrefixSpan.scala b/mllib/src/main/scala/org/apache/spark/mllib/fpm/PrefixSpan.scala
index 4455681e50..4344ab1bad 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/fpm/PrefixSpan.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/fpm/PrefixSpan.scala
@@ -23,12 +23,22 @@ import java.util.concurrent.atomic.AtomicInteger
 import scala.collection.JavaConverters._
 import scala.collection.mutable
 import scala.reflect.ClassTag
+import scala.reflect.runtime.universe._
 
+import org.json4s.DefaultFormats
+import org.json4s.JsonDSL._
+import org.json4s.jackson.JsonMethods.{compact, render}
+
+import org.apache.spark.SparkContext
 import org.apache.spark.annotation.{Experimental, Since}
 import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.api.java.JavaSparkContext.fakeClassTag
 import org.apache.spark.internal.Logging
+import org.apache.spark.mllib.util.{Loader, Saveable}
 import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.{DataFrame, Row, SQLContext}
+import org.apache.spark.sql.catalyst.ScalaReflection
+import org.apache.spark.sql.types._
 import org.apache.spark.storage.StorageLevel
 
 /**
@@ -566,4 +576,88 @@ object PrefixSpan extends Logging {
 @Since("1.5.0")
 class PrefixSpanModel[Item] @Since("1.5.0") (
     @Since("1.5.0") val freqSequences: RDD[PrefixSpan.FreqSequence[Item]])
-  extends Serializable
+  extends Saveable with Serializable {
+
+  /**
+   * Save this model to the given path.
+   * It only works for Item datatypes supported by DataFrames.
+   *
+   * This saves:
+   *  - human-readable (JSON) model metadata to path/metadata/
+   *  - Parquet formatted data to path/data/
+   *
+   * The model may be loaded using [[PrefixSpanModel.load]].
+   *
+   * @param sc  Spark context used to save model data.
+   * @param path  Path specifying the directory in which to save this model.
+   *              If the directory already exists, this method throws an exception.
+   */
+  @Since("2.0.0")
+  override def save(sc: SparkContext, path: String): Unit = {
+    PrefixSpanModel.SaveLoadV1_0.save(this, path)
+  }
+
+  override protected val formatVersion: String = "1.0"
+}
+
+@Since("2.0.0")
+object PrefixSpanModel extends Loader[PrefixSpanModel[_]] {
+
+  @Since("2.0.0")
+  override def load(sc: SparkContext, path: String): PrefixSpanModel[_] = {
+    PrefixSpanModel.SaveLoadV1_0.load(sc, path)
+  }
+
+  private[fpm] object SaveLoadV1_0 {
+
+    private val thisFormatVersion = "1.0"
+
+    private val thisClassName = "org.apache.spark.mllib.fpm.PrefixSpanModel"
+
+    def save(model: PrefixSpanModel[_], path: String): Unit = {
+      val sc = model.freqSequences.sparkContext
+      val sqlContext = SQLContext.getOrCreate(sc)
+
+      val metadata = compact(render(
+        ("class" -> thisClassName) ~ ("version" -> thisFormatVersion)))
+      sc.parallelize(Seq(metadata), 1).saveAsTextFile(Loader.metadataPath(path))
+
+      // Get the type of item class
+      val sample = model.freqSequences.first().sequence(0)(0)
+      val className = sample.getClass.getCanonicalName
+      val classSymbol = runtimeMirror(getClass.getClassLoader).staticClass(className)
+      val tpe = classSymbol.selfType
+
+      val itemType = ScalaReflection.schemaFor(tpe).dataType
+      val fields = Array(StructField("sequence", ArrayType(ArrayType(itemType))),
+        StructField("freq", LongType))
+      val schema = StructType(fields)
+      val rowDataRDD = model.freqSequences.map { x =>
+        Row(x.sequence, x.freq)
+      }
+      sqlContext.createDataFrame(rowDataRDD, schema).write.parquet(Loader.dataPath(path))
+    }
+
+    def load(sc: SparkContext, path: String): PrefixSpanModel[_] = {
+      implicit val formats = DefaultFormats
+      val sqlContext = SQLContext.getOrCreate(sc)
+
+      val (className, formatVersion, metadata) = Loader.loadMetadata(sc, path)
+      assert(className == thisClassName)
+      assert(formatVersion == thisFormatVersion)
+
+      val freqSequences = sqlContext.read.parquet(Loader.dataPath(path))
+      val sample = freqSequences.select("sequence").head().get(0)
+      loadImpl(freqSequences, sample)
+    }
+
+    def loadImpl[Item: ClassTag](freqSequences: DataFrame, sample: Item): PrefixSpanModel[Item] = {
+      val freqSequencesRDD = freqSequences.select("sequence", "freq").rdd.map { x =>
+        val sequence = x.getAs[Seq[Seq[Item]]](0).map(_.toArray).toArray
+        val freq = x.getLong(1)
+        new PrefixSpan.FreqSequence(sequence, freq)
+      }
+      new PrefixSpanModel(freqSequencesRDD)
+    }
+  }
+}
diff --git a/mllib/src/test/java/org/apache/spark/mllib/fpm/JavaPrefixSpanSuite.java b/mllib/src/test/java/org/apache/spark/mllib/fpm/JavaPrefixSpanSuite.java
index 34daf5fbde..8a67793abc 100644
--- a/mllib/src/test/java/org/apache/spark/mllib/fpm/JavaPrefixSpanSuite.java
+++ b/mllib/src/test/java/org/apache/spark/mllib/fpm/JavaPrefixSpanSuite.java
@@ -17,6 +17,7 @@
 
