[SPARK-10574][ML][MLLIB] HashingTF supports MurmurHash3

## What changes were proposed in this pull request?
As the discussion at [SPARK-10574](https://issues.apache.org/jira/browse/SPARK-10574), ```HashingTF``` should support MurmurHash3 and make it as the default hash algorithm. We should also expose set/get API for ```hashAlgorithm```, then users can choose the hash method.

Note: The problem that ```mllib.feature.HashingTF``` behaves differently between Scala/Java and Python will be resolved in the followup work.

## How was this patch tested?
unit tests.

cc jkbradley MLnick

Author: Yanbo Liang <ybliang8@gmail.com>
Author: Joseph K. Bradley <joseph@databricks.com>

Closes #12498 from yanboliang/spark-10574.

5 files changed, 162 insertions, 30 deletions

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 467ad73074..6fc08aee13 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.{BooleanParam, IntParam, ParamMap, ParamValidators}
+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.feature
@@ -31,6 +31,12 @@ import org.apache.spark.sql.types.{ArrayType, StructType}
 /**
  * :: Experimental ::
  * Maps a sequence of terms to their term frequencies using the hashing trick.
+ * Currently we support two hash algorithms: "murmur3" (default) and "native".
+ * "murmur3" calculates a hash code value for the term object using
+ * Austin Appleby's MurmurHash 3 algorithm (MurmurHash3_x86_32);
+ * "native" calculates the hash code value using the native Scala implementation.
+ * In Spark 1.6 and earlier, "native" is the default hash algorithm;
+ * after Spark 2.0, we use "murmur3" as the default one.
  */
 @Experimental
 class HashingTF(override val uid: String)
@@ -63,7 +69,20 @@ class HashingTF(override val uid: String)
     "This is useful for discrete probabilistic models that model binary events rather " +
     "than integer counts")
 
-  setDefault(numFeatures -> (1 << 18), binary -> false)
+  /**
+   * The hash algorithm used when mapping term to integer.
+   * Supported options: "murmur3" and "native". We use "native" as default hash algorithm
+   * in Spark 1.6 and earlier. After Spark 2.0, we use "murmur3" as default one.
+   * (Default = "murmur3")
+   * @group expertParam
+   */
+  val hashAlgorithm = new Param[String](this, "hashAlgorithm", "The hash algorithm used when " +
+    "mapping term to integer. Supported options: " +
+    s"${feature.HashingTF.supportedHashAlgorithms.mkString(",")}.",
+    ParamValidators.inArray[String](feature.HashingTF.supportedHashAlgorithms))
+
+  setDefault(numFeatures -> (1 << 18), binary -> false,
+    hashAlgorithm -> feature.HashingTF.Murmur3)
 
   /** @group getParam */
   def getNumFeatures: Int = $(numFeatures)
@@ -77,10 +96,18 @@ class HashingTF(override val uid: String)
   /** @group setParam */
   def setBinary(value: Boolean): this.type = set(binary, value)
 
+  /** @group expertGetParam */
+  def getHashAlgorithm: String = $(hashAlgorithm)
+
+  /** @group expertSetParam */
+  def setHashAlgorithm(value: String): this.type = set(hashAlgorithm, value)
+
   @Since("2.0.0")
   override def transform(dataset: Dataset[_]): DataFrame = {
     val outputSchema = transformSchema(dataset.schema)
-    val hashingTF = new feature.HashingTF($(numFeatures)).setBinary($(binary))
+    val hashingTF = new feature.HashingTF($(numFeatures))
+      .setBinary($(binary))
+      .setHashAlgorithm($(hashAlgorithm))
     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 47c9e850a0..321f11d9f9 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
@@ -22,10 +22,13 @@ import java.lang.{Iterable => JavaIterable}
 import scala.collection.JavaConverters._
 import scala.collection.mutable
 
+import org.apache.spark.SparkException
 import org.apache.spark.annotation.Since
 import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.rdd.RDD
+import org.apache.spark.unsafe.hash.Murmur3_x86_32._
+import org.apache.spark.unsafe.types.UTF8String
 import org.apache.spark.util.Utils
 
 /**
@@ -36,7 +39,10 @@ import org.apache.spark.util.Utils
 @Since("1.1.0")
 class HashingTF(val numFeatures: Int) extends Serializable {
 
+  import HashingTF._
+
   private var binary = false
+  private var hashAlgorithm = HashingTF.Murmur3
 
   /**
    */
@@ -54,10 +60,34 @@ class HashingTF(val numFeatures: Int) extends Serializable {
   }
 
   /**
+   * Set the hash algorithm used when mapping term to integer.
+   * (default: murmur3)
+   */
+  @Since("2.0.0")
+  def setHashAlgorithm(value: String): this.type = {
+    hashAlgorithm = value
+    this
+  }
+
+  /**
    * Returns the index of the input term.
    */
   @Since("1.1.0")
-  def indexOf(term: Any): Int = Utils.nonNegativeMod(term.##, numFeatures)
+  def indexOf(term: Any): Int = {
+    Utils.nonNegativeMod(getHashFunction(term), numFeatures)
+  }
+
+  /**
+   * Get the hash function corresponding to the current [[hashAlgorithm]] setting.
+   */
+  private def getHashFunction: Any => Int = hashAlgorithm match {
+    case Murmur3 => murmur3Hash
+    case Native => nativeHash
+    case _ =>
+      // This should never happen.
+      throw new IllegalArgumentException(
+        s"HashingTF does not recognize hash algorithm $hashAlgorithm")
+  }
 
