[SPARK-8468] [ML] Take the negative of some metrics in RegressionEvaluator to get correct cross validation

JIRA: https://issues.apache.org/jira/browse/SPARK-8468

Author: Liang-Chi Hsieh <viirya@gmail.com>

Closes #6905 from viirya/cv_min and squashes the following commits:

930d3db [Liang-Chi Hsieh] Fix python unit test and add document.
d632135 [Liang-Chi Hsieh] Merge remote-tracking branch 'upstream/master' into cv_min
16e3b2c [Liang-Chi Hsieh] Take the negative instead of reciprocal.
c3dd8d9 [Liang-Chi Hsieh] For comments.
b5f52c1 [Liang-Chi Hsieh] Add param to CrossValidator for choosing whether to maximize evaulation value.

(cherry picked from commit 0b89951)
Signed-off-by: Joseph K. Bradley <joseph@databricks.com>

Author: viirya

mllib/src/main/scala/org/apache/spark/ml/evaluation/RegressionEvaluator.scala

@@ -37,6 +37,10 @@ final class RegressionEvaluator(override val uid: String)
 
   /**
    * param for metric name in evaluation (supports `"rmse"` (default), `"mse"`, `"r2"`, and `"mae"`)
+   *
+   * Because we will maximize evaluation value (ref: `CrossValidator`),
+   * when we evaluate a metric that is needed to minimize (e.g., `"rmse"`, `"mse"`, `"mae"`),
+   * we take and output the negative of this metric.
    * @group param
    */
   val metricName: Param[String] = {
@@ -70,13 +74,13 @@ final class RegressionEvaluator(override val uid: String)
     val metrics = new RegressionMetrics(predictionAndLabels)
     val metric = $(metricName) match {
       case "rmse" =>
-        metrics.rootMeanSquaredError
+        -metrics.rootMeanSquaredError
       case "mse" =>
-        metrics.meanSquaredError
+        -metrics.meanSquaredError
       case "r2" =>
         metrics.r2
       case "mae" =>
-        metrics.meanAbsoluteError
+        -metrics.meanAbsoluteError
     }
     metric
   }

mllib/src/main/scala/org/apache/spark/ml/param/params.scala

@@ -297,7 +297,7 @@ class DoubleArrayParam(parent: Params, name: String, doc: String, isValid: Array
 
 /**
  * :: Experimental ::
- * A param amd its value.
+ * A param and its value.
  */
 @Experimental
 case class ParamPair[T](param: Param[T], value: T) {

mllib/src/test/scala/org/apache/spark/ml/evaluation/RegressionEvaluatorSuite.scala

@@ -63,14 +63,14 @@ class RegressionEvaluatorSuite extends SparkFunSuite with MLlibTestSparkContext
 
     // default = rmse
     val evaluator = new RegressionEvaluator()
-    assert(evaluator.evaluate(predictions) ~== 0.1019382 absTol 0.001)
+    assert(evaluator.evaluate(predictions) ~== -0.1019382 absTol 0.001)
 
     // r2 score
     evaluator.setMetricName("r2")
     assert(evaluator.evaluate(predictions) ~== 0.9998196 absTol 0.001)
 
     // mae
     evaluator.setMetricName("mae")
-    assert(evaluator.evaluate(predictions) ~== 0.08036075 absTol 0.001)
+    assert(evaluator.evaluate(predictions) ~== -0.08036075 absTol 0.001)
   }
 }

mllib/src/test/scala/org/apache/spark/ml/tuning/CrossValidatorSuite.scala

@@ -20,11 +20,12 @@ package org.apache.spark.ml.tuning
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.ml.{Estimator, Model}
 import org.apache.spark.ml.classification.LogisticRegression
-import org.apache.spark.ml.evaluation.{BinaryClassificationEvaluator, Evaluator}
+import org.apache.spark.ml.evaluation.{BinaryClassificationEvaluator, Evaluator, RegressionEvaluator}
 import org.apache.spark.ml.param.ParamMap
 import org.apache.spark.ml.param.shared.HasInputCol
+import org.apache.spark.ml.regression.LinearRegression
 import org.apache.spark.mllib.classification.LogisticRegressionSuite.generateLogisticInput
-import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.{LinearDataGenerator, MLlibTestSparkContext}
 import org.apache.spark.sql.{DataFrame, SQLContext}
 import org.apache.spark.sql.types.StructType
 
@@ -57,6 +58,36 @@ class CrossValidatorSuite extends SparkFunSuite with MLlibTestSparkContext {
     assert(parent.getMaxIter === 10)
   }
 
+  test("cross validation with linear regression") {
+    val dataset = sqlContext.createDataFrame(
+      sc.parallelize(LinearDataGenerator.generateLinearInput(
+        6.3, Array(4.7, 7.2), Array(0.9, -1.3), Array(0.7, 1.2), 100, 42, 0.1), 2))
+
+    val trainer = new LinearRegression
+    val lrParamMaps = new ParamGridBuilder()
+      .addGrid(trainer.regParam, Array(1000.0, 0.001))
+      .addGrid(trainer.maxIter, Array(0, 10))
+      .build()
+    val eval = new RegressionEvaluator()
+    val cv = new CrossValidator()
+      .setEstimator(trainer)
+      .setEstimatorParamMaps(lrParamMaps)
+      .setEvaluator(eval)
+      .setNumFolds(3)
+    val cvModel = cv.fit(dataset)
+    val parent = cvModel.bestModel.parent.asInstanceOf[LinearRegression]
+    assert(parent.getRegParam === 0.001)
+    assert(parent.getMaxIter === 10)
+    assert(cvModel.avgMetrics.length === lrParamMaps.length)
+
+    eval.setMetricName("r2")
+    val cvModel2 = cv.fit(dataset)
+    val parent2 = cvModel2.bestModel.parent.asInstanceOf[LinearRegression]
+    assert(parent2.getRegParam === 0.001)
+    assert(parent2.getMaxIter === 10)
+    assert(cvModel2.avgMetrics.length === lrParamMaps.length)
+  }
+
   test("validateParams should check estimatorParamMaps") {
     import CrossValidatorSuite._
 

python/pyspark/ml/evaluation.py

@@ -160,13 +160,15 @@ class RegressionEvaluator(JavaEvaluator, HasLabelCol, HasPredictionCol):
     ...
     >>> evaluator = RegressionEvaluator(predictionCol="raw")
     >>> evaluator.evaluate(dataset)
-    2.842...
+    -2.842...
     >>> evaluator.evaluate(dataset, {evaluator.metricName: "r2"})
     0.993...
     >>> evaluator.evaluate(dataset, {evaluator.metricName: "mae"})
-    2.649...
+    -2.649...
     """
-    # a placeholder to make it appear in the generated doc
+    # Because we will maximize evaluation value (ref: `CrossValidator`),
+    # when we evaluate a metric that is needed to minimize (e.g., `"rmse"`, `"mse"`, `"mae"`),
+    # we take and output the negative of this metric.
     metricName = Param(Params._dummy(), "metricName",
                        "metric name in evaluation (mse|rmse|r2|mae)")