You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@spark.apache.org by me...@apache.org on 2014/09/16 02:39:36 UTC
git commit: [SPARK-3396][MLLIB] Use SquaredL2Updater in
LogisticRegressionWithSGD
Repository: spark
Updated Branches:
refs/heads/master 37d925280 -> 3b9312813
[SPARK-3396][MLLIB] Use SquaredL2Updater in LogisticRegressionWithSGD
SimpleUpdater ignores the regularizer, which leads to an unregularized
LogReg. To enable the common L2 regularizer (and the corresponding
regularization parameter) for logistic regression the SquaredL2Updater
has to be used in SGD (see, e.g., [SVMWithSGD])
Author: Christoph Sawade <ch...@sawade.me>
Closes #2398 from BigCrunsh/fix-regparam-logreg and squashes the following commits:
0820c04 [Christoph Sawade] Use SquaredL2Updater in LogisticRegressionWithSGD
Project: http://git-wip-us.apache.org/repos/asf/spark/repo
Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/3b931281
Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/3b931281
Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/3b931281
Branch: refs/heads/master
Commit: 3b93128139e8d303f1d7bfd04e9a99a11a5b6404
Parents: 37d9252
Author: Christoph Sawade <ch...@sawade.me>
Authored: Mon Sep 15 17:39:31 2014 -0700
Committer: Xiangrui Meng <me...@databricks.com>
Committed: Mon Sep 15 17:39:31 2014 -0700
----------------------------------------------------------------------
.../classification/LogisticRegression.scala | 2 +-
.../LogisticRegressionSuite.scala | 44 ++++++++++++++++++--
2 files changed, 42 insertions(+), 4 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/spark/blob/3b931281/mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala
----------------------------------------------------------------------
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 486bdbf..84d3c7c 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
@@ -84,7 +84,7 @@ class LogisticRegressionWithSGD private (
extends GeneralizedLinearAlgorithm[LogisticRegressionModel] with Serializable {
private val gradient = new LogisticGradient()
- private val updater = new SimpleUpdater()
+ private val updater = new SquaredL2Updater()
override val optimizer = new GradientDescent(gradient, updater)
.setStepSize(stepSize)
.setNumIterations(numIterations)
http://git-wip-us.apache.org/repos/asf/spark/blob/3b931281/mllib/src/test/scala/org/apache/spark/mllib/classification/LogisticRegressionSuite.scala
----------------------------------------------------------------------
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/classification/LogisticRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/classification/LogisticRegressionSuite.scala
index 8621786..e954baa 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/classification/LogisticRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/classification/LogisticRegressionSuite.scala
@@ -43,7 +43,7 @@ object LogisticRegressionSuite {
offset: Double,
scale: Double,
nPoints: Int,
- seed: Int): Seq[LabeledPoint] = {
+ seed: Int): Seq[LabeledPoint] = {
val rnd = new Random(seed)
val x1 = Array.fill[Double](nPoints)(rnd.nextGaussian())
@@ -58,12 +58,15 @@ object LogisticRegressionSuite {
}
class LogisticRegressionSuite extends FunSuite with LocalSparkContext with Matchers {
- def validatePrediction(predictions: Seq[Double], input: Seq[LabeledPoint]) {
+ def validatePrediction(
+ predictions: Seq[Double],
+ input: Seq[LabeledPoint],
+ expectedAcc: Double = 0.83) {
val numOffPredictions = predictions.zip(input).count { case (prediction, expected) =>
prediction != expected.label
}
// At least 83% of the predictions should be on.
- ((input.length - numOffPredictions).toDouble / input.length) should be > 0.83
+ ((input.length - numOffPredictions).toDouble / input.length) should be > expectedAcc
}
// Test if we can correctly learn A, B where Y = logistic(A + B*X)
@@ -155,6 +158,41 @@ class LogisticRegressionSuite extends FunSuite with LocalSparkContext with Match
validatePrediction(validationData.map(row => model.predict(row.features)), validationData)
}
+ test("logistic regression with initial weights and non-default regularization parameter") {
+ val nPoints = 10000
+ val A = 2.0
+ val B = -1.5
+
+ val testData = LogisticRegressionSuite.generateLogisticInput(A, B, nPoints, 42)
+
+ val initialB = -1.0
+ val initialWeights = Vectors.dense(initialB)
+
+ val testRDD = sc.parallelize(testData, 2)
+ testRDD.cache()
+
+ // Use half as many iterations as the previous test.
+ val lr = new LogisticRegressionWithSGD().setIntercept(true)
+ lr.optimizer.
+ setStepSize(10.0).
+ setNumIterations(10).
+ setRegParam(1.0)
+
+ val model = lr.run(testRDD, initialWeights)
+
+ // Test the weights
+ assert(model.weights(0) ~== -430000.0 relTol 20000.0)
+ assert(model.intercept ~== 370000.0 relTol 20000.0)
+
+ val validationData = LogisticRegressionSuite.generateLogisticInput(A, B, nPoints, 17)
+ val validationRDD = sc.parallelize(validationData, 2)
+ // Test prediction on RDD.
+ validatePrediction(model.predict(validationRDD.map(_.features)).collect(), validationData, 0.8)
+
+ // Test prediction on Array.
+ validatePrediction(validationData.map(row => model.predict(row.features)), validationData, 0.8)
+ }
+
test("logistic regression with initial weights with LBFGS") {
val nPoints = 10000
val A = 2.0
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@spark.apache.org
For additional commands, e-mail: commits-help@spark.apache.org