diff --git a/.gitignore b/.gitignore
index 34939e3a97aaa..3b9086c7187dc 100644
--- a/.gitignore
+++ b/.gitignore
@@ -5,6 +5,7 @@
 *.ipr
 *.iml
 *.iws
+*.pyc
 .idea/
 .idea_modules/
 sbt/*.jar
@@ -49,7 +50,9 @@ dependency-reduced-pom.xml
 checkpoint
 derby.log
 dist/
-spark-*-bin.tar.gz
+dev/create-release/*txt
+dev/create-release/*new
+spark-*-bin-*.tgz
 unit-tests.log
 /lib/
 rat-results.txt
diff --git a/bin/beeline.cmd b/bin/beeline.cmd
new file mode 100644
index 0000000000000..8293f311029dd
--- /dev/null
+++ b/bin/beeline.cmd
@@ -0,0 +1,21 @@
+@echo off
+
+rem
+rem Licensed to the Apache Software Foundation (ASF) under one or more
+rem contributor license agreements.  See the NOTICE file distributed with
+rem this work for additional information regarding copyright ownership.
+rem The ASF licenses this file to You under the Apache License, Version 2.0
+rem (the "License"); you may not use this file except in compliance with
+rem the License.  You may obtain a copy of the License at
+rem
+rem    http://www.apache.org/licenses/LICENSE-2.0
+rem
+rem Unless required by applicable law or agreed to in writing, software
+rem distributed under the License is distributed on an "AS IS" BASIS,
+rem WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+rem See the License for the specific language governing permissions and
+rem limitations under the License.
+rem
+
+set SPARK_HOME=%~dp0..
+cmd /V /E /C %SPARK_HOME%\bin\spark-class.cmd org.apache.hive.beeline.BeeLine %*
diff --git a/bin/spark-sql b/bin/spark-sql
index 63d00437d508d..3b6cc420fea81 100755
--- a/bin/spark-sql
+++ b/bin/spark-sql
@@ -23,6 +23,8 @@
 # Enter posix mode for bash
 set -o posix
 
+# NOTE: This exact class name is matched downstream by SparkSubmit.
+# Any changes need to be reflected there.
 CLASS="org.apache.spark.sql.hive.thriftserver.SparkSQLCLIDriver"
 
 # Figure out where Spark is installed
diff --git a/core/src/main/scala/org/apache/spark/SparkConf.scala b/core/src/main/scala/org/apache/spark/SparkConf.scala
index 4c6c86c7bad78..c14764f773982 100644
--- a/core/src/main/scala/org/apache/spark/SparkConf.scala
+++ b/core/src/main/scala/org/apache/spark/SparkConf.scala
@@ -61,7 +61,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging {
       throw new NullPointerException("null key")
     }
     if (value == null) {
-      throw new NullPointerException("null value")
+      throw new NullPointerException("null value for " + key)
     }
     settings(key) = value
     this
diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala
index 9b0d5be7a7ab2..aded7c12e274e 100644
--- a/core/src/main/scala/org/apache/spark/SparkContext.scala
+++ b/core/src/main/scala/org/apache/spark/SparkContext.scala
@@ -1630,28 +1630,28 @@ object SparkContext extends Logging {
   // following ones.
 
   @deprecated("Replaced by implicit objects in AccumulatorParam. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   object DoubleAccumulatorParam extends AccumulatorParam[Double] {
     def addInPlace(t1: Double, t2: Double): Double = t1 + t2
     def zero(initialValue: Double) = 0.0
   }
 
   @deprecated("Replaced by implicit objects in AccumulatorParam. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   object IntAccumulatorParam extends AccumulatorParam[Int] {
     def addInPlace(t1: Int, t2: Int): Int = t1 + t2
     def zero(initialValue: Int) = 0
   }
 
   @deprecated("Replaced by implicit objects in AccumulatorParam. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   object LongAccumulatorParam extends AccumulatorParam[Long] {
     def addInPlace(t1: Long, t2: Long) = t1 + t2
     def zero(initialValue: Long) = 0L
   }
 
   @deprecated("Replaced by implicit objects in AccumulatorParam. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   object FloatAccumulatorParam extends AccumulatorParam[Float] {
     def addInPlace(t1: Float, t2: Float) = t1 + t2
     def zero(initialValue: Float) = 0f
@@ -1662,34 +1662,34 @@ object SparkContext extends Logging {
   // and just call the corresponding functions in `object RDD`.
 
   @deprecated("Replaced by implicit functions in the RDD companion object. This is " +
-    "kept here only for backward compatibility.", "1.2.0")
+    "kept here only for backward compatibility.", "1.3.0")
   def rddToPairRDDFunctions[K, V](rdd: RDD[(K, V)])
       (implicit kt: ClassTag[K], vt: ClassTag[V], ord: Ordering[K] = null) = {
     RDD.rddToPairRDDFunctions(rdd)
   }
 
   @deprecated("Replaced by implicit functions in the RDD companion object. This is " +
-    "kept here only for backward compatibility.", "1.2.0")
+    "kept here only for backward compatibility.", "1.3.0")
   def rddToAsyncRDDActions[T: ClassTag](rdd: RDD[T]) = RDD.rddToAsyncRDDActions(rdd)
 
   @deprecated("Replaced by implicit functions in the RDD companion object. This is " +
-    "kept here only for backward compatibility.", "1.2.0")
+    "kept here only for backward compatibility.", "1.3.0")
   def rddToSequenceFileRDDFunctions[K <% Writable: ClassTag, V <% Writable: ClassTag](
       rdd: RDD[(K, V)]) =
     RDD.rddToSequenceFileRDDFunctions(rdd)
 
   @deprecated("Replaced by implicit functions in the RDD companion object. This is " +
-    "kept here only for backward compatibility.", "1.2.0")
+    "kept here only for backward compatibility.", "1.3.0")
   def rddToOrderedRDDFunctions[K : Ordering : ClassTag, V: ClassTag](
       rdd: RDD[(K, V)]) =
     RDD.rddToOrderedRDDFunctions(rdd)
 
   @deprecated("Replaced by implicit functions in the RDD companion object. This is " +
-    "kept here only for backward compatibility.", "1.2.0")
+    "kept here only for backward compatibility.", "1.3.0")
   def doubleRDDToDoubleRDDFunctions(rdd: RDD[Double]) = RDD.doubleRDDToDoubleRDDFunctions(rdd)
 
   @deprecated("Replaced by implicit functions in the RDD companion object. This is " +
-    "kept here only for backward compatibility.", "1.2.0")
+    "kept here only for backward compatibility.", "1.3.0")
   def numericRDDToDoubleRDDFunctions[T](rdd: RDD[T])(implicit num: Numeric[T]) =
     RDD.numericRDDToDoubleRDDFunctions(rdd)
 
@@ -1722,43 +1722,43 @@ object SparkContext extends Logging {
   // and just call the corresponding functions in `object WritableConverter`.
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def intWritableConverter(): WritableConverter[Int] =
     WritableConverter.intWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def longWritableConverter(): WritableConverter[Long] =
     WritableConverter.longWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def doubleWritableConverter(): WritableConverter[Double] =
     WritableConverter.doubleWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def floatWritableConverter(): WritableConverter[Float] =
     WritableConverter.floatWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def booleanWritableConverter(): WritableConverter[Boolean] =
     WritableConverter.booleanWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def bytesWritableConverter(): WritableConverter[Array[Byte]] =
     WritableConverter.bytesWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
+    "backward compatibility.", "1.3.0")
   def stringWritableConverter(): WritableConverter[String] =
     WritableConverter.stringWritableConverter()
 
   @deprecated("Replaced by implicit functions in WritableConverter. This is kept here only for " +
-    "backward compatibility.", "1.2.0")
-  def writableWritableConverter[T <: Writable]() =
+    "backward compatibility.", "1.3.0")
+  def writableWritableConverter[T <: Writable](): WritableConverter[T] =
     WritableConverter.writableWritableConverter()
 
   /**
@@ -2017,15 +2017,15 @@ object WritableConverter {
     simpleWritableConverter[Boolean, BooleanWritable](_.get)
 
   implicit def bytesWritableConverter(): WritableConverter[Array[Byte]] = {
-    simpleWritableConverter[Array[Byte], BytesWritable](bw =>
+    simpleWritableConverter[Array[Byte], BytesWritable] { bw =>
       // getBytes method returns array which is longer then data to be returned
       Arrays.copyOfRange(bw.getBytes, 0, bw.getLength)
-    )
+    }
   }
 
   implicit def stringWritableConverter(): WritableConverter[String] =
     simpleWritableConverter[String, Text](_.toString)
 
-  implicit def writableWritableConverter[T <: Writable]() =
+  implicit def writableWritableConverter[T <: Writable](): WritableConverter[T] =
     new WritableConverter[T](_.runtimeClass.asInstanceOf[Class[T]], _.asInstanceOf[T])
 }
diff --git a/core/src/main/scala/org/apache/spark/api/java/JavaRDDLike.scala b/core/src/main/scala/org/apache/spark/api/java/JavaRDDLike.scala
index 5a8e5bb1f721a..bd451634e53d2 100644
--- a/core/src/main/scala/org/apache/spark/api/java/JavaRDDLike.scala
+++ b/core/src/main/scala/org/apache/spark/api/java/JavaRDDLike.scala
@@ -28,7 +28,6 @@ import com.google.common.base.Optional
 import org.apache.hadoop.io.compress.CompressionCodec
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 import org.apache.spark.annotation.Experimental
 import org.apache.spark.api.java.JavaPairRDD._
 import org.apache.spark.api.java.JavaSparkContext.fakeClassTag
@@ -212,8 +211,9 @@ trait JavaRDDLike[T, This <: JavaRDDLike[T, This]] extends Serializable {
    * Return an RDD of grouped elements. Each group consists of a key and a sequence of elements
    * mapping to that key.
    */
-  def groupBy[K](f: JFunction[T, K]): JavaPairRDD[K, JIterable[T]] = {
-    implicit val ctagK: ClassTag[K] = fakeClassTag
+  def groupBy[U](f: JFunction[T, U]): JavaPairRDD[U, JIterable[T]] = {
+    // The type parameter is U instead of K in order to work around a compiler bug; see SPARK-4459
+    implicit val ctagK: ClassTag[U] = fakeClassTag
     implicit val ctagV: ClassTag[JList[T]] = fakeClassTag
     JavaPairRDD.fromRDD(groupByResultToJava(rdd.groupBy(f)(fakeClassTag)))
   }
@@ -222,10 +222,11 @@ trait JavaRDDLike[T, This <: JavaRDDLike[T, This]] extends Serializable {
    * Return an RDD of grouped elements. Each group consists of a key and a sequence of elements
    * mapping to that key.
    */
-  def groupBy[K](f: JFunction[T, K], numPartitions: Int): JavaPairRDD[K, JIterable[T]] = {
-    implicit val ctagK: ClassTag[K] = fakeClassTag
+  def groupBy[U](f: JFunction[T, U], numPartitions: Int): JavaPairRDD[U, JIterable[T]] = {
+    // The type parameter is U instead of K in order to work around a compiler bug; see SPARK-4459
+    implicit val ctagK: ClassTag[U] = fakeClassTag
     implicit val ctagV: ClassTag[JList[T]] = fakeClassTag
-    JavaPairRDD.fromRDD(groupByResultToJava(rdd.groupBy(f, numPartitions)(fakeClassTag[K])))
+    JavaPairRDD.fromRDD(groupByResultToJava(rdd.groupBy(f, numPartitions)(fakeClassTag[U])))
   }
 
