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 e9304b5e5c65074325fd27238f53c8768ef3b4fb..482dd4b272d1d31a6ef16a279e103a98138963ac 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 @@ -140,6 +140,7 @@ private class RandomForest ( logDebug("maxBins = " + metadata.maxBins) logDebug("featureSubsetStrategy = " + featureSubsetStrategy) logDebug("numFeaturesPerNode = " + metadata.numFeaturesPerNode) + logDebug("subsamplingRate = " + strategy.subsamplingRate) // Find the splits and the corresponding bins (interval between the splits) using a sample // of the input data. @@ -155,19 +156,12 @@ private class RandomForest ( // Cache input RDD for speedup during multiple passes. val treeInput = TreePoint.convertToTreeRDD(retaggedInput, bins, metadata) - val (subsample, withReplacement) = { - // TODO: Have a stricter check for RF in the strategy - val isRandomForest = numTrees > 1 - if (isRandomForest) { - (1.0, true) - } else { - (strategy.subsamplingRate, false) - } - } + val withReplacement = if (numTrees > 1) true else false val baggedInput - = BaggedPoint.convertToBaggedRDD(treeInput, subsample, numTrees, withReplacement, seed) - .persist(StorageLevel.MEMORY_AND_DISK) + = BaggedPoint.convertToBaggedRDD(treeInput, + strategy.subsamplingRate, numTrees, + withReplacement, seed).persist(StorageLevel.MEMORY_AND_DISK) // depth of the decision tree val maxDepth = strategy.maxDepth diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala index 972959885f39607af67696c9631fcbec6b0beb1b..3308adb6752ffd8b14a8af00846fdea47a62d3a5 100644 --- a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala +++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala @@ -156,6 +156,9 @@ class Strategy ( s"DecisionTree Strategy requires minInstancesPerNode >= 1 but was given $minInstancesPerNode") require(maxMemoryInMB <= 10240, s"DecisionTree Strategy requires maxMemoryInMB <= 10240, but was given $maxMemoryInMB") + require(subsamplingRate > 0 && subsamplingRate <= 1, + s"DecisionTree Strategy requires subsamplingRate <=1 and >0, but was given " + + s"$subsamplingRate") } /** Returns a shallow copy of this instance. */ diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala index f7f0f20c6c1251bace76bd1b7c5fe677d9e3bc7e..55e963977b54fd596777270d551f0d4fa724a340 100644 --- a/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala +++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala @@ -196,6 +196,22 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext { featureSubsetStrategy = "sqrt", seed = 12345) EnsembleTestHelper.validateClassifier(model, arr, 1.0) } + + test("subsampling rate in RandomForest"){ + val arr = EnsembleTestHelper.generateOrderedLabeledPoints(5, 20) + val rdd = sc.parallelize(arr) + val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 2, + numClasses = 2, categoricalFeaturesInfo = Map.empty[Int, Int], + useNodeIdCache = true) + + val rf1 = RandomForest.trainClassifier(rdd, strategy, numTrees = 3, + featureSubsetStrategy = "auto", seed = 123) + strategy.subsamplingRate = 0.5 + val rf2 = RandomForest.trainClassifier(rdd, strategy, numTrees = 3, + featureSubsetStrategy = "auto", seed = 123) + assert(rf1.toDebugString != rf2.toDebugString) + } + }