   /**
    * Transforms the input document into a sparse term frequency vector.
@@ -66,8 +96,9 @@ class HashingTF(val numFeatures: Int) extends Serializable {
   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
+    val hashFunc: Any => Int = getHashFunction
     document.foreach { term =>
-      val i = indexOf(term)
+      val i = Utils.nonNegativeMod(hashFunc(term), numFeatures)
       termFrequencies.put(i, setTF(i))
     }
     Vectors.sparse(numFeatures, termFrequencies.toSeq)
@@ -97,3 +128,41 @@ class HashingTF(val numFeatures: Int) extends Serializable {
     dataset.rdd.map(this.transform).toJavaRDD()
   }
 }
+
+object HashingTF {
+
+  private[spark] val Native: String = "native"
+
+  private[spark] val Murmur3: String = "murmur3"
+
+  private[spark] val supportedHashAlgorithms: Array[String] = Array(Native, Murmur3)
+
+  private val seed = 42
+
+  /**
+   * Calculate a hash code value for the term object using the native Scala implementation.
+   */
+  private[spark] def nativeHash(term: Any): Int = term.##
+
+  /**
+   * Calculate a hash code value for the term object using
+   * Austin Appleby's MurmurHash 3 algorithm (MurmurHash3_x86_32).
+   */
+  private[spark] def murmur3Hash(term: Any): Int = {
+    term match {
+      case null => seed
+      case b: Boolean => hashInt(if (b) 1 else 0, seed)
+      case b: Byte => hashInt(b, seed)
+      case s: Short => hashInt(s, seed)
+      case i: Int => hashInt(i, seed)
+      case l: Long => hashLong(l, seed)
+      case f: Float => hashInt(java.lang.Float.floatToIntBits(f), seed)
+      case d: Double => hashLong(java.lang.Double.doubleToLongBits(d), seed)
+      case s: String =>
+        val utf8 = UTF8String.fromString(s)
+        hashUnsafeBytes(utf8.getBaseObject, utf8.getBaseOffset, utf8.numBytes(), seed)
+      case _ => throw new SparkException("HashingTF with murmur3 algorithm does not " +
+        s"support type ${term.getClass.getCanonicalName} of input data.")
+    }
+  }
+}
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 addd733c20..e32b862af7 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
@@ -21,6 +21,7 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.attribute.AttributeGroup
 import org.apache.spark.ml.param.ParamsSuite
 import org.apache.spark.ml.util.DefaultReadWriteTest
+import org.apache.spark.mllib.feature.{HashingTF => MLlibHashingTF}
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
 import org.apache.spark.mllib.util.TestingUtils._
@@ -37,19 +38,26 @@ class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
       (0, "a a b b c d".split(" ").toSeq)
     )).toDF("id", "words")
     val n = 100
-    val hashingTF = new HashingTF()
-      .setInputCol("words")
-      .setOutputCol("features")
-      .setNumFeatures(n)
-    val output = hashingTF.transform(df)
-    val attrGroup = AttributeGroup.fromStructField(output.schema("features"))
-    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 => 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)
+    Seq("murmur3", "native").foreach { hashAlgorithm =>
+      val hashingTF = new HashingTF()
+        .setInputCol("words")
+        .setOutputCol("features")
+        .setNumFeatures(n)
+        .setHashAlgorithm(hashAlgorithm)
+      val output = hashingTF.transform(df)
+      val attrGroup = AttributeGroup.fromStructField(output.schema("features"))
+      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 => Int = if (hashAlgorithm == "murmur3") {
+        murmur3FeatureIdx(n)
+      } else {
+        nativeFeatureIdx(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") {
@@ -64,7 +72,7 @@ class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
         .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
+    def idx: Any => Int = murmur3FeatureIdx(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)
@@ -78,7 +86,11 @@ class HashingTFSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
     testDefaultReadWrite(t)
   }
 