   /**
@@ -459,8 +460,9 @@ trait JavaRDDLike[T, This <: JavaRDDLike[T, This]] extends Serializable {
   /**
    * Creates tuples of the elements in this RDD by applying `f`.
    */
-  def keyBy[K](f: JFunction[T, K]): JavaPairRDD[K, T] = {
-    implicit val ctag: ClassTag[K] = fakeClassTag
+  def keyBy[U](f: JFunction[T, U]): JavaPairRDD[U, T] = {
+    // The type parameter is U instead of K in order to work around a compiler bug; see SPARK-4459
+    implicit val ctag: ClassTag[U] = fakeClassTag
     JavaPairRDD.fromRDD(rdd.keyBy(f))
   }
 
diff --git a/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala b/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
index e0bc00e1eb249..bad40e6529f74 100644
--- a/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
+++ b/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
@@ -34,7 +34,6 @@ import org.apache.hadoop.io.compress.CompressionCodec
 import org.apache.hadoop.mapred.{InputFormat, OutputFormat, JobConf}
 import org.apache.hadoop.mapreduce.{InputFormat => NewInputFormat, OutputFormat => NewOutputFormat}
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 import org.apache.spark.api.java.{JavaSparkContext, JavaPairRDD, JavaRDD}
 import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.rdd.RDD
diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala
index 00f291823e984..955cbd6dab96d 100644
--- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala
@@ -142,6 +142,8 @@ object SparkSubmit {
         printErrorAndExit("Cluster deploy mode is currently not supported for python applications.")
       case (_, CLUSTER) if isShell(args.primaryResource) =>
         printErrorAndExit("Cluster deploy mode is not applicable to Spark shells.")
+      case (_, CLUSTER) if isSqlShell(args.mainClass) =>
+        printErrorAndExit("Cluster deploy mode is not applicable to Spark SQL shell.")
       case _ =>
     }
 
@@ -279,6 +281,11 @@ object SparkSubmit {
       sysProps.getOrElseUpdate(k, v)
     }
 
+    // Ignore invalid spark.driver.host in cluster modes.
+    if (deployMode == CLUSTER) {
+      sysProps -= ("spark.driver.host")
+    }
+
     // Resolve paths in certain spark properties
     val pathConfigs = Seq(
       "spark.jars",
@@ -393,6 +400,13 @@ object SparkSubmit {
     primaryResource == SPARK_SHELL || primaryResource == PYSPARK_SHELL
   }
 
+  /**
+   * Return whether the given main class represents a sql shell.
+   */
+  private[spark] def isSqlShell(mainClass: String): Boolean = {
+    mainClass == "org.apache.spark.sql.hive.thriftserver.SparkSQLCLIDriver"
+  }
+
   /**
    * Return whether the given primary resource requires running python.
    */
diff --git a/core/src/main/scala/org/apache/spark/deploy/client/AppClient.scala b/core/src/main/scala/org/apache/spark/deploy/client/AppClient.scala
index 98a93d1fcb2a3..4efebcaa350fe 100644
--- a/core/src/main/scala/org/apache/spark/deploy/client/AppClient.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/client/AppClient.scala
@@ -134,6 +134,7 @@ private[spark] class AppClient(
         val fullId = appId + "/" + id
         logInfo("Executor added: %s on %s (%s) with %d cores".format(fullId, workerId, hostPort,
           cores))
+        master ! ExecutorStateChanged(appId, id, ExecutorState.RUNNING, None, None)
         listener.executorAdded(fullId, workerId, hostPort, cores, memory)
 
       case ExecutorUpdated(id, state, message, exitStatus) =>
diff --git a/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala b/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
index 8ba6a01bbcb97..f4fedc6327ab9 100644
--- a/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala
@@ -144,8 +144,6 @@ private[spark] class ExecutorRunner(
       Files.write(header, stderr, UTF_8)
       stderrAppender = FileAppender(process.getErrorStream, stderr, conf)
 
-      state = ExecutorState.RUNNING
-      worker ! ExecutorStateChanged(appId, execId, state, None, None)
       // Wait for it to exit; executor may exit with code 0 (when driver instructs it to shutdown)
       // or with nonzero exit code
       val exitCode = process.waitFor()
diff --git a/core/src/main/scala/org/apache/spark/package.scala b/core/src/main/scala/org/apache/spark/package.scala
index 436dbed1730bc..5ad73c3d27f47 100644
--- a/core/src/main/scala/org/apache/spark/package.scala
+++ b/core/src/main/scala/org/apache/spark/package.scala
@@ -27,8 +27,8 @@ package org.apache
  * contains operations available only on RDDs of Doubles; and
  * [[org.apache.spark.rdd.SequenceFileRDDFunctions]] contains operations available on RDDs that can
  * be saved as SequenceFiles. These operations are automatically available on any RDD of the right
- * type (e.g. RDD[(Int, Int)] through implicit conversions when you
- * `import org.apache.spark.SparkContext._`.
+ * type (e.g. RDD[(Int, Int)] through implicit conversions except `saveAsSequenceFile`. You need to
+ * `import org.apache.spark.SparkContext._` to make `saveAsSequenceFile` work.
  *
  * Java programmers should reference the [[org.apache.spark.api.java]] package
  * for Spark programming APIs in Java.
diff --git a/core/src/main/scala/org/apache/spark/rdd/AsyncRDDActions.scala b/core/src/main/scala/org/apache/spark/rdd/AsyncRDDActions.scala
index 9f9f10b7ebc3a..646df283ac069 100644
--- a/core/src/main/scala/org/apache/spark/rdd/AsyncRDDActions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/AsyncRDDActions.scala
@@ -27,7 +27,6 @@ import org.apache.spark.{ComplexFutureAction, FutureAction, Logging}
 
 /**
  * A set of asynchronous RDD actions available through an implicit conversion.
- * Import `org.apache.spark.SparkContext._` at the top of your program to use these functions.
  */
 class AsyncRDDActions[T: ClassTag](self: RDD[T]) extends Serializable with Logging {
 
diff --git a/core/src/main/scala/org/apache/spark/rdd/BinaryFileRDD.scala b/core/src/main/scala/org/apache/spark/rdd/BinaryFileRDD.scala
index 6e66ddbdef788..1f755db485812 100644
--- a/core/src/main/scala/org/apache/spark/rdd/BinaryFileRDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/BinaryFileRDD.scala
@@ -24,12 +24,12 @@ import org.apache.spark.input.StreamFileInputFormat
 import org.apache.spark.{ Partition, SparkContext }
 
 private[spark] class BinaryFileRDD[T](
-  sc: SparkContext,
-  inputFormatClass: Class[_ <: StreamFileInputFormat[T]],
-  keyClass: Class[String],
-  valueClass: Class[T],
-  @transient conf: Configuration,
-  minPartitions: Int)
+    sc: SparkContext,
+    inputFormatClass: Class[_ <: StreamFileInputFormat[T]],
+    keyClass: Class[String],
+    valueClass: Class[T],
+    @transient conf: Configuration,
+    minPartitions: Int)
   extends NewHadoopRDD[String, T](sc, inputFormatClass, keyClass, valueClass, conf) {
 
   override def getPartitions: Array[Partition] = {
diff --git a/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala
index e0494ee39657c..e66f83bb34e30 100644
--- a/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/DoubleRDDFunctions.scala
@@ -27,7 +27,6 @@ import org.apache.spark.util.StatCounter
 