 package org.apache.spark.mllib.fpm;
 
+import java.io.File;
 import java.util.Arrays;
 import java.util.List;
 
@@ -28,6 +29,7 @@ import org.junit.Test;
 import org.apache.spark.api.java.JavaRDD;
 import org.apache.spark.api.java.JavaSparkContext;
 import org.apache.spark.mllib.fpm.PrefixSpan.FreqSequence;
+import org.apache.spark.util.Utils;
 
 public class JavaPrefixSpanSuite {
   private transient JavaSparkContext sc;
@@ -64,4 +66,39 @@ public class JavaPrefixSpanSuite {
       long freq = freqSeq.freq();
     }
   }
+
+  @Test
+  public void runPrefixSpanSaveLoad() {
+    JavaRDD<List<List<Integer>>> sequences = sc.parallelize(Arrays.asList(
+      Arrays.asList(Arrays.asList(1, 2), Arrays.asList(3)),
+      Arrays.asList(Arrays.asList(1), Arrays.asList(3, 2), Arrays.asList(1, 2)),
+      Arrays.asList(Arrays.asList(1, 2), Arrays.asList(5)),
+      Arrays.asList(Arrays.asList(6))
+    ), 2);
+    PrefixSpan prefixSpan = new PrefixSpan()
+      .setMinSupport(0.5)
+      .setMaxPatternLength(5);
+    PrefixSpanModel<Integer> model = prefixSpan.run(sequences);
+
+    File tempDir = Utils.createTempDir(
+      System.getProperty("java.io.tmpdir"), "JavaPrefixSpanSuite");
+    String outputPath = tempDir.getPath();
+
+    try {
+      model.save(sc.sc(), outputPath);
+      PrefixSpanModel newModel = PrefixSpanModel.load(sc.sc(), outputPath);
+      JavaRDD<FreqSequence<Integer>> freqSeqs = newModel.freqSequences().toJavaRDD();
+      List<FreqSequence<Integer>> localFreqSeqs = freqSeqs.collect();
+      Assert.assertEquals(5, localFreqSeqs.size());
+      // Check that each frequent sequence could be materialized.
+      for (PrefixSpan.FreqSequence<Integer> freqSeq: localFreqSeqs) {
+        List<List<Integer>> seq = freqSeq.javaSequence();
+        long freq = freqSeq.freq();
+      }
+    } finally {
+      Utils.deleteRecursively(tempDir);
+    }
+
+
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/fpm/PrefixSpanSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/fpm/PrefixSpanSuite.scala
index a83e543859..6d8c7b47d8 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/fpm/PrefixSpanSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/fpm/PrefixSpanSuite.scala
@@ -18,6 +18,7 @@ package org.apache.spark.mllib.fpm
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.util.Utils
 
 class PrefixSpanSuite extends SparkFunSuite with MLlibTestSparkContext {
 
@@ -357,6 +358,36 @@ class PrefixSpanSuite extends SparkFunSuite with MLlibTestSparkContext {
     compareResults(expected, model.freqSequences.collect())
   }
 
