More mima excludes, added lots of warnings to not use impurity

Author: vlad17

mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala

@@ -59,9 +59,8 @@ import org.apache.spark.sql.types.DoubleType
  *  - In the case of squared error loss, variance impurity and mean leaf estimates happen
  *    to make the SGB and TreeBoost algorithms identical.
  *
- * [[GBTClassifier]] will use the usual `"loss-based"` impurity by default, conforming to
+ * [[GBTClassifier]] will use the `"loss-based"` impurity by default, conforming to
  * TreeBoost behavior. For SGB, set impurity to `"variance"`.
- * To use of TreeBoost, set impurity to `"loss-based"`.
  *
  * Currently, however, even TreeBoost behavior uses variance impurity for split selection for
  * ease and speed. This is the approach `R`'s
@@ -70,7 +69,7 @@ import org.apache.spark.sql.types.DoubleType
 @Since("1.4.0")
 class GBTClassifier @Since("1.4.0") (
     @Since("1.4.0") override val uid: String)
-  extends Classifier[Vector, GBTClassifier, GBTClassificationModel]
+  extends Predictor[Vector, GBTClassifier, GBTClassificationModel]
   with GBTClassifierParams with DefaultParamsWritable with Logging {
 
   @Since("1.4.0")
@@ -102,6 +101,18 @@ class GBTClassifier @Since("1.4.0") (
   @Since("1.4.0")
   override def setCheckpointInterval(value: Int): this.type = super.setCheckpointInterval(value)
 
+  /**
+   * Impurity-setting is currently only offered as a way to recover pre-2.0.2 Spark GBT
+   * behavior (which is Stochastic Gradient Boosting): set impurity to `"variance"` for this.
+   * @param value new impurity value
+   * @return this
+   */
+  @Since("1.4.0")
+  @deprecated(
+    "Control over impurity will be removed, as it is an implementation detail of GBTs",
+    "2.0.2")
+  override def setImpurity(value: String): this.type = super.setImpurity(value)
+
   // Parameters from TreeEnsembleParams:
 
   @Since("1.4.0")

mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala

@@ -98,10 +98,16 @@ class GBTRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: String)
 
   /**
    * Note that the loss-based impurity is currently NOT compatible with absolute loss.
+   *
+   * Impurity-setting is currently only offered as a way to recover pre-2.0.2 Spark GBT
+   * behavior (which is Stochastic Gradient Boosting): set impurity to `"variance"` for this.
    * @param value new impurity value
    * @return this
    */
   @Since("1.4.0")
+  @deprecated(
+    "Control over impurity will be removed, as it is an implementation detail of GBTs",
+    "2.0.2")
   override def setImpurity(value: String): this.type = super.setImpurity(value)
 
   // Parameters from TreeEnsembleParams:

mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala

@@ -516,8 +516,11 @@ private[ml] trait GBTClassifierParams extends GBTParams with TreeClassifierParam
    * Also used for terminal leaf value prediction.
    * Supported: "loss-based" (default) and "variance"
    *
-   * @group param
+   * @group expertParam
    */
+  @deprecated(
+      "Control over impurity will be removed, as it is an implementation detail of GBTs",
+      "2.0.2")
   override val impurity: Param[String] = new Param[String](this, "impurity", "Criterion used for" +
     " information gain calculation (case-insensitive). Supported options:" +
     s" ${GBTClassifierParams.supportedImpurities.mkString(", ")}",
@@ -590,8 +593,11 @@ private[ml] trait GBTRegressorParams extends GBTParams with TreeRegressorParams
    * Also used for terminal leaf value prediction.
    * Supported: "loss-based" and "variance" (default)
    *
-   * @group param
+   * @group expertParam
    */
+  @deprecated(
+    "Control over impurity will be removed, as it is an implementation detail of GBTs",
+    "2.0.2")
   override val impurity: Param[String] = new Param[String](this, "impurity", "Criterion used for" +
     " information gain calculation (case-insensitive). Supported options:" +
     s" ${GBTRegressorParams.supportedImpurities.mkString(", ")}",

project/MimaExcludes.scala

@@ -869,7 +869,13 @@ object MimaExcludes {
       ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.ml.classification.GBTClassifier.getOldBoostingStrategy"),
       ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.ml.classification.GBTClassificationModel.getOldBoostingStrategy"),
       ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.ml.regression.GBTRegressor.getOldBoostingStrategy"),
-      ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.ml.regression.GBTRegressionModel.getOldBoostingStrategy")
+      ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.ml.regression.GBTRegressionModel.getOldBoostingStrategy"),
+ProblemFilters.exclude[InheritedNewAbstractMethodProblem]("org.apache.spark.ml.tree.TreeRegressorParamsWithDefault.org$apache$spark$ml$tree$TreeRegressorParamsWithDefault$_setter_$impurity_="),
+ProblemFilters.exclude[InheritedNewAbstractMethodProblem]("org.apache.spark.ml.tree.TreeClassifierParamsWithDefault.org$apache$spark$ml$tree$TreeClassifierParamsWithDefault$_setter_$impurity_="),
+ProblemFilters.exclude[InheritedNewAbstractMethodProblem]("org.apache.spark.ml.tree.TreeClassifierParamsWithDefault.org$apache$spark$ml$tree$TreeClassifierParamsWithDefault$_setter_$impurity_="),
+ProblemFilters.exclude[InheritedNewAbstractMethodProblem]("org.apache.spark.ml.tree.TreeRegressorParamsWithDefault.org$apache$spark$ml$tree$TreeRegressorParamsWithDefault$_setter_$impurity_="),
+ProblemFilters.exclude[InheritedNewAbstractMethodProblem]("org.apache.spark.ml.tree.TreeClassifierParamsWithDefault.org$apache$spark$ml$tree$TreeClassifierParamsWithDefault$_setter_$impurity_="),
+ProblemFilters.exclude[InheritedNewAbstractMethodProblem]("org.apache.spark.ml.tree.TreeRegressorParamsWithDefault.org$apache$spark$ml$tree$TreeRegressorParamsWithDefault$_setter_$impurity_=")
     )
   }