 /**
  * Extra functions available on RDDs of Doubles through an implicit conversion.
- * Import `org.apache.spark.SparkContext._` at the top of your program to use these functions.
  */
 class DoubleRDDFunctions(self: RDD[Double]) extends Logging with Serializable {
   /** Add up the elements in this RDD. */
diff --git a/core/src/main/scala/org/apache/spark/rdd/FilteredRDD.scala b/core/src/main/scala/org/apache/spark/rdd/FilteredRDD.scala
deleted file mode 100644
index 9e41b3d1e2d4f..0000000000000
--- a/core/src/main/scala/org/apache/spark/rdd/FilteredRDD.scala
+++ /dev/null
@@ -1,35 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.rdd
-
-import scala.reflect.ClassTag
-
-import org.apache.spark.{Partition, TaskContext}
-
-private[spark] class FilteredRDD[T: ClassTag](
-    prev: RDD[T],
-    f: T => Boolean)
-  extends RDD[T](prev) {
-
-  override def getPartitions: Array[Partition] = firstParent[T].partitions
-
-  override val partitioner = prev.partitioner    // Since filter cannot change a partition's keys
-
-  override def compute(split: Partition, context: TaskContext) =
-    firstParent[T].iterator(split, context).filter(f)
-}
diff --git a/core/src/main/scala/org/apache/spark/rdd/FlatMappedRDD.scala b/core/src/main/scala/org/apache/spark/rdd/FlatMappedRDD.scala
deleted file mode 100644
index d8f87d4e3690e..0000000000000
--- a/core/src/main/scala/org/apache/spark/rdd/FlatMappedRDD.scala
+++ /dev/null
@@ -1,34 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.rdd
-
-import scala.reflect.ClassTag
-
-import org.apache.spark.{Partition, TaskContext}
-
-private[spark]
-class FlatMappedRDD[U: ClassTag, T: ClassTag](
-    prev: RDD[T],
-    f: T => TraversableOnce[U])
-  extends RDD[U](prev) {
-
-  override def getPartitions: Array[Partition] = firstParent[T].partitions
-
-  override def compute(split: Partition, context: TaskContext) =
-    firstParent[T].iterator(split, context).flatMap(f)
-}
diff --git a/core/src/main/scala/org/apache/spark/rdd/FlatMappedValuesRDD.scala b/core/src/main/scala/org/apache/spark/rdd/FlatMappedValuesRDD.scala
deleted file mode 100644
index 7c9023f62d3b6..0000000000000
--- a/core/src/main/scala/org/apache/spark/rdd/FlatMappedValuesRDD.scala
+++ /dev/null
@@ -1,35 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.rdd
-
-import org.apache.spark.{Partition, TaskContext}
-
-private[spark]
-class FlatMappedValuesRDD[K, V, U](prev: RDD[_ <: Product2[K, V]], f: V => TraversableOnce[U])
-  extends RDD[(K, U)](prev) {
-
-  override def getPartitions = firstParent[Product2[K, V]].partitions
-
-  override val partitioner = firstParent[Product2[K, V]].partitioner
-
-  override def compute(split: Partition, context: TaskContext) = {
-    firstParent[Product2[K, V]].iterator(split, context).flatMap { case Product2(k, v) =>
-      f(v).map(x => (k, x))
-    }
-  }
-}
diff --git a/core/src/main/scala/org/apache/spark/rdd/GlommedRDD.scala b/core/src/main/scala/org/apache/spark/rdd/GlommedRDD.scala
deleted file mode 100644
index f6463fa715a71..0000000000000
--- a/core/src/main/scala/org/apache/spark/rdd/GlommedRDD.scala
+++ /dev/null
@@ -1,31 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.rdd
-
-import scala.reflect.ClassTag
-
-import org.apache.spark.{Partition, TaskContext}
-
-private[spark] class GlommedRDD[T: ClassTag](prev: RDD[T])
-  extends RDD[Array[T]](prev) {
-
-  override def getPartitions: Array[Partition] = firstParent[T].partitions
-
-  override def compute(split: Partition, context: TaskContext) =
-    Array(firstParent[T].iterator(split, context).toArray).iterator
-}
diff --git a/core/src/main/scala/org/apache/spark/rdd/MappedValuesRDD.scala b/core/src/main/scala/org/apache/spark/rdd/MappedValuesRDD.scala
deleted file mode 100644
index a60952eee5901..0000000000000
--- a/core/src/main/scala/org/apache/spark/rdd/MappedValuesRDD.scala
+++ /dev/null
@@ -1,33 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one or more
- * contributor license agreements.  See the NOTICE file distributed with
- * this work for additional information regarding copyright ownership.
- * The ASF licenses this file to You under the Apache License, Version 2.0
- * (the "License"); you may not use this file except in compliance with
- * the License.  You may obtain a copy of the License at
- *
- *    http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-package org.apache.spark.rdd
-
-import org.apache.spark.{Partition, TaskContext}
-
-private[spark]
-class MappedValuesRDD[K, V, U](prev: RDD[_ <: Product2[K, V]], f: V => U)
-  extends RDD[(K, U)](prev) {
-
-  override def getPartitions = firstParent[Product2[K, U]].partitions
-
-  override val partitioner = firstParent[Product2[K, U]].partitioner
-
-  override def compute(split: Partition, context: TaskContext): Iterator[(K, U)] = {
-    firstParent[Product2[K, V]].iterator(split, context).map { pair => (pair._1, f(pair._2)) }
-  }
-}
diff --git a/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala
index d0dbfef35d03c..144f679a59460 100644
--- a/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/OrderedRDDFunctions.scala
@@ -24,10 +24,9 @@ import org.apache.spark.annotation.DeveloperApi
 
 /**
  * Extra functions available on RDDs of (key, value) pairs where the key is sortable through
- * an implicit conversion. Import `org.apache.spark.SparkContext._` at the top of your program to
- * use these functions. They will work with any key type `K` that has an implicit `Ordering[K]` in
- * scope.  Ordering objects already exist for all of the standard primitive types.  Users can also
- * define their own orderings for custom types, or to override the default ordering.  The implicit
+ * an implicit conversion. They will work with any key type `K` that has an implicit `Ordering[K]`
+ * in scope. Ordering objects already exist for all of the standard primitive types. Users can also
+ * define their own orderings for custom types, or to override the default ordering. The implicit
  * ordering that is in the closest scope will be used.
  *
  * {{{
diff --git a/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala
index 8c2c959e73bb6..c43e1f2fe135e 100644
--- a/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala
@@ -37,7 +37,6 @@ RecordWriter => NewRecordWriter}
 
 import org.apache.spark._
 import org.apache.spark.Partitioner.defaultPartitioner
-import org.apache.spark.SparkContext._
 import org.apache.spark.annotation.Experimental
 import org.apache.spark.deploy.SparkHadoopUtil
 import org.apache.spark.executor.{DataWriteMethod, OutputMetrics}
@@ -50,7 +49,6 @@ import org.apache.spark.util.random.StratifiedSamplingUtils
 
 /**
  * Extra functions available on RDDs of (key, value) pairs through an implicit conversion.
- * Import `org.apache.spark.SparkContext._` at the top of your program to use these functions.
  */
 class PairRDDFunctions[K, V](self: RDD[(K, V)])
     (implicit kt: ClassTag[K], vt: ClassTag[V], ord: Ordering[K] = null)
@@ -662,7 +660,9 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)])
    */
   def mapValues[U](f: V => U): RDD[(K, U)] = {
     val cleanF = self.context.clean(f)
-    new MappedValuesRDD(self, cleanF)
+    new MapPartitionsRDD[(K, U), (K, V)](self,
+      (context, pid, iter) => iter.map { case (k, v) => (k, cleanF(v)) },
+      preservesPartitioning = true)
   }
 
   /**
@@ -671,7 +671,11 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)])
    */
   def flatMapValues[U](f: V => TraversableOnce[U]): RDD[(K, U)] = {
     val cleanF = self.context.clean(f)
-    new FlatMappedValuesRDD(self, cleanF)
+    new MapPartitionsRDD[(K, U), (K, V)](self,
+      (context, pid, iter) => iter.flatMap { case (k, v) =>
+        cleanF(v).map(x => (k, x))
+      },
+      preservesPartitioning = true)
   }
 
   /**
diff --git a/core/src/main/scala/org/apache/spark/rdd/PipedRDD.scala b/core/src/main/scala/org/apache/spark/rdd/PipedRDD.scala
index 56ac7a69be0d3..ed79032893d33 100644
--- a/core/src/main/scala/org/apache/spark/rdd/PipedRDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/PipedRDD.scala
@@ -63,7 +63,7 @@ private[spark] class PipedRDD[T: ClassTag](
 
   /**
    * A FilenameFilter that accepts anything that isn't equal to the name passed in.
-   * @param name of file or directory to leave out
+   * @param filterName of file or directory to leave out
    */
   class NotEqualsFileNameFilter(filterName: String) extends FilenameFilter {
     def accept(dir: File, name: String): Boolean = {
diff --git a/core/src/main/scala/org/apache/spark/rdd/RDD.scala b/core/src/main/scala/org/apache/spark/rdd/RDD.scala
index 3add4a76192ca..214f22bc5b603 100644
--- a/core/src/main/scala/org/apache/spark/rdd/RDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/RDD.scala
@@ -17,7 +17,7 @@
 
 package org.apache.spark.rdd
 
-import java.util.{Properties, Random}
+import java.util.Random
 
 import scala.collection.{mutable, Map}
 import scala.collection.mutable.ArrayBuffer
@@ -34,16 +34,14 @@ import org.apache.hadoop.mapred.TextOutputFormat
 
 import org.apache.spark._
 import org.apache.spark.Partitioner._
-import org.apache.spark.SparkContext._
 import org.apache.spark.annotation.{DeveloperApi, Experimental}
 import org.apache.spark.api.java.JavaRDD
-import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.partial.BoundedDouble
 import org.apache.spark.partial.CountEvaluator
 import org.apache.spark.partial.GroupedCountEvaluator
 import org.apache.spark.partial.PartialResult
 import org.apache.spark.storage.StorageLevel
-import org.apache.spark.util.{BoundedPriorityQueue, Utils, CallSite}
+import org.apache.spark.util.{BoundedPriorityQueue, Utils}
 import org.apache.spark.util.collection.OpenHashMap
 import org.apache.spark.util.random.{BernoulliSampler, PoissonSampler, BernoulliCellSampler,
   SamplingUtils}
@@ -58,8 +56,9 @@ import org.apache.spark.util.random.{BernoulliSampler, PoissonSampler, Bernoulli
  * Doubles; and
  * [[org.apache.spark.rdd.SequenceFileRDDFunctions]] contains operations available on RDDs that
  * can be saved as SequenceFiles.
- * These operations are automatically available on any RDD of the right type (e.g. RDD[(Int, Int)]
- * through implicit conversions when you `import org.apache.spark.SparkContext._`.
+ * All operations are automatically available on any RDD of the right type (e.g. RDD[(Int, Int)]
+ * through implicit conversions except `saveAsSequenceFile`. You need to
+ * `import org.apache.spark.SparkContext._` to make `saveAsSequenceFile` work.
  *
  * Internally, each RDD is characterized by five main properties:
  *
@@ -270,19 +269,30 @@ abstract class RDD[T: ClassTag](
   /**
    * Return a new RDD by applying a function to all elements of this RDD.
    */
-  def map[U: ClassTag](f: T => U): RDD[U] = new MappedRDD(this, sc.clean(f))
+  def map[U: ClassTag](f: T => U): RDD[U] = {
+    val cleanF = sc.clean(f)
+    new MapPartitionsRDD[U, T](this, (context, pid, iter) => iter.map(cleanF))
+  }
 
   /**
    *  Return a new RDD by first applying a function to all elements of this
    *  RDD, and then flattening the results.
    */
-  def flatMap[U: ClassTag](f: T => TraversableOnce[U]): RDD[U] =
-    new FlatMappedRDD(this, sc.clean(f))
+  def flatMap[U: ClassTag](f: T => TraversableOnce[U]): RDD[U] = {
+    val cleanF = sc.clean(f)
+    new MapPartitionsRDD[U, T](this, (context, pid, iter) => iter.flatMap(cleanF))
+  }
 
