fix discrete boosting symmetry

core/src/main/scala/org/apache/spark/ml/classification/BoostingClassifier.scala

@@ -367,7 +367,15 @@ class BoostingClassificationModel(
     val res = Array.fill(numClasses)(0.0)
     var i = 0
     while (i < numModels) {
-      res(models(i).predict(features).toInt) += weights(i)
+      val prediction = models(i).predict(features).toInt
+      val weight = weights(i)
+      var c = 0
+      while (c < numClasses) {
+        if (prediction == c) {
+          res(c) += weight
+        } else { res(c) -= 1.0 / (numClasses - 1) * weight }
+        c += 1
+      }
       i += 1
     }
     Vectors.dense(res)

core/src/test/scala/org/apache/spark/ml/classification/BoostingClassifierSuite.scala

@@ -25,6 +25,7 @@ import org.scalatest.funsuite.AnyFunSuite
 import org.scalatest.matchers.should.Matchers._
 
 import scala.collection.mutable.ListBuffer
+import org.apache.spark.ml.linalg.DenseVector
 
 class BoostingClassifierSuite extends AnyFunSuite with BeforeAndAfterAll {
 
@@ -81,7 +82,7 @@ class BoostingClassifierSuite extends AnyFunSuite with BeforeAndAfterAll {
           bcModel.models.take(i))
         metrics += mce.evaluate(model.transform(test))
       })
-    
+
     assert(
       metrics.toList
         .sliding(2)
@@ -122,6 +123,36 @@ class BoostingClassifierSuite extends AnyFunSuite with BeforeAndAfterAll {
       mce.evaluate(bcrModel.transform(test)) === mce.evaluate(bcdModel.transform(test)) +- 0.02)
   }
 
+  List("discrete", "real").foreach { algorithm =>
+    test(
+      f"$algorithm boosting decision function respects the symmetric constraint for weak learners") {
+      val data =
+        spark.read
+          .format("libsvm")
+          .load("../data/letter/letter.svm")
+          .withColumn("label", col("label") - lit(1))
+          .cache()
+      data.count()
+      val numClasses = data.select("label").distinct().count()
+
+      val lr = new DecisionTreeClassifier().setMaxDepth(1)
+      val boost = new BoostingClassifier()
+        .setBaseLearner(lr)
+        .setNumBaseLearners(10)
+        .setAlgorithm(algorithm)
+      val boostModel = boost.fit(data)
+
+      val predictions = boostModel
+        .transform(data)
+        .select("rawPrediction")
+        .distinct()
+        .collect()
+        .map(_.getAs[DenseVector](0).toArray)
+
+      predictions.foreach(raw => assert(raw.sum === 0.0 +- 1e-6))
+    }
+  }
+
   test("read/write") {
     val data =
       spark.read

docs/bagging.md

@@ -28,7 +28,7 @@ new BaggingClassifier()
         .setBaseLearner(new DecisionTreeClassifier()) //Base learner used by the meta-estimator.
         .setNumBaseLearners(10) //Number of base learners.
         .setSubsampleRatio(0.8) //Ratio sampling of examples.
-        .setReplacement(true) //Exemples drawn with replacement or not.
+        .setReplacement(true) //Samples drawn with replacement or not.
         .setSubspaceRatio(0.8) //Ratio sampling of features.
         .setVotingStrategy("soft") //Soft or Hard majority vote.
         .setParallelism(4) //Number of base learners trained simultaneously.
@@ -37,7 +37,7 @@ new BaggingRegressor()
         .setBaseLearner(new DecisionTreeRegressor()) //Base learner used by the meta-estimator.
         .setNumBaseLearners(10) //Number of base learners.
         .setSubsampleRatio(0.8) //Sampling ratio of examples.
-        .setReplacement(true) //Exemples drawn with replacement or not.
+        .setReplacement(true) //Samples drawn with replacement or not.
         .setSubspaceRatio(0.8) //Sampling ratio of features.
         .setParallelism(4) //Number of base learners trained simultaneously.
 ```