[jvm-packages] Implemented early stopping (#2710)

* Allowed subsampling test from the training data frame/RDD

The implementation requires storing 1 - trainTestRatio points in memory
to make the sampling work.

An alternative approach would be to construct the full DMatrix and then
slice it deterministically into train/test. The peak memory consumption
of such scenario, however, is twice the dataset size.

* Removed duplication from 'XGBoost.train'

Scala callers can (and should) use names to supply a subset of
parameters. Method overloading is not required.

* Reuse XGBoost seed parameter to stabilize train/test splitting

* Added early stopping support to non-distributed XGBoost

Closes #1544

* Added early-stopping to distributed XGBoost

* Moved construction of 'watches' into a separate method

This commit also fixes the handling of 'baseMargin' which previously
was not added to the validation matrix.

* Addressed review comments

jvm-packages/xgboost4j-example/src/main/scala/ml/dmlc/xgboost4j/scala/example/BasicWalkThrough.scala

@@ -85,7 +85,7 @@ object BasicWalkThrough {
     val watches2 = new mutable.HashMap[String, DMatrix]
     watches2 += "train" -> trainMax2
     watches2 += "test" -> testMax2
-    val booster3 = XGBoost.train(trainMax2, params.toMap, round, watches2.toMap, null, null)
+    val booster3 = XGBoost.train(trainMax2, params.toMap, round, watches2.toMap)
     val predicts3 = booster3.predict(testMax2)
     println(checkPredicts(predicts, predicts3))
   }

jvm-packages/xgboost4j-example/src/main/scala/ml/dmlc/xgboost4j/scala/example/CrossValidation.scala

@@ -41,6 +41,6 @@ object CrossValidation {
     val metrics: Array[String] = null
 
     val evalHist: Array[String] =
-      XGBoost.crossValidation(trainMat, params.toMap, round, nfold, metrics, null, null)
+      XGBoost.crossValidation(trainMat, params.toMap, round, nfold, metrics)
   }
 }

jvm-packages/xgboost4j-example/src/main/scala/ml/dmlc/xgboost4j/scala/example/CustomObjective.scala

@@ -151,7 +151,8 @@ object CustomObjective {
     val round = 2
     // train a model
     val booster = XGBoost.train(trainMat, params.toMap, round, watches.toMap)
-    XGBoost.train(trainMat, params.toMap, round, watches.toMap, new LogRegObj, new EvalError)
+    XGBoost.train(trainMat, params.toMap, round, watches.toMap,
+      obj = new LogRegObj, eval = new EvalError)
   }
 
 }

jvm-packages/xgboost4j-example/src/main/scala/ml/dmlc/xgboost4j/scala/example/ExternalMemory.scala

@@ -54,6 +54,6 @@ object ExternalMemory {
     testMat.setBaseMargin(testPred)
 
     System.out.println("result of running from initial prediction")
-    val booster2 = XGBoost.train(trainMat, params.toMap, 1, watches.toMap, null, null)
+    val booster2 = XGBoost.train(trainMat, params.toMap, 1, watches.toMap)
   }
 }

jvm-packages/xgboost4j-example/src/main/scala/ml/dmlc/xgboost4j/scala/example/GeneralizedLinearModel.scala

@@ -52,7 +52,7 @@ object GeneralizedLinearModel {
     watches += "test" -> testMat
 
     val round = 4
-    val booster = XGBoost.train(trainMat, params.toMap, 1, watches.toMap, null, null)
+    val booster = XGBoost.train(trainMat, params.toMap, 1, watches.toMap)
     val predicts = booster.predict(testMat)
     val eval = new CustomEval
     println(s"error=${eval.eval(predicts, testMat)}")