   /**
    * Return a new RDD containing only the elements that satisfy a predicate.
    */
-  def filter(f: T => Boolean): RDD[T] = new FilteredRDD(this, sc.clean(f))
+  def filter(f: T => Boolean): RDD[T] = {
+    val cleanF = sc.clean(f)
+    new MapPartitionsRDD[T, T](
+      this,
+      (context, pid, iter) => iter.filter(cleanF),
+      preservesPartitioning = true)
+  }
 
   /**
    * Return a new RDD containing the distinct elements in this RDD.
@@ -503,7 +513,9 @@ abstract class RDD[T: ClassTag](
   /**
    * Return an RDD created by coalescing all elements within each partition into an array.
    */
-  def glom(): RDD[Array[T]] = new GlommedRDD(this)
+  def glom(): RDD[Array[T]] = {
+    new MapPartitionsRDD[Array[T], T](this, (context, pid, iter) => Iterator(iter.toArray))
+  }
 
   /**
    * Return the Cartesian product of this RDD and another one, that is, the RDD of all pairs of
@@ -1386,6 +1398,13 @@ abstract class RDD[T: ClassTag](
   }
 }
 
+
+/**
+ * Defines implicit functions that provide extra functionalities on RDDs of specific types.
+ *
+ * For example, [[RDD.rddToPairRDDFunctions]] converts an RDD into a [[PairRDDFunctions]] for
+ * key-value-pair RDDs, and enabling extra functionalities such as [[PairRDDFunctions.reduceByKey]].
+ */
 object RDD {
 
   // The following implicit functions were in SparkContext before 1.2 and users had to
@@ -1394,22 +1413,30 @@ object RDD {
   // compatibility and forward to the following functions directly.
 
   implicit def rddToPairRDDFunctions[K, V](rdd: RDD[(K, V)])
-      (implicit kt: ClassTag[K], vt: ClassTag[V], ord: Ordering[K] = null) = {
+    (implicit kt: ClassTag[K], vt: ClassTag[V], ord: Ordering[K] = null): PairRDDFunctions[K, V] = {
     new PairRDDFunctions(rdd)
   }
 
-  implicit def rddToAsyncRDDActions[T: ClassTag](rdd: RDD[T]) = new AsyncRDDActions(rdd)
+  implicit def rddToAsyncRDDActions[T: ClassTag](rdd: RDD[T]): AsyncRDDActions[T] = {
+    new AsyncRDDActions(rdd)
+  }
 
   implicit def rddToSequenceFileRDDFunctions[K <% Writable: ClassTag, V <% Writable: ClassTag](
-      rdd: RDD[(K, V)]) =
+      rdd: RDD[(K, V)]): SequenceFileRDDFunctions[K, V] = {
     new SequenceFileRDDFunctions(rdd)
+  }
 
-  implicit def rddToOrderedRDDFunctions[K : Ordering : ClassTag, V: ClassTag](
-      rdd: RDD[(K, V)]) =
+  implicit def rddToOrderedRDDFunctions[K : Ordering : ClassTag, V: ClassTag](rdd: RDD[(K, V)])
+    : OrderedRDDFunctions[K, V, (K, V)] = {
     new OrderedRDDFunctions[K, V, (K, V)](rdd)
+  }
 
-  implicit def doubleRDDToDoubleRDDFunctions(rdd: RDD[Double]) = new DoubleRDDFunctions(rdd)
+  implicit def doubleRDDToDoubleRDDFunctions(rdd: RDD[Double]): DoubleRDDFunctions = {
+    new DoubleRDDFunctions(rdd)
+  }
 
-  implicit def numericRDDToDoubleRDDFunctions[T](rdd: RDD[T])(implicit num: Numeric[T]) =
+  implicit def numericRDDToDoubleRDDFunctions[T](rdd: RDD[T])(implicit num: Numeric[T])
+    : DoubleRDDFunctions = {
     new DoubleRDDFunctions(rdd.map(x => num.toDouble(x)))
+  }
 }
diff --git a/core/src/main/scala/org/apache/spark/rdd/SequenceFileRDDFunctions.scala b/core/src/main/scala/org/apache/spark/rdd/SequenceFileRDDFunctions.scala
index 9a1efc83cbe6a..2b48916951430 100644
--- a/core/src/main/scala/org/apache/spark/rdd/SequenceFileRDDFunctions.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/SequenceFileRDDFunctions.scala
@@ -24,7 +24,6 @@ import org.apache.hadoop.mapred.JobConf
 import org.apache.hadoop.mapred.SequenceFileOutputFormat
 
 import org.apache.spark.Logging
-import org.apache.spark.SparkContext._
 
 /**
  * Extra functions available on RDDs of (key, value) pairs to create a Hadoop SequenceFile,
diff --git a/core/src/main/scala/org/apache/spark/rdd/ZippedPartitionsRDD.scala b/core/src/main/scala/org/apache/spark/rdd/ZippedPartitionsRDD.scala
index 996f2cd3f34a3..95b2dd954e9f4 100644
--- a/core/src/main/scala/org/apache/spark/rdd/ZippedPartitionsRDD.scala
+++ b/core/src/main/scala/org/apache/spark/rdd/ZippedPartitionsRDD.scala
@@ -77,7 +77,7 @@ private[spark] abstract class ZippedPartitionsBaseRDD[V: ClassTag](
 
 private[spark] class ZippedPartitionsRDD2[A: ClassTag, B: ClassTag, V: ClassTag](
     sc: SparkContext,
-    f: (Iterator[A], Iterator[B]) => Iterator[V],
+    var f: (Iterator[A], Iterator[B]) => Iterator[V],
     var rdd1: RDD[A],
     var rdd2: RDD[B],
     preservesPartitioning: Boolean = false)
@@ -92,13 +92,14 @@ private[spark] class ZippedPartitionsRDD2[A: ClassTag, B: ClassTag, V: ClassTag]
     super.clearDependencies()
     rdd1 = null
     rdd2 = null
+    f = null
   }
 }
 
 private[spark] class ZippedPartitionsRDD3
   [A: ClassTag, B: ClassTag, C: ClassTag, V: ClassTag](
     sc: SparkContext,
-    f: (Iterator[A], Iterator[B], Iterator[C]) => Iterator[V],
+    var f: (Iterator[A], Iterator[B], Iterator[C]) => Iterator[V],
     var rdd1: RDD[A],
     var rdd2: RDD[B],
     var rdd3: RDD[C],
@@ -117,13 +118,14 @@ private[spark] class ZippedPartitionsRDD3
     rdd1 = null
     rdd2 = null
     rdd3 = null
+    f = null
   }
 }
 
 private[spark] class ZippedPartitionsRDD4
   [A: ClassTag, B: ClassTag, C: ClassTag, D:ClassTag, V: ClassTag](
     sc: SparkContext,
-    f: (Iterator[A], Iterator[B], Iterator[C], Iterator[D]) => Iterator[V],
+    var f: (Iterator[A], Iterator[B], Iterator[C], Iterator[D]) => Iterator[V],
     var rdd1: RDD[A],
     var rdd2: RDD[B],
     var rdd3: RDD[C],
@@ -145,5 +147,6 @@ private[spark] class ZippedPartitionsRDD4
     rdd2 = null
     rdd3 = null
     rdd4 = null
+    f = null
   }
 }
diff --git a/core/src/main/scala/org/apache/spark/shuffle/ShuffleMemoryManager.scala b/core/src/main/scala/org/apache/spark/shuffle/ShuffleMemoryManager.scala
index ee91a368b76ea..3bcc7178a3d8b 100644
--- a/core/src/main/scala/org/apache/spark/shuffle/ShuffleMemoryManager.scala
+++ b/core/src/main/scala/org/apache/spark/shuffle/ShuffleMemoryManager.scala
@@ -66,8 +66,9 @@ private[spark] class ShuffleMemoryManager(maxMemory: Long) extends Logging {
       val curMem = threadMemory(threadId)
       val freeMemory = maxMemory - threadMemory.values.sum
 
-      // How much we can grant this thread; don't let it grow to more than 1 / numActiveThreads
-      val maxToGrant = math.min(numBytes, (maxMemory / numActiveThreads) - curMem)
+      // How much we can grant this thread; don't let it grow to more than 1 / numActiveThreads;
+      // don't let it be negative
+      val maxToGrant = math.min(numBytes, math.max(0, (maxMemory / numActiveThreads) - curMem))
 
       if (curMem < maxMemory / (2 * numActiveThreads)) {
         // We want to let each thread get at least 1 / (2 * numActiveThreads) before blocking;
diff --git a/core/src/main/scala/org/apache/spark/storage/BlockManagerMasterActor.scala b/core/src/main/scala/org/apache/spark/storage/BlockManagerMasterActor.scala
index 685b2e11440fb..9cbda41223a8b 100644
--- a/core/src/main/scala/org/apache/spark/storage/BlockManagerMasterActor.scala
+++ b/core/src/main/scala/org/apache/spark/storage/BlockManagerMasterActor.scala
@@ -73,9 +73,8 @@ class BlockManagerMasterActor(val isLocal: Boolean, conf: SparkConf, listenerBus
 
     case UpdateBlockInfo(
       blockManagerId, blockId, storageLevel, deserializedSize, size, tachyonSize) =>
-      // TODO: Ideally we want to handle all the message replies in receive instead of in the
-      // individual private methods.
-      updateBlockInfo(blockManagerId, blockId, storageLevel, deserializedSize, size, tachyonSize)
+      sender ! updateBlockInfo(
+        blockManagerId, blockId, storageLevel, deserializedSize, size, tachyonSize)
 
     case GetLocations(blockId) =>
       sender ! getLocations(blockId)
@@ -355,23 +354,21 @@ class BlockManagerMasterActor(val isLocal: Boolean, conf: SparkConf, listenerBus
       storageLevel: StorageLevel,
       memSize: Long,
       diskSize: Long,
-      tachyonSize: Long) {
+      tachyonSize: Long): Boolean = {
 