-  private def featureIdx(numFeatures: Int)(term: Any): Int = {
-    Utils.nonNegativeMod(term.##, numFeatures)
+  private def nativeFeatureIdx(numFeatures: Int)(term: Any): Int = {
+    Utils.nonNegativeMod(MLlibHashingTF.nativeHash(term), numFeatures)
+  }
+
+  private def murmur3FeatureIdx(numFeatures: Int)(term: Any): Int = {
+    Utils.nonNegativeMod(MLlibHashingTF.murmur3Hash(term), numFeatures)
   }
 }
diff --git a/python/pyspark/ml/feature.py b/python/pyspark/ml/feature.py
index 1b298e639d..0e578d48ca 100644
--- a/python/pyspark/ml/feature.py
+++ b/python/pyspark/ml/feature.py
@@ -523,12 +523,12 @@ class HashingTF(JavaTransformer, HasInputCol, HasOutputCol, HasNumFeatures, Java
     >>> df = sqlContext.createDataFrame([(["a", "b", "c"],)], ["words"])
     >>> hashingTF = HashingTF(numFeatures=10, inputCol="words", outputCol="features")
     >>> hashingTF.transform(df).head().features
-    SparseVector(10, {7: 1.0, 8: 1.0, 9: 1.0})
+    SparseVector(10, {0: 1.0, 1: 1.0, 2: 1.0})
     >>> hashingTF.setParams(outputCol="freqs").transform(df).head().freqs
-    SparseVector(10, {7: 1.0, 8: 1.0, 9: 1.0})
+    SparseVector(10, {0: 1.0, 1: 1.0, 2: 1.0})
     >>> params = {hashingTF.numFeatures: 5, hashingTF.outputCol: "vector"}
     >>> hashingTF.transform(df, params).head().vector
-    SparseVector(5, {2: 1.0, 3: 1.0, 4: 1.0})
+    SparseVector(5, {0: 1.0, 1: 1.0, 2: 1.0})
     >>> hashingTFPath = temp_path + "/hashing-tf"
     >>> hashingTF.save(hashingTFPath)
     >>> loadedHashingTF = HashingTF.load(hashingTFPath)
@@ -543,22 +543,30 @@ class HashingTF(JavaTransformer, HasInputCol, HasOutputCol, HasNumFeatures, Java
                    "rather than integer counts. Default False.",
                    typeConverter=TypeConverters.toBoolean)
 
+    hashAlgorithm = Param(Params._dummy(), "hashAlgorithm", "The hash algorithm used when " +
+                          "mapping term to integer. Supported options: murmur3(default) " +
+                          "and native.", typeConverter=TypeConverters.toString)
+
     @keyword_only
-    def __init__(self, numFeatures=1 << 18, binary=False, inputCol=None, outputCol=None):
+    def __init__(self, numFeatures=1 << 18, binary=False, inputCol=None, outputCol=None,
+                 hashAlgorithm="murmur3"):
         """
-        __init__(self, numFeatures=1 << 18, inputCol=None, outputCol=None)
+        __init__(self, numFeatures=1 << 18, binary=False, inputCol=None, outputCol=None, \
+                 hashAlgorithm="murmur3")
         """
         super(HashingTF, self).__init__()
         self._java_obj = self._new_java_obj("org.apache.spark.ml.feature.HashingTF", self.uid)
-        self._setDefault(numFeatures=1 << 18, binary=False)
+        self._setDefault(numFeatures=1 << 18, binary=False, hashAlgorithm="murmur3")
         kwargs = self.__init__._input_kwargs
         self.setParams(**kwargs)
 
     @keyword_only
     @since("1.3.0")
-    def setParams(self, numFeatures=1 << 18, inputCol=None, outputCol=None):
+    def setParams(self, numFeatures=1 << 18, binary=False, inputCol=None, outputCol=None,
+                  hashAlgorithm="murmur3"):
         """
-        setParams(self, numFeatures=1 << 18, inputCol=None, outputCol=None)
+        setParams(self, numFeatures=1 << 18, binary=False, inputCol=None, outputCol=None, \
+                  hashAlgorithm="murmur3")
         Sets params for this HashingTF.
         """
         kwargs = self.setParams._input_kwargs
@@ -579,6 +587,21 @@ class HashingTF(JavaTransformer, HasInputCol, HasOutputCol, HasNumFeatures, Java
         """
         return self.getOrDefault(self.binary)
 
+    @since("2.0.0")
+    def setHashAlgorithm(self, value):
+        """
+        Sets the value of :py:attr:`hashAlgorithm`.
+        """
+        self._set(hashAlgorithm=value)
+        return self
+
+    @since("2.0.0")
+    def getHashAlgorithm(self):
+        """
+        Gets the value of hashAlgorithm or its default value.
+        """
+        return self.getOrDefault(self.hashAlgorithm)
+
 
 @inherit_doc
 class IDF(JavaEstimator, HasInputCol, HasOutputCol, JavaMLReadable, JavaMLWritable):
diff --git a/python/pyspark/ml/tests.py b/python/pyspark/ml/tests.py
index e95458699d..8954e96df9 100644
--- a/python/pyspark/ml/tests.py
+++ b/python/pyspark/ml/tests.py
@@ -918,7 +918,8 @@ class HashingTFTest(PySparkTestCase):
         df = sqlContext.createDataFrame([(0, ["a", "a", "b", "c", "c", "c"])], ["id", "words"])
         n = 100
         hashingTF = HashingTF()
-        hashingTF.setInputCol("words").setOutputCol("features").setNumFeatures(n).setBinary(True)
+        hashingTF.setInputCol("words").setOutputCol("features").setNumFeatures(n)\
+            .setBinary(True).setHashAlgorithm("native")
         output = hashingTF.transform(df)
         features = output.select("features").first().features.toArray()
         expected = Vectors.sparse(n, {(ord("a") % n): 1.0,