+  test("model save/load") {
+    val sequences = Seq(
+      Array(Array(1, 2), Array(3)),
+      Array(Array(1), Array(3, 2), Array(1, 2)),
+      Array(Array(1, 2), Array(5)),
+      Array(Array(6)))
+    val rdd = sc.parallelize(sequences, 2).cache()
+
+    val prefixSpan = new PrefixSpan()
+      .setMinSupport(0.5)
+      .setMaxPatternLength(5)
+    val model = prefixSpan.run(rdd)
+
+    val tempDir = Utils.createTempDir()
+    val path = tempDir.toURI.toString
+    try {
+      model.save(sc, path)
+      val newModel = PrefixSpanModel.load(sc, path)
+      val originalSet = model.freqSequences.collect().map { x =>
+        (x.sequence.map(_.toSet).toSeq, x.freq)
+      }.toSet
+      val newSet = newModel.freqSequences.collect().map { x =>
+        (x.sequence.map(_.toSet).toSeq, x.freq)
+      }.toSet
+      assert(originalSet === newSet)
+    } finally {
+      Utils.deleteRecursively(tempDir)
+    }
+  }
+
   private def compareResults[Item](
       expectedValue: Array[(Array[Array[Item]], Long)],
       actualValue: Array[PrefixSpan.FreqSequence[Item]]): Unit = {
-- 
cgit v1.2.3


From 0d17593b32c12c3e39575430aa85cf20e56fae6a Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Wed, 13 Apr 2016 13:20:29 -0700
Subject: [SPARK-14461][ML] GLM training summaries should provide solver

## What changes were proposed in this pull request?
GLM training summaries should provide solver.

## How was this patch tested?
Unit tests.

cc jkbradley

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12253 from yanboliang/spark-14461.
---
 .../spark/ml/regression/GeneralizedLinearRegression.scala      | 10 +++++++---
 .../spark/ml/regression/GeneralizedLinearRegressionSuite.scala |  4 ++++
 2 files changed, 11 insertions(+), 3 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
index 00cf25dc54..e92a3e7fa1 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/regression/GeneralizedLinearRegression.scala
@@ -237,7 +237,8 @@ class GeneralizedLinearRegression @Since("2.0.0") (@Since("2.0.0") override val
         predictionColName,
         model,
         wlsModel.diagInvAtWA.toArray,
-        1)
+        1,
+        getSolver)
       return model.setSummary(trainingSummary)
     }
 
@@ -257,7 +258,8 @@ class GeneralizedLinearRegression @Since("2.0.0") (@Since("2.0.0") override val
       predictionColName,
       model,
       irlsModel.diagInvAtWA.toArray,
-      irlsModel.numIterations)
+      irlsModel.numIterations,
+      getSolver)
 
     model.setSummary(trainingSummary)
   }
@@ -781,6 +783,7 @@ object GeneralizedLinearRegressionModel extends MLReadable[GeneralizedLinearRegr
  * @param model the model that should be summarized
  * @param diagInvAtWA diagonal of matrix (A^T * W * A)^-1 in the last iteration
  * @param numIterations number of iterations
+ * @param solver the solver algorithm used for model training
  */
 @Since("2.0.0")
 @Experimental
@@ -789,7 +792,8 @@ class GeneralizedLinearRegressionSummary private[regression] (
     @Since("2.0.0") val predictionCol: String,
     @Since("2.0.0") val model: GeneralizedLinearRegressionModel,
     private val diagInvAtWA: Array[Double],
-    @Since("2.0.0") val numIterations: Int) extends Serializable {
+    @Since("2.0.0") val numIterations: Int,
+    @Since("2.0.0") val solver: String) extends Serializable {
 
   import GeneralizedLinearRegression._
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
index 4905f3e068..3ecc210abd 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala
@@ -626,6 +626,7 @@ class GeneralizedLinearRegressionSuite
     assert(summary.residualDegreeOfFreedom === residualDegreeOfFreedomR)
     assert(summary.residualDegreeOfFreedomNull === residualDegreeOfFreedomNullR)
     assert(summary.aic ~== aicR absTol 1E-3)
+    assert(summary.solver === "irls")
   }
 
   test("glm summary: binomial family with weight") {
@@ -739,6 +740,7 @@ class GeneralizedLinearRegressionSuite
     assert(summary.residualDegreeOfFreedom === residualDegreeOfFreedomR)
     assert(summary.residualDegreeOfFreedomNull === residualDegreeOfFreedomNullR)
     assert(summary.aic ~== aicR absTol 1E-3)
+    assert(summary.solver === "irls")
   }
 