     if (!blockManagerInfo.contains(blockManagerId)) {
       if (blockManagerId.isDriver && !isLocal) {
         // We intentionally do not register the master (except in local mode),
         // so we should not indicate failure.
-        sender ! true
+        return true
       } else {
-        sender ! false
+        return false
       }
-      return
     }
 
     if (blockId == null) {
       blockManagerInfo(blockManagerId).updateLastSeenMs()
-      sender ! true
-      return
+      return true
     }
 
     blockManagerInfo(blockManagerId).updateBlockInfo(
@@ -395,7 +392,7 @@ class BlockManagerMasterActor(val isLocal: Boolean, conf: SparkConf, listenerBus
     if (locations.size == 0) {
       blockLocations.remove(blockId)
     }
-    sender ! true
+    true
   }
 
   private def getLocations(blockId: BlockId): Seq[BlockManagerId] = {
diff --git a/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala b/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
index 83170f7c5a4ab..2499c11a65b0e 100644
--- a/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
+++ b/core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
@@ -17,6 +17,7 @@
 
 package org.apache.spark.storage
 
+import java.io.{InputStream, IOException}
 import java.util.concurrent.LinkedBlockingQueue
 
 import scala.collection.mutable.{ArrayBuffer, HashSet, Queue}
@@ -289,17 +290,22 @@ final class ShuffleBlockFetcherIterator(
     }
 
     val iteratorTry: Try[Iterator[Any]] = result match {
-      case FailureFetchResult(_, e) => Failure(e)
-      case SuccessFetchResult(blockId, _, buf) => {
-        val is = blockManager.wrapForCompression(blockId, buf.createInputStream())
-        val iter = serializer.newInstance().deserializeStream(is).asIterator
-        Success(CompletionIterator[Any, Iterator[Any]](iter, {
-          // Once the iterator is exhausted, release the buffer and set currentResult to null
-          // so we don't release it again in cleanup.
-          currentResult = null
-          buf.release()
-        }))
-      }
+      case FailureFetchResult(_, e) =>
+        Failure(e)
+      case SuccessFetchResult(blockId, _, buf) =>
+        // There is a chance that createInputStream can fail (e.g. fetching a local file that does
+        // not exist, SPARK-4085). In that case, we should propagate the right exception so
+        // the scheduler gets a FetchFailedException.
+        Try(buf.createInputStream()).map { is0 =>
+          val is = blockManager.wrapForCompression(blockId, is0)
+          val iter = serializer.newInstance().deserializeStream(is).asIterator
+          CompletionIterator[Any, Iterator[Any]](iter, {
+            // Once the iterator is exhausted, release the buffer and set currentResult to null
+            // so we don't release it again in cleanup.
+            currentResult = null
+            buf.release()
+          })
+        }
     }
 
     (result.blockId, iteratorTry)
diff --git a/core/src/main/scala/org/apache/spark/ui/UIUtils.scala b/core/src/main/scala/org/apache/spark/ui/UIUtils.scala
index 09079bbd43f6f..315327c3c6b7c 100644
--- a/core/src/main/scala/org/apache/spark/ui/UIUtils.scala
+++ b/core/src/main/scala/org/apache/spark/ui/UIUtils.scala
@@ -212,6 +212,11 @@ private[spark] object UIUtils extends Logging {
           </div>
           {content}
         </div>
+        <div id="footer">
+          <div class="container-fluid">
+            <p class="muted credit">Spark {org.apache.spark.SPARK_VERSION}</p>
+          </div>
+        </div>
       </body>
     </html>
   }
@@ -238,6 +243,11 @@ private[spark] object UIUtils extends Logging {
           </div>
           {content}
         </div>
+        <div id="footer">
+          <div class="container-fluid">
+            <p class="muted credit">Spark {org.apache.spark.SPARK_VERSION}</p>
+          </div>
+        </div>
       </body>
     </html>
   }
diff --git a/core/src/main/scala/org/apache/spark/ui/UIWorkloadGenerator.scala b/core/src/main/scala/org/apache/spark/ui/UIWorkloadGenerator.scala
index 18d2b5075aa08..b4677447c8872 100644
--- a/core/src/main/scala/org/apache/spark/ui/UIWorkloadGenerator.scala
+++ b/core/src/main/scala/org/apache/spark/ui/UIWorkloadGenerator.scala
@@ -20,7 +20,6 @@ package org.apache.spark.ui
 import scala.util.Random
 
 import org.apache.spark.{SparkConf, SparkContext}
-import org.apache.spark.SparkContext._
 import org.apache.spark.scheduler.SchedulingMode
 
 /**
diff --git a/core/src/main/scala/org/apache/spark/util/AkkaUtils.scala b/core/src/main/scala/org/apache/spark/util/AkkaUtils.scala
index 10010bdfa1a51..8c2457f56bffe 100644
--- a/core/src/main/scala/org/apache/spark/util/AkkaUtils.scala
+++ b/core/src/main/scala/org/apache/spark/util/AkkaUtils.scala
@@ -134,9 +134,16 @@ private[spark] object AkkaUtils extends Logging {
     Duration.create(conf.getLong("spark.akka.lookupTimeout", 30), "seconds")
   }
 
+  private val AKKA_MAX_FRAME_SIZE_IN_MB = Int.MaxValue / 1024 / 1024
+
   /** Returns the configured max frame size for Akka messages in bytes. */
   def maxFrameSizeBytes(conf: SparkConf): Int = {
-    conf.getInt("spark.akka.frameSize", 10) * 1024 * 1024
+    val frameSizeInMB = conf.getInt("spark.akka.frameSize", 10)
+    if (frameSizeInMB > AKKA_MAX_FRAME_SIZE_IN_MB) {
+      throw new IllegalArgumentException("spark.akka.frameSize should not be greater than "
+        + AKKA_MAX_FRAME_SIZE_IN_MB + "MB")
+    }
+    frameSizeInMB * 1024 * 1024
   }
 
   /** Space reserved for extra data in an Akka message besides serialized task or task result. */
diff --git a/core/src/main/scala/org/apache/spark/util/collection/CompactBuffer.scala b/core/src/main/scala/org/apache/spark/util/collection/CompactBuffer.scala
index d44e15e3c97ea..4d43d8d5cc8d8 100644
--- a/core/src/main/scala/org/apache/spark/util/collection/CompactBuffer.scala
+++ b/core/src/main/scala/org/apache/spark/util/collection/CompactBuffer.scala
@@ -17,6 +17,8 @@
 
 package org.apache.spark.util.collection
 
+import scala.reflect.ClassTag
+
 /**
  * An append-only buffer similar to ArrayBuffer, but more memory-efficient for small buffers.
  * ArrayBuffer always allocates an Object array to store the data, with 16 entries by default,
@@ -25,7 +27,7 @@ package org.apache.spark.util.collection
  * entries than that. This makes it more efficient for operations like groupBy where we expect
  * some keys to have very few elements.
  */
-private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
+private[spark] class CompactBuffer[T: ClassTag] extends Seq[T] with Serializable {
   // First two elements
   private var element0: T = _
   private var element1: T = _
@@ -34,7 +36,7 @@ private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
   private var curSize = 0
 
   // Array for extra elements
-  private var otherElements: Array[AnyRef] = null
+  private var otherElements: Array[T] = null
 
   def apply(position: Int): T = {
     if (position < 0 || position >= curSize) {
@@ -45,7 +47,7 @@ private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
     } else if (position == 1) {
       element1
     } else {
-      otherElements(position - 2).asInstanceOf[T]
+      otherElements(position - 2)
     }
   }
 
@@ -58,7 +60,7 @@ private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
     } else if (position == 1) {
       element1 = value
     } else {
-      otherElements(position - 2) = value.asInstanceOf[AnyRef]
+      otherElements(position - 2) = value
     }
   }
 
@@ -72,7 +74,7 @@ private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
       curSize = 2
     } else {
       growToSize(curSize + 1)
-      otherElements(newIndex - 2) = value.asInstanceOf[AnyRef]
+      otherElements(newIndex - 2) = value
     }
     this
   }
@@ -139,7 +141,7 @@ private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
           newArrayLen = Int.MaxValue - 2
         }
       }
-      val newArray = new Array[AnyRef](newArrayLen)
+      val newArray = new Array[T](newArrayLen)
       if (otherElements != null) {
         System.arraycopy(otherElements, 0, newArray, 0, otherElements.length)
       }
@@ -150,9 +152,9 @@ private[spark] class CompactBuffer[T] extends Seq[T] with Serializable {
 }
 
 private[spark] object CompactBuffer {
-  def apply[T](): CompactBuffer[T] = new CompactBuffer[T]
+  def apply[T: ClassTag](): CompactBuffer[T] = new CompactBuffer[T]
 
-  def apply[T](value: T): CompactBuffer[T] = {
+  def apply[T: ClassTag](value: T): CompactBuffer[T] = {
     val buf = new CompactBuffer[T]
     buf += value
   }
diff --git a/core/src/main/scala/org/apache/spark/util/random/StratifiedSamplingUtils.scala b/core/src/main/scala/org/apache/spark/util/random/StratifiedSamplingUtils.scala
index 4fa357edd6f07..2ae308dacf1ae 100644
--- a/core/src/main/scala/org/apache/spark/util/random/StratifiedSamplingUtils.scala
+++ b/core/src/main/scala/org/apache/spark/util/random/StratifiedSamplingUtils.scala
@@ -25,7 +25,6 @@ import scala.reflect.ClassTag
 import org.apache.commons.math3.distribution.PoissonDistribution
 
 import org.apache.spark.Logging
-import org.apache.spark.SparkContext._
 import org.apache.spark.rdd.RDD
 
 /**
diff --git a/core/src/test/java/org/apache/spark/JavaAPISuite.java b/core/src/test/java/org/apache/spark/JavaAPISuite.java
index 59c86eecac5e8..3ad4f2f193af4 100644
--- a/core/src/test/java/org/apache/spark/JavaAPISuite.java
+++ b/core/src/test/java/org/apache/spark/JavaAPISuite.java
@@ -323,6 +323,47 @@ public Boolean call(Integer x) {
     Assert.assertEquals(5, Iterables.size(oddsAndEvens.lookup(false).get(0))); // Odds
   }
 
