From db56f2df1b8027171da1b8d2571d1f2ef1e103b6 Mon Sep 17 00:00:00 2001 From: Michael Giannakopoulos Date: Sun, 20 Jul 2014 20:48:44 -0700 Subject: [SPARK-1945][MLLIB] Documentation Improvements for Spark 1.0 Standalone application examples are added to 'mllib-linear-methods.md' file written in Java. This commit is related to the issue [Add full Java Examples in MLlib docs](https://issues.apache.org/jira/browse/SPARK-1945). Also I changed the name of the sigmoid function from 'logit' to 'f'. This is because the logit function is the inverse of sigmoid. Thanks, Michael Author: Michael Giannakopoulos Closes #1311 from miccagiann/master and squashes the following commits: 8ffe5ab [Michael Giannakopoulos] Update code so as to comply with code standards. f7ad5cc [Michael Giannakopoulos] Merge remote-tracking branch 'upstream/master' 38d92c7 [Michael Giannakopoulos] Adding PCA, SVD and LBFGS examples in Java. Performing minor updates in the already committed examples so as to eradicate the call of 'productElement' function whenever is possible. cc0a089 [Michael Giannakopoulos] Modyfied Java examples so as to comply with coding standards. b1141b2 [Michael Giannakopoulos] Added Java examples for Clustering and Collaborative Filtering [mllib-clustering.md & mllib-collaborative-filtering.md]. 837f7a8 [Michael Giannakopoulos] Merge remote-tracking branch 'upstream/master' 15f0eb4 [Michael Giannakopoulos] Java examples included in 'mllib-linear-methods.md' file. --- docs/mllib-optimization.md | 96 +++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 95 insertions(+), 1 deletion(-) (limited to 'docs/mllib-optimization.md') diff --git a/docs/mllib-optimization.md b/docs/mllib-optimization.md index 651958c781..26ce5f3c50 100644 --- a/docs/mllib-optimization.md +++ b/docs/mllib-optimization.md @@ -207,6 +207,10 @@ the loss computed for every iteration. Here is an example to train binary logistic regression with L2 regularization using L-BFGS optimizer. + +
+ +
{% highlight scala %} import org.apache.spark.SparkContext import org.apache.spark.mllib.evaluation.BinaryClassificationMetrics @@ -263,7 +267,97 @@ println("Loss of each step in training process") loss.foreach(println) println("Area under ROC = " + auROC) {% endhighlight %} - +
+ +
+{% highlight java %} +import java.util.Arrays; +import java.util.Random; + +import scala.Tuple2; + +import org.apache.spark.api.java.*; +import org.apache.spark.api.java.function.Function; +import org.apache.spark.mllib.classification.LogisticRegressionModel; +import org.apache.spark.mllib.evaluation.BinaryClassificationMetrics; +import org.apache.spark.mllib.linalg.Vector; +import org.apache.spark.mllib.linalg.Vectors; +import org.apache.spark.mllib.optimization.*; +import org.apache.spark.mllib.regression.LabeledPoint; +import org.apache.spark.mllib.util.MLUtils; +import org.apache.spark.SparkConf; +import org.apache.spark.SparkContext; + +public class LBFGSExample { + public static void main(String[] args) { + SparkConf conf = new SparkConf().setAppName("L-BFGS Example"); + SparkContext sc = new SparkContext(conf); + String path = "data/mllib/sample_libsvm_data.txt"; + JavaRDD data = MLUtils.loadLibSVMFile(sc, path).toJavaRDD(); + int numFeatures = data.take(1).get(0).features().size(); + + // Split initial RDD into two... [60% training data, 40% testing data]. + JavaRDD trainingInit = data.sample(false, 0.6, 11L); + JavaRDD test = data.subtract(trainingInit); + + // Append 1 into the training data as intercept. + JavaRDD> training = data.map( + new Function>() { + public Tuple2 call(LabeledPoint p) { + return new Tuple2(p.label(), MLUtils.appendBias(p.features())); + } + }); + training.cache(); + + // Run training algorithm to build the model. + int numCorrections = 10; + double convergenceTol = 1e-4; + int maxNumIterations = 20; + double regParam = 0.1; + Vector initialWeightsWithIntercept = Vectors.dense(new double[numFeatures + 1]); + + Tuple2 result = LBFGS.runLBFGS( + training.rdd(), + new LogisticGradient(), + new SquaredL2Updater(), + numCorrections, + convergenceTol, + maxNumIterations, + regParam, + initialWeightsWithIntercept); + Vector weightsWithIntercept = result._1(); + double[] loss = result._2(); + + final LogisticRegressionModel model = new LogisticRegressionModel( + Vectors.dense(Arrays.copyOf(weightsWithIntercept.toArray(), weightsWithIntercept.size() - 1)), + (weightsWithIntercept.toArray())[weightsWithIntercept.size() - 1]); + + // Clear the default threshold. + model.clearThreshold(); + + // Compute raw scores on the test set. + JavaRDD> scoreAndLabels = test.map( + new Function>() { + public Tuple2 call(LabeledPoint p) { + Double score = model.predict(p.features()); + return new Tuple2(score, p.label()); + } + }); + + // Get evaluation metrics. + BinaryClassificationMetrics metrics = + new BinaryClassificationMetrics(scoreAndLabels.rdd()); + double auROC = metrics.areaUnderROC(); + + System.out.println("Loss of each step in training process"); + for (double l : loss) + System.out.println(l); + System.out.println("Area under ROC = " + auROC); + } +} +{% endhighlight %} +
+
#### Developer's note Since the Hessian is constructed approximately from previous gradient evaluations, the objective function can not be changed during the optimization process. -- cgit v1.2.3