   test("glm summary: poisson family with weight") {
@@ -855,6 +857,7 @@ class GeneralizedLinearRegressionSuite
     assert(summary.residualDegreeOfFreedom === residualDegreeOfFreedomR)
     assert(summary.residualDegreeOfFreedomNull === residualDegreeOfFreedomNullR)
     assert(summary.aic ~== aicR absTol 1E-3)
+    assert(summary.solver === "irls")
   }
 
   test("glm summary: gamma family with weight") {
@@ -968,6 +971,7 @@ class GeneralizedLinearRegressionSuite
     assert(summary.residualDegreeOfFreedom === residualDegreeOfFreedomR)
     assert(summary.residualDegreeOfFreedomNull === residualDegreeOfFreedomNullR)
     assert(summary.aic ~== aicR absTol 1E-3)
+    assert(summary.solver === "irls")
   }
 
   test("read/write") {
-- 
cgit v1.2.3


From a91aaf5a8cca18811c0cccc20f4e77f36231b344 Mon Sep 17 00:00:00 2001
From: Yanbo Liang <ybliang8@gmail.com>
Date: Wed, 13 Apr 2016 13:23:10 -0700
Subject: [SPARK-14375][ML] Unit test for spark.ml KMeansSummary

## What changes were proposed in this pull request?
* Modify ```KMeansSummary.clusterSizes``` method to make it robust to empty clusters.
* Add unit test for spark.ml ```KMeansSummary```.
* Add Since tag.

## How was this patch tested?
unit tests.

cc jkbradley

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #12254 from yanboliang/spark-14375.
---
 .../org/apache/spark/ml/clustering/KMeans.scala    | 35 ++++++++++++++++++----
 .../org/apache/spark/ml/r/KMeansWrapper.scala      |  2 +-
 .../apache/spark/ml/clustering/KMeansSuite.scala   | 18 ++++++++++-
 3 files changed, 47 insertions(+), 8 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala b/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
index d716bc6887..b324196842 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/clustering/KMeans.scala
@@ -143,6 +143,12 @@ class KMeansModel private[ml] (
     this
   }
 
+  /**
+   * Return true if there exists summary of model.
+   */
+  @Since("2.0.0")
+  def hasSummary: Boolean = trainingSummary.nonEmpty
+
   /**
    * Gets summary of model on training set. An exception is
    * thrown if `trainingSummary == None`.
@@ -267,7 +273,8 @@ class KMeans @Since("1.5.0") (
       .setEpsilon($(tol))
     val parentModel = algo.run(rdd)
     val model = copyValues(new KMeansModel(uid, parentModel).setParent(this))
-    val summary = new KMeansSummary(model.transform(dataset), $(predictionCol), $(featuresCol))
+    val summary = new KMeansSummary(
+      model.transform(dataset), $(predictionCol), $(featuresCol), $(k))
     model.setSummary(summary)
   }
 
@@ -284,10 +291,22 @@ object KMeans extends DefaultParamsReadable[KMeans] {
   override def load(path: String): KMeans = super.load(path)
 }
 
+/**
+ * :: Experimental ::
+ * Summary of KMeans.
+ *
+ * @param predictions  [[DataFrame]] produced by [[KMeansModel.transform()]]
+ * @param predictionCol  Name for column of predicted clusters in `predictions`
+ * @param featuresCol  Name for column of features in `predictions`
+ * @param k  Number of clusters
+ */
+@Since("2.0.0")
+@Experimental
 class KMeansSummary private[clustering] (
     @Since("2.0.0") @transient val predictions: DataFrame,
     @Since("2.0.0") val predictionCol: String,
-    @Since("2.0.0") val featuresCol: String) extends Serializable {
+    @Since("2.0.0") val featuresCol: String,
+    @Since("2.0.0") val k: Int) extends Serializable {
 
   /**
    * Cluster centers of the transformed data.
@@ -296,11 +315,15 @@ class KMeansSummary private[clustering] (
   @transient lazy val cluster: DataFrame = predictions.select(predictionCol)
 