+  @Test
+  public void groupByOnPairRDD() {
+    // Regression test for SPARK-4459
+    JavaRDD<Integer> rdd = sc.parallelize(Arrays.asList(1, 1, 2, 3, 5, 8, 13));
+    Function<Tuple2<Integer, Integer>, Boolean> areOdd =
+      new Function<Tuple2<Integer, Integer>, Boolean>() {
+        @Override
+        public Boolean call(Tuple2<Integer, Integer> x) {
+          return (x._1() % 2 == 0) && (x._2() % 2 == 0);
+        }
+      };
+    JavaPairRDD<Integer, Integer> pairRDD = rdd.zip(rdd);
+    JavaPairRDD<Boolean, Iterable<Tuple2<Integer, Integer>>> oddsAndEvens = pairRDD.groupBy(areOdd);
+    Assert.assertEquals(2, oddsAndEvens.count());
+    Assert.assertEquals(2, Iterables.size(oddsAndEvens.lookup(true).get(0)));  // Evens
+    Assert.assertEquals(5, Iterables.size(oddsAndEvens.lookup(false).get(0))); // Odds
+
+    oddsAndEvens = pairRDD.groupBy(areOdd, 1);
+    Assert.assertEquals(2, oddsAndEvens.count());
+    Assert.assertEquals(2, Iterables.size(oddsAndEvens.lookup(true).get(0)));  // Evens
+    Assert.assertEquals(5, Iterables.size(oddsAndEvens.lookup(false).get(0))); // Odds
+  }
+
+  @SuppressWarnings("unchecked")
+  @Test
+  public void keyByOnPairRDD() {
+    // Regression test for SPARK-4459
+    JavaRDD<Integer> rdd = sc.parallelize(Arrays.asList(1, 1, 2, 3, 5, 8, 13));
+    Function<Tuple2<Integer, Integer>, String> sumToString =
+      new Function<Tuple2<Integer, Integer>, String>() {
+        @Override
+        public String call(Tuple2<Integer, Integer> x) {
+          return String.valueOf(x._1() + x._2());
+        }
+      };
+    JavaPairRDD<Integer, Integer> pairRDD = rdd.zip(rdd);
+    JavaPairRDD<String, Tuple2<Integer, Integer>> keyed = pairRDD.keyBy(sumToString);
+    Assert.assertEquals(7, keyed.count());
+    Assert.assertEquals(1, (long) keyed.lookup("2").get(0)._1());
+  }
+
   @SuppressWarnings("unchecked")
   @Test
   public void cogroup() {
diff --git a/core/src/test/scala/org/apache/spark/AccumulatorSuite.scala b/core/src/test/scala/org/apache/spark/AccumulatorSuite.scala
index 52d1d5277658e..f087fc550dde3 100644
--- a/core/src/test/scala/org/apache/spark/AccumulatorSuite.scala
+++ b/core/src/test/scala/org/apache/spark/AccumulatorSuite.scala
@@ -22,7 +22,6 @@ import scala.collection.mutable
 import org.scalatest.FunSuite
 import org.scalatest.Matchers
 
-import org.apache.spark.SparkContext._
 
 class AccumulatorSuite extends FunSuite with Matchers with LocalSparkContext {
 
diff --git a/core/src/test/scala/org/apache/spark/CheckpointSuite.scala b/core/src/test/scala/org/apache/spark/CheckpointSuite.scala
index a41914a1a9d0c..3b10b3a042317 100644
--- a/core/src/test/scala/org/apache/spark/CheckpointSuite.scala
+++ b/core/src/test/scala/org/apache/spark/CheckpointSuite.scala
@@ -23,7 +23,6 @@ import scala.reflect.ClassTag
 
 import org.scalatest.FunSuite
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.rdd._
 import org.apache.spark.storage.{BlockId, StorageLevel, TestBlockId}
 import org.apache.spark.util.Utils
diff --git a/core/src/test/scala/org/apache/spark/ContextCleanerSuite.scala b/core/src/test/scala/org/apache/spark/ContextCleanerSuite.scala
index 2e3fc5ef0e336..ae2ae7ed0d3aa 100644
--- a/core/src/test/scala/org/apache/spark/ContextCleanerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/ContextCleanerSuite.scala
@@ -28,7 +28,6 @@ import org.scalatest.concurrent.{PatienceConfiguration, Eventually}
 import org.scalatest.concurrent.Eventually._
 import org.scalatest.time.SpanSugar._
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.rdd.RDD
 import org.apache.spark.storage._
 import org.apache.spark.shuffle.hash.HashShuffleManager
diff --git a/core/src/test/scala/org/apache/spark/DistributedSuite.scala b/core/src/test/scala/org/apache/spark/DistributedSuite.scala
index 429199f2075c6..998f3008ec0ea 100644
--- a/core/src/test/scala/org/apache/spark/DistributedSuite.scala
+++ b/core/src/test/scala/org/apache/spark/DistributedSuite.scala
@@ -23,7 +23,6 @@ import org.scalatest.concurrent.Timeouts._
 import org.scalatest.Matchers
 import org.scalatest.time.{Millis, Span}
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.storage.{RDDBlockId, StorageLevel}
 
 class NotSerializableClass
diff --git a/core/src/test/scala/org/apache/spark/ExternalShuffleServiceSuite.scala b/core/src/test/scala/org/apache/spark/ExternalShuffleServiceSuite.scala
index 55799f55146cb..bac6fdbcdc976 100644
--- a/core/src/test/scala/org/apache/spark/ExternalShuffleServiceSuite.scala
+++ b/core/src/test/scala/org/apache/spark/ExternalShuffleServiceSuite.scala
@@ -17,11 +17,8 @@
 
 package org.apache.spark
 
-import java.util.concurrent.atomic.AtomicInteger
-
 import org.scalatest.BeforeAndAfterAll
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.network.TransportContext
 import org.apache.spark.network.netty.SparkTransportConf
 import org.apache.spark.network.server.TransportServer
diff --git a/core/src/test/scala/org/apache/spark/FailureSuite.scala b/core/src/test/scala/org/apache/spark/FailureSuite.scala
index 2229e6acc425d..1212d0b43207d 100644
--- a/core/src/test/scala/org/apache/spark/FailureSuite.scala
+++ b/core/src/test/scala/org/apache/spark/FailureSuite.scala
@@ -19,7 +19,6 @@ package org.apache.spark
 
 import org.scalatest.FunSuite
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.util.NonSerializable
 
 import java.io.NotSerializableException
diff --git a/core/src/test/scala/org/apache/spark/FileServerSuite.scala b/core/src/test/scala/org/apache/spark/FileServerSuite.scala
index 379c2a6ea4b55..49426545c767e 100644
--- a/core/src/test/scala/org/apache/spark/FileServerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/FileServerSuite.scala
@@ -23,7 +23,6 @@ import java.util.jar.{JarEntry, JarOutputStream}
 import com.google.common.io.ByteStreams
 import org.scalatest.FunSuite
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.util.Utils
 
 class FileServerSuite extends FunSuite with LocalSparkContext {
diff --git a/core/src/test/scala/org/apache/spark/FutureActionSuite.scala b/core/src/test/scala/org/apache/spark/FutureActionSuite.scala
index db9c25fc457a4..f5cdb01ec9504 100644
--- a/core/src/test/scala/org/apache/spark/FutureActionSuite.scala
+++ b/core/src/test/scala/org/apache/spark/FutureActionSuite.scala
@@ -22,7 +22,6 @@ import scala.concurrent.duration.Duration
 
 import org.scalatest.{BeforeAndAfter, FunSuite, Matchers}
 
-import org.apache.spark.SparkContext._
 
 class FutureActionSuite extends FunSuite with BeforeAndAfter with Matchers with LocalSparkContext {
 
diff --git a/core/src/test/scala/org/apache/spark/ImplicitOrderingSuite.scala b/core/src/test/scala/org/apache/spark/ImplicitOrderingSuite.scala
index 8e4a9e2c9f56c..d895230ecf330 100644
--- a/core/src/test/scala/org/apache/spark/ImplicitOrderingSuite.scala
+++ b/core/src/test/scala/org/apache/spark/ImplicitOrderingSuite.scala
@@ -20,7 +20,6 @@ package org.apache.spark
 import org.scalatest.FunSuite
 
 import org.apache.spark.rdd.RDD
-import org.apache.spark.SparkContext._
 
 class ImplicitOrderingSuite extends FunSuite with LocalSparkContext {
   // Tests that PairRDDFunctions grabs an implicit Ordering in various cases where it should.
diff --git a/core/src/test/scala/org/apache/spark/JobCancellationSuite.scala b/core/src/test/scala/org/apache/spark/JobCancellationSuite.scala
index a57430e829ced..41ed2bce55ce1 100644
--- a/core/src/test/scala/org/apache/spark/JobCancellationSuite.scala
+++ b/core/src/test/scala/org/apache/spark/JobCancellationSuite.scala
@@ -27,7 +27,6 @@ import scala.concurrent.future
 import org.scalatest.{BeforeAndAfter, FunSuite}
 import org.scalatest.Matchers
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.scheduler.{SparkListener, SparkListenerTaskStart}
 
