Github user sethah commented on a diff in the pull request:
https://github.com/apache/spark/pull/16441#discussion_r95287318
--- Diff:
mllib/src/test/scala/org/apache/spark/ml/classification/GBTClassifierSuite.scala
---
@@ -66,10 +70,79 @@ class GBTClassifierSuite extends SparkFunSuite with
MLlibTestSparkContext
ParamsSuite.checkParams(new GBTClassifier)
val model = new GBTClassificationModel("gbtc",
Array(new DecisionTreeRegressionModel("dtr", new LeafNode(0.0, 0.0,
null), 1)),
- Array(1.0), 1)
+ Array(1.0), 1, 2)
ParamsSuite.checkParams(model)
}
+ test("GBTClassifier: Predictor, Classifier methods") {
+ val rawPredictionCol = "rawPrediction"
+ val predictionCol = "prediction"
+ val labelCol = "label"
+ val featuresCol = "features"
+ val probabilityCol = "probability"
+
+ val gbt = new GBTClassifier().setSeed(123)
+ val trainingDataset = trainData.toDF(labelCol, featuresCol)
+ val gbtModel = gbt.fit(trainingDataset)
+ assert(gbtModel.numClasses === 2)
+ val numFeatures =
trainingDataset.select(featuresCol).first().getAs[Vector](0).size
+ assert(gbtModel.numFeatures === numFeatures)
+
+ val blas = BLAS.getInstance()
+
+ val validationDataset = validationData.toDF(labelCol, featuresCol)
+ val results = gbtModel.transform(validationDataset)
+ // check that raw prediction is tree predictions dot tree weights
+ results.select(rawPredictionCol, featuresCol).collect().foreach {
+ case Row(raw: Vector, features: Vector) =>
+ assert(raw.size === 2)
+ val treePredictions =
gbtModel.trees.map(_.rootNode.predictImpl(features).prediction)
+ val prediction = blas.ddot(gbtModel.numTrees, treePredictions, 1,
gbtModel.treeWeights, 1)
+ assert(raw ~== Vectors.dense(-prediction, prediction) relTol eps)
+ }
+
+ // Compare rawPrediction with probability
+ results.select(rawPredictionCol, probabilityCol).collect().foreach {
+ case Row(raw: Vector, prob: Vector) =>
+ assert(raw.size === 2)
+ assert(prob.size === 2)
+ val prodFromRaw = raw.toDense.values.map(value => 1 / (1 +
math.exp(-2 * value)))
+ assert(prob(0) ~== prodFromRaw(0) relTol eps)
+ assert(prob(1) ~== prodFromRaw(1) relTol eps)
+ }
+
+ // Compare prediction with probability
+ results.select(predictionCol, probabilityCol).collect().foreach {
+ case Row(pred: Double, prob: Vector) =>
+ val predFromProb = prob.toArray.zipWithIndex.maxBy(_._1)._2
+ assert(pred == predFromProb)
+ }
+
+ // force it to use raw2prediction
+ gbtModel.setRawPredictionCol(rawPredictionCol).setProbabilityCol("")
+ val resultsUsingRaw2Predict =
+
gbtModel.transform(validationDataset).select(predictionCol).as[Double].collect()
+
resultsUsingRaw2Predict.zip(results.select(predictionCol).as[Double].collect()).foreach
{
+ case (pred1, pred2) => assert(pred1 === pred2)
+ }
+
+ // force it to use probability2prediction
+ gbtModel.setRawPredictionCol("").setProbabilityCol(probabilityCol)
+ val resultsUsingProb2Predict =
+
gbtModel.transform(validationDataset).select(predictionCol).as[Double].collect()
+
resultsUsingProb2Predict.zip(results.select(predictionCol).as[Double].collect()).foreach
{
+ case (pred1, pred2) => assert(pred1 === pred2)
+ }
+
+ // force it to use predict
+ gbtModel.setRawPredictionCol("").setProbabilityCol("")
+ val resultsUsingPredict =
+
gbtModel.transform(validationDataset).select(predictionCol).as[Double].collect()
+
resultsUsingPredict.zip(results.select(predictionCol).as[Double].collect()).foreach
{
+ case (pred1, pred2) => assert(pred1 === pred2)
+ }
+ }
+
--- End diff --
Shall we add a "default params" for parity with other suites like
LogisticRegression?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