   /**
-   * Size of each cluster.
+   * Size of (number of data points in) each cluster.
    */
   @Since("2.0.0")
-  lazy val clusterSizes: Array[Int] = cluster.rdd.map {
-    case Row(clusterIdx: Int) => (clusterIdx, 1)
-  }.reduceByKey(_ + _).collect().sortBy(_._1).map(_._2)
+  lazy val clusterSizes: Array[Long] = {
+    val sizes = Array.fill[Long](k)(0)
+    cluster.groupBy(predictionCol).count().select(predictionCol, "count").collect().foreach {
+      case Row(cluster: Int, count: Long) => sizes(cluster) = count
+    }
+    sizes
+  }
 
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
index ee513579ce..9e2b81ee20 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/r/KMeansWrapper.scala
@@ -37,7 +37,7 @@ private[r] class KMeansWrapper private (
 
   lazy val k: Int = kMeansModel.getK
 
-  lazy val size: Array[Int] = kMeansModel.summary.clusterSizes
+  lazy val size: Array[Long] = kMeansModel.summary.clusterSizes
 
   lazy val cluster: DataFrame = kMeansModel.summary.cluster
 
diff --git a/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala
index 2076c745e2..2ca386e422 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/clustering/KMeansSuite.scala
@@ -82,7 +82,7 @@ class KMeansSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultR
     }
   }
 
-  test("fit & transform") {
+  test("fit, transform, and summary") {
     val predictionColName = "kmeans_prediction"
     val kmeans = new KMeans().setK(k).setPredictionCol(predictionColName).setSeed(1)
     val model = kmeans.fit(dataset)
@@ -99,6 +99,22 @@ class KMeansSuite extends SparkFunSuite with MLlibTestSparkContext with DefaultR
     assert(clusters === Set(0, 1, 2, 3, 4))
     assert(model.computeCost(dataset) < 0.1)
     assert(model.hasParent)
+
+    // Check validity of model summary
+    val numRows = dataset.count()
+    assert(model.hasSummary)
+    val summary: KMeansSummary = model.summary
+    assert(summary.predictionCol === predictionColName)
+    assert(summary.featuresCol === "features")
+    assert(summary.predictions.count() === numRows)
+    for (c <- Array(predictionColName, "features")) {
+      assert(summary.predictions.columns.contains(c))
+    }
+    assert(summary.cluster.columns === Array(predictionColName))
+    val clusterSizes = summary.clusterSizes
+    assert(clusterSizes.length === k)
+    assert(clusterSizes.sum === numRows)
+    assert(clusterSizes.forall(_ >= 0))
   }
 
   test("read/write") {
-- 
cgit v1.2.3


From bf65c87f706019d235d7093637341668a13b1be1 Mon Sep 17 00:00:00 2001
From: "Joseph K. Bradley" <joseph@databricks.com>
Date: Thu, 14 Apr 2016 12:44:59 -0700
Subject: [SPARK-14618][ML][DOC] Updated RegressionEvaluator.metricName param
 doc

## What changes were proposed in this pull request?

In Spark 1.4, we negated some metrics from RegressionEvaluator since CrossValidator always maximized metrics. This was fixed in 1.5, but the docs were not updated. This PR updates the docs.

## How was this patch tested?

no tests

Author: Joseph K. Bradley <joseph@databricks.com>

Closes #12377 from jkbradley/regeval-doc.
---
 .../org/apache/spark/ml/evaluation/RegressionEvaluator.scala     | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

(limited to 'mllib/src/main/scala/org')

diff --git a/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala b/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala
index 4134e2dbc5..ed04b67bcc 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala
@@ -39,11 +39,12 @@ final class RegressionEvaluator @Since("1.4.0") (@Since("1.4.0") override val ui
   def this() = this(Identifiable.randomUID("regEval"))
 
   /**
-   * param for metric name in evaluation (supports `"rmse"` (default), `"mse"`, `"r2"`, and `"mae"`)
+   * Param for metric name in evaluation. Supports:
+   *  - `"rmse"` (default): root mean squared error
+   *  - `"mse"`: mean squared error
+   *  - `"r2"`: R^2^ metric
+   *  - `"mae"`: mean absolute error
    *
-   * Because we will maximize evaluation value (ref: `CrossValidator`),
-   * when we evaluate a metric that is needed to minimize (e.g., `"rmse"`, `"mse"`, `"mae"`),
-   * we take and output the negative of this metric.
    * @group param
    */
   @Since("1.4.0")
-- 
cgit v1.2.3