 /**
diff --git a/core/src/test/scala/org/apache/spark/PartitioningSuite.scala b/core/src/test/scala/org/apache/spark/PartitioningSuite.scala
index 646ede30ae6ff..b7532314ada01 100644
--- a/core/src/test/scala/org/apache/spark/PartitioningSuite.scala
+++ b/core/src/test/scala/org/apache/spark/PartitioningSuite.scala
@@ -22,7 +22,6 @@ import scala.math.abs
 
 import org.scalatest.{FunSuite, PrivateMethodTester}
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.rdd.RDD
 import org.apache.spark.util.StatCounter
 
diff --git a/core/src/test/scala/org/apache/spark/ShuffleSuite.scala b/core/src/test/scala/org/apache/spark/ShuffleSuite.scala
index 85e5f9ab444b3..58a96245a9b53 100644
--- a/core/src/test/scala/org/apache/spark/ShuffleSuite.scala
+++ b/core/src/test/scala/org/apache/spark/ShuffleSuite.scala
@@ -20,10 +20,10 @@ package org.apache.spark
 import org.scalatest.FunSuite
 import org.scalatest.Matchers
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.ShuffleSuite.NonJavaSerializableClass
 import org.apache.spark.rdd.{CoGroupedRDD, OrderedRDDFunctions, RDD, ShuffledRDD, SubtractedRDD}
 import org.apache.spark.serializer.KryoSerializer
+import org.apache.spark.storage.{ShuffleDataBlockId, ShuffleBlockId}
 import org.apache.spark.util.MutablePair
 
 abstract class ShuffleSuite extends FunSuite with Matchers with LocalSparkContext {
@@ -264,13 +264,34 @@ abstract class ShuffleSuite extends FunSuite with Matchers with LocalSparkContex
       }
     }
   }
+
+  test("[SPARK-4085] rerun map stage if reduce stage cannot find its local shuffle file") {
+    val myConf = conf.clone().set("spark.test.noStageRetry", "false")
+    sc = new SparkContext("local", "test", myConf)
+    val rdd = sc.parallelize(1 to 10, 2).map((_, 1)).reduceByKey(_ + _)
+    rdd.count()
+
+    // Delete one of the local shuffle blocks.
+    val hashFile = sc.env.blockManager.diskBlockManager.getFile(new ShuffleBlockId(0, 0, 0))
+    val sortFile = sc.env.blockManager.diskBlockManager.getFile(new ShuffleDataBlockId(0, 0, 0))
+    assert(hashFile.exists() || sortFile.exists())
+
+    if (hashFile.exists()) {
+      hashFile.delete()
+    }
+    if (sortFile.exists()) {
+      sortFile.delete()
+    }
+
+    // This count should retry the execution of the previous stage and rerun shuffle.
+    rdd.count()
+  }
 }
 
 object ShuffleSuite {
 
   def mergeCombineException(x: Int, y: Int): Int = {
     throw new SparkException("Exception for map-side combine.")
-    x + y
   }
 
   class NonJavaSerializableClass(val value: Int) extends Comparable[NonJavaSerializableClass] {
diff --git a/core/src/test/scala/org/apache/spark/StatusTrackerSuite.scala b/core/src/test/scala/org/apache/spark/StatusTrackerSuite.scala
index 8577e4ac7e33e..41d6ea29d5b06 100644
--- a/core/src/test/scala/org/apache/spark/StatusTrackerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/StatusTrackerSuite.scala
@@ -25,7 +25,6 @@ import org.scalatest.{Matchers, FunSuite}
 import org.scalatest.concurrent.Eventually._
 
 import org.apache.spark.JobExecutionStatus._
-import org.apache.spark.SparkContext._
 
 class StatusTrackerSuite extends FunSuite with Matchers with LocalSparkContext {
 
diff --git a/core/src/test/scala/org/apache/spark/rdd/AsyncRDDActionsSuite.scala b/core/src/test/scala/org/apache/spark/rdd/AsyncRDDActionsSuite.scala
index 3b833f2e41867..f2b0ea1063a72 100644
--- a/core/src/test/scala/org/apache/spark/rdd/AsyncRDDActionsSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rdd/AsyncRDDActionsSuite.scala
@@ -27,7 +27,6 @@ import org.scalatest.{BeforeAndAfterAll, FunSuite}
 import org.scalatest.concurrent.Timeouts
 import org.scalatest.time.SpanSugar._
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.{SparkContext, SparkException, LocalSparkContext}
 
 class AsyncRDDActionsSuite extends FunSuite with BeforeAndAfterAll with Timeouts {
diff --git a/core/src/test/scala/org/apache/spark/rdd/DoubleRDDSuite.scala b/core/src/test/scala/org/apache/spark/rdd/DoubleRDDSuite.scala
index f89bdb6e07dea..de306533752c1 100644
--- a/core/src/test/scala/org/apache/spark/rdd/DoubleRDDSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rdd/DoubleRDDSuite.scala
@@ -20,7 +20,6 @@ package org.apache.spark.rdd
 import org.scalatest.FunSuite
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 
 class DoubleRDDSuite extends FunSuite with SharedSparkContext {
   // Verify tests on the histogram functionality. We test with both evenly
diff --git a/core/src/test/scala/org/apache/spark/rdd/PairRDDFunctionsSuite.scala b/core/src/test/scala/org/apache/spark/rdd/PairRDDFunctionsSuite.scala
index 3620e251cc139..108f70af43f37 100644
--- a/core/src/test/scala/org/apache/spark/rdd/PairRDDFunctionsSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rdd/PairRDDFunctionsSuite.scala
@@ -29,7 +29,6 @@ import org.apache.hadoop.mapreduce.{JobContext => NewJobContext, OutputCommitter
 OutputFormat => NewOutputFormat, RecordWriter => NewRecordWriter,
 TaskAttemptContext => NewTaskAttempContext}
 import org.apache.spark.{Partitioner, SharedSparkContext}
-import org.apache.spark.SparkContext._
 import org.apache.spark.util.Utils
 
 import org.scalatest.FunSuite
diff --git a/core/src/test/scala/org/apache/spark/rdd/RDDSuite.scala b/core/src/test/scala/org/apache/spark/rdd/RDDSuite.scala
index e079ca3b1e896..46fcb80fa1845 100644
--- a/core/src/test/scala/org/apache/spark/rdd/RDDSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rdd/RDDSuite.scala
@@ -24,11 +24,9 @@ import scala.reflect.ClassTag
 import org.scalatest.FunSuite
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
-import org.apache.spark.util.Utils
-
 import org.apache.spark.api.java.{JavaRDD, JavaSparkContext}
 import org.apache.spark.rdd.RDDSuiteUtils._
+import org.apache.spark.util.Utils
 
 class RDDSuite extends FunSuite with SharedSparkContext {
 
@@ -38,8 +36,8 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     assert(nums.toLocalIterator.toList === List(1, 2, 3, 4))
     val dups = sc.makeRDD(Array(1, 1, 2, 2, 3, 3, 4, 4), 2)
     assert(dups.distinct().count() === 4)
-    assert(dups.distinct.count === 4)  // Can distinct and count be called without parentheses?
-    assert(dups.distinct.collect === dups.distinct().collect)
+    assert(dups.distinct().count === 4)  // Can distinct and count be called without parentheses?
+    assert(dups.distinct().collect === dups.distinct().collect)
     assert(dups.distinct(2).collect === dups.distinct().collect)
     assert(nums.reduce(_ + _) === 10)
     assert(nums.fold(0)(_ + _) === 10)
@@ -97,7 +95,6 @@ class RDDSuite extends FunSuite with SharedSparkContext {
   }
 
   test("partitioner aware union") {
-    import SparkContext._
     def makeRDDWithPartitioner(seq: Seq[Int]) = {
       sc.makeRDD(seq, 1)
         .map(x => (x, null))
@@ -619,9 +616,9 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     for(seed <- 1 to 5) {
       val splits = data.randomSplit(Array(1.0, 2.0, 3.0), seed)
       assert(splits.size == 3, "wrong number of splits")
-      assert(splits.flatMap(_.collect).sorted.toList == data.collect.toList,
+      assert(splits.flatMap(_.collect()).sorted.toList == data.collect().toList,
         "incomplete or wrong split")
-      val s = splits.map(_.count)
+      val s = splits.map(_.count())
       assert(math.abs(s(0) - 100) < 50) // std =  9.13
       assert(math.abs(s(1) - 200) < 50) // std = 11.55
       assert(math.abs(s(2) - 300) < 50) // std = 12.25
@@ -764,8 +761,8 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     val rdd3 = rdd2.map(_ + 1)
     val rdd4 = new UnionRDD(sc, List(rdd1, rdd2, rdd3))
     assert(rdd4.parent(0).isInstanceOf[ParallelCollectionRDD[_]])
-    assert(rdd4.parent(1).isInstanceOf[FilteredRDD[_]])
-    assert(rdd4.parent(2).isInstanceOf[MappedRDD[_, _]])
+    assert(rdd4.parent[Int](1) === rdd2)
+    assert(rdd4.parent[Int](2) === rdd3)
   }
 
   test("getNarrowAncestors") {
@@ -783,20 +780,18 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     // Simple dependency tree with a single branch
     assert(ancestors1.size === 0)
     assert(ancestors2.size === 2)
-    assert(ancestors2.count(_.isInstanceOf[ParallelCollectionRDD[_]]) === 1)
-    assert(ancestors2.count(_.isInstanceOf[FilteredRDD[_]]) === 1)
+    assert(ancestors2.count(_ === rdd1) === 1)
+    assert(ancestors2.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 1)
     assert(ancestors3.size === 5)
-    assert(ancestors3.count(_.isInstanceOf[ParallelCollectionRDD[_]]) === 1)
-    assert(ancestors3.count(_.isInstanceOf[FilteredRDD[_]]) === 2)
-    assert(ancestors3.count(_.isInstanceOf[MappedRDD[_, _]]) === 2)
+    assert(ancestors3.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 4)
 
     // Any ancestors before the shuffle are not considered
     assert(ancestors4.size === 0)
     assert(ancestors4.count(_.isInstanceOf[ShuffledRDD[_, _, _]]) === 0)
     assert(ancestors5.size === 3)
     assert(ancestors5.count(_.isInstanceOf[ShuffledRDD[_, _, _]]) === 1)
-    assert(ancestors5.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 0)
-    assert(ancestors5.count(_.isInstanceOf[MappedValuesRDD[_, _, _]]) === 2)
+    assert(ancestors5.count(_ === rdd3) === 0)
+    assert(ancestors5.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 2)
   }
 
   test("getNarrowAncestors with multiple parents") {
@@ -817,16 +812,16 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     // Simple dependency tree with multiple branches
     assert(ancestors6.size === 3)
     assert(ancestors6.count(_.isInstanceOf[ParallelCollectionRDD[_]]) === 2)
-    assert(ancestors6.count(_.isInstanceOf[MappedRDD[_, _]]) === 1)
+    assert(ancestors6.count(_ === rdd2) === 1)
     assert(ancestors7.size === 5)
     assert(ancestors7.count(_.isInstanceOf[ParallelCollectionRDD[_]]) === 3)
-    assert(ancestors7.count(_.isInstanceOf[MappedRDD[_, _]]) === 1)
-    assert(ancestors7.count(_.isInstanceOf[FilteredRDD[_]]) === 1)
+    assert(ancestors7.count(_ === rdd2) === 1)
+    assert(ancestors7.count(_ === rdd3) === 1)
 
