Add param to CrossValidator for choosing whether to maximize evaulation value.

Author: viirya

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/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala

@@ -76,6 +76,20 @@ private[ml] trait CrossValidatorParams extends Params {
   def getNumFolds: Int = $(numFolds)
 
   setDefault(numFolds -> 3)
+
+  /**
+   * Param for whether maximize the evaluation value during cross validation.
+   * If false, turn to minimize the evaluation value.
+   * Default: true
+   * @group param
+   */
+  val useMax: BooleanParam = new BooleanParam(this, "useMax",
+    "whether maximize the evaluation value durin cross validation")
+
+  /** @group getParam */
+  def getUseMax: Boolean = $(useMax)
+
+  setDefault(useMax -> true)
 }
 
 /**
@@ -102,6 +116,9 @@ class CrossValidator(override val uid: String) extends Estimator[CrossValidatorM
   /** @group setParam */
   def setNumFolds(value: Int): this.type = set(numFolds, value)
 
+  /** @group setParam */
+  def setUseMax(value: Boolean): this.type = set(useMax, value)
+
   override def fit(dataset: DataFrame): CrossValidatorModel = {
     val schema = dataset.schema
     transformSchema(schema, logging = true)
@@ -131,7 +148,11 @@ class CrossValidator(override val uid: String) extends Estimator[CrossValidatorM
     }
     f2jBLAS.dscal(numModels, 1.0 / $(numFolds), metrics, 1)
     logInfo(s"Average cross-validation metrics: ${metrics.toSeq}")
-    val (bestMetric, bestIndex) = metrics.zipWithIndex.maxBy(_._1)
+    val (bestMetric, bestIndex) = if ($(useMax)) {
+      metrics.zipWithIndex.maxBy(_._1)
+    } else {
+      metrics.zipWithIndex.minBy(_._1)
+    }
     logInfo(s"Best set of parameters:\n${epm(bestIndex)}")
     logInfo(s"Best cross-validation metric: $bestMetric.")
     val bestModel = est.fit(dataset, epm(bestIndex)).asInstanceOf[Model[_]]

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

@@ -21,11 +21,12 @@ 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
 
@@ -59,6 +60,30 @@ class CrossValidatorSuite extends SparkFunSuite with MLlibTestSparkContext {
     assert(cvModel.avgMetrics.length === lrParamMaps.length)
   }
 
+  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)
+      .setUseMax(false)
+    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)
+  }
+
   test("validateParams should check estimatorParamMaps") {
     import CrossValidatorSuite._