[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-spark/src/test/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostDFSuite.scala

@@ -18,6 +18,7 @@ package ml.dmlc.xgboost4j.scala.spark
 
 import ml.dmlc.xgboost4j.scala.{DMatrix, XGBoost => ScalaXGBoost}
 import ml.dmlc.xgboost4j.{LabeledPoint => XGBLabeledPoint}
+
 import org.apache.spark.ml.linalg.DenseVector
 import org.apache.spark.ml.param.ParamMap
 import org.apache.spark.sql._
@@ -201,7 +202,8 @@ class XGBoostDFSuite extends FunSuite with PerTest {
     val trainingDfWithMargin = trainingDf.withColumn("margin", functions.rand())
     val testRDD = sc.parallelize(Classification.test.map(_.features))
     val paramMap = Map("eta" -> "1", "max_depth" -> "6", "silent" -> "1",
-      "objective" -> "binary:logistic", "baseMarginCol" -> "margin")
+      "objective" -> "binary:logistic", "baseMarginCol" -> "margin",
+      "testTrainSplit" -> 0.5)
 
     def trainPredict(df: Dataset[_]): Array[Float] = {
       XGBoost.trainWithDataFrame(df, paramMap, round = 1, nWorkers = numWorkers)