     // Dependency tree with duplicate nodes (e.g. rdd1 should not be reported twice)
     assert(ancestors8.size === 7)
-    assert(ancestors8.count(_.isInstanceOf[MappedRDD[_, _]]) === 1)
-    assert(ancestors8.count(_.isInstanceOf[FilteredRDD[_]]) === 1)
+    assert(ancestors8.count(_ === rdd2) === 1)
+    assert(ancestors8.count(_ === rdd3) === 1)
     assert(ancestors8.count(_.isInstanceOf[UnionRDD[_]]) === 2)
     assert(ancestors8.count(_.isInstanceOf[ParallelCollectionRDD[_]]) === 3)
     assert(ancestors8.count(_ == rdd1) === 1)
@@ -836,7 +831,6 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     // Any ancestors before the shuffle are not considered
     assert(ancestors9.size === 2)
     assert(ancestors9.count(_.isInstanceOf[CoGroupedRDD[_]]) === 1)
-    assert(ancestors9.count(_.isInstanceOf[MappedValuesRDD[_, _, _]]) === 1)
   }
 
   /**
@@ -870,12 +864,10 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     val ancestors3 = rdd3.getNarrowAncestors
     val ancestors4 = rdd4.getNarrowAncestors
     assert(ancestors3.size === 4)
-    assert(ancestors3.count(_.isInstanceOf[MappedRDD[_, _]]) === 2)
-    assert(ancestors3.count(_.isInstanceOf[FilteredRDD[_]]) === 2)
+    assert(ancestors3.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 4)
     assert(ancestors3.count(_ == rdd3) === 0)
     assert(ancestors4.size === 4)
-    assert(ancestors4.count(_.isInstanceOf[MappedRDD[_, _]]) === 2)
-    assert(ancestors4.count(_.isInstanceOf[FilteredRDD[_]]) === 1)
+    assert(ancestors4.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 3)
     assert(ancestors4.count(_.isInstanceOf[CyclicalDependencyRDD[_]]) === 1)
     assert(ancestors4.count(_ == rdd3) === 1)
     assert(ancestors4.count(_ == rdd4) === 0)
@@ -883,8 +875,7 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     // Cycles that do not involve the root
     val ancestors5 = rdd5.getNarrowAncestors
     assert(ancestors5.size === 6)
-    assert(ancestors5.count(_.isInstanceOf[MappedRDD[_, _]]) === 3)
-    assert(ancestors5.count(_.isInstanceOf[FilteredRDD[_]]) === 2)
+    assert(ancestors5.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 5)
     assert(ancestors5.count(_.isInstanceOf[CyclicalDependencyRDD[_]]) === 1)
     assert(ancestors4.count(_ == rdd3) === 1)
 
@@ -892,8 +883,7 @@ class RDDSuite extends FunSuite with SharedSparkContext {
     val ancestors6 = rdd6.getNarrowAncestors
     assert(ancestors6.size === 12)
     assert(ancestors6.count(_.isInstanceOf[UnionRDD[_]]) === 2)
-    assert(ancestors6.count(_.isInstanceOf[MappedRDD[_, _]]) === 4)
-    assert(ancestors6.count(_.isInstanceOf[FilteredRDD[_]]) === 3)
+    assert(ancestors6.count(_.isInstanceOf[MapPartitionsRDD[_, _]]) === 7)
     assert(ancestors6.count(_.isInstanceOf[CyclicalDependencyRDD[_]]) === 3)
   }
 
diff --git a/core/src/test/scala/org/apache/spark/rdd/SortingSuite.scala b/core/src/test/scala/org/apache/spark/rdd/SortingSuite.scala
index 656917628f7a8..a40f2ffeffdf9 100644
--- a/core/src/test/scala/org/apache/spark/rdd/SortingSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rdd/SortingSuite.scala
@@ -21,7 +21,6 @@ import org.scalatest.FunSuite
 import org.scalatest.Matchers
 
 import org.apache.spark.{Logging, SharedSparkContext}
-import org.apache.spark.SparkContext._
 
 class SortingSuite extends FunSuite with SharedSparkContext with Matchers with Logging {
 
diff --git a/core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala b/core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala
index bdd721dc7eaf7..436eea4f1fdcf 100644
--- a/core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala
@@ -27,7 +27,6 @@ import org.scalatest.concurrent.Timeouts
 import org.scalatest.time.SpanSugar._
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 import org.apache.spark.rdd.RDD
 import org.apache.spark.scheduler.SchedulingMode.SchedulingMode
 import org.apache.spark.storage.{BlockId, BlockManagerId, BlockManagerMaster}
diff --git a/core/src/test/scala/org/apache/spark/scheduler/ReplayListenerSuite.scala b/core/src/test/scala/org/apache/spark/scheduler/ReplayListenerSuite.scala
index e05f373392d4a..90bdfe07f61c9 100644
--- a/core/src/test/scala/org/apache/spark/scheduler/ReplayListenerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/scheduler/ReplayListenerSuite.scala
@@ -22,7 +22,6 @@ import java.io.{File, PrintWriter}
 import org.json4s.jackson.JsonMethods._
 import org.scalatest.{BeforeAndAfter, FunSuite}
 
-import org.apache.spark.SparkContext._
 import org.apache.spark.{SparkConf, SparkContext}
 import org.apache.spark.deploy.SparkHadoopUtil
 import org.apache.spark.io.CompressionCodec
diff --git a/core/src/test/scala/org/apache/spark/scheduler/SparkListenerSuite.scala b/core/src/test/scala/org/apache/spark/scheduler/SparkListenerSuite.scala
index abe0dc35b07e2..b276343cb412c 100644
--- a/core/src/test/scala/org/apache/spark/scheduler/SparkListenerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/scheduler/SparkListenerSuite.scala
@@ -25,7 +25,6 @@ import org.scalatest.{BeforeAndAfter, BeforeAndAfterAll, FunSuite}
 import org.scalatest.Matchers
 
 import org.apache.spark.{LocalSparkContext, SparkContext}
-import org.apache.spark.SparkContext._
 import org.apache.spark.executor.TaskMetrics
 
 class SparkListenerSuite extends FunSuite with LocalSparkContext with Matchers
diff --git a/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala b/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala
index 11e8c9c4cb37f..855f1b6276089 100644
--- a/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala
+++ b/core/src/test/scala/org/apache/spark/serializer/KryoSerializerDistributedSuite.scala
@@ -23,7 +23,6 @@ import com.esotericsoftware.kryo.Kryo
 import org.scalatest.FunSuite
 
 import org.apache.spark.{LocalSparkContext, SparkConf, SparkContext, SparkEnv, TestUtils}
-import org.apache.spark.SparkContext._
 import org.apache.spark.serializer.KryoDistributedTest._
 
 class KryoSerializerDistributedSuite extends FunSuite {
diff --git a/core/src/test/scala/org/apache/spark/shuffle/ShuffleMemoryManagerSuite.scala b/core/src/test/scala/org/apache/spark/shuffle/ShuffleMemoryManagerSuite.scala
index d31bc22ee74f7..e0e646f0a3652 100644
--- a/core/src/test/scala/org/apache/spark/shuffle/ShuffleMemoryManagerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/shuffle/ShuffleMemoryManagerSuite.scala
@@ -159,7 +159,7 @@ class ShuffleMemoryManagerSuite extends FunSuite with Timeouts {
 
   test("threads can block to get at least 1 / 2N memory") {
     // t1 grabs 1000 bytes and then waits until t2 is ready to make a request. It sleeps
-    // for a bit and releases 250 bytes, which should then be greanted to t2. Further requests
+    // for a bit and releases 250 bytes, which should then be granted to t2. Further requests
     // by t2 will return false right away because it now has 1 / 2N of the memory.
 
     val manager = new ShuffleMemoryManager(1000L)
@@ -291,4 +291,19 @@ class ShuffleMemoryManagerSuite extends FunSuite with Timeouts {
       assert(state.t2WaitTime > 200, s"t2 waited less than 200 ms (${state.t2WaitTime})")
     }
   }
+
+  test("threads should not be granted a negative size") {
+    val manager = new ShuffleMemoryManager(1000L)
+    manager.tryToAcquire(700L)
+
+    val latch = new CountDownLatch(1)
+    startThread("t1") {
+      manager.tryToAcquire(300L)
+      latch.countDown()
+    }
+    latch.await() // Wait until `t1` calls `tryToAcquire`
+
+    val granted = manager.tryToAcquire(300L)
+    assert(0 === granted, "granted is negative")
+  }
 }
diff --git a/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala b/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala
index d2857b8b55664..787f4c2b5a8b2 100644
--- a/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala
+++ b/core/src/test/scala/org/apache/spark/ui/UISeleniumSuite.scala
@@ -27,7 +27,6 @@ import org.scalatest.selenium.WebBrowser
 import org.scalatest.time.SpanSugar._
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 import org.apache.spark.LocalSparkContext._
 import org.apache.spark.api.java.StorageLevels
 import org.apache.spark.shuffle.FetchFailedException
diff --git a/core/src/test/scala/org/apache/spark/util/collection/ExternalAppendOnlyMapSuite.scala b/core/src/test/scala/org/apache/spark/util/collection/ExternalAppendOnlyMapSuite.scala
index 511d76c9144cc..48f79ea651018 100644
--- a/core/src/test/scala/org/apache/spark/util/collection/ExternalAppendOnlyMapSuite.scala
+++ b/core/src/test/scala/org/apache/spark/util/collection/ExternalAppendOnlyMapSuite.scala
@@ -22,7 +22,6 @@ import scala.collection.mutable.ArrayBuffer
 import org.scalatest.FunSuite
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 import org.apache.spark.io.CompressionCodec
 
 class ExternalAppendOnlyMapSuite extends FunSuite with LocalSparkContext {
diff --git a/core/src/test/scala/org/apache/spark/util/collection/ExternalSorterSuite.scala b/core/src/test/scala/org/apache/spark/util/collection/ExternalSorterSuite.scala
index 3cb42d416de4f..72d96798b1141 100644
--- a/core/src/test/scala/org/apache/spark/util/collection/ExternalSorterSuite.scala
+++ b/core/src/test/scala/org/apache/spark/util/collection/ExternalSorterSuite.scala
@@ -22,7 +22,6 @@ import scala.collection.mutable.ArrayBuffer
 import org.scalatest.{PrivateMethodTester, FunSuite}
 
 import org.apache.spark._
-import org.apache.spark.SparkContext._
 
 import scala.util.Random
 
diff --git a/dev/create-release/create-release.sh b/dev/create-release/create-release.sh
index e0aca467ac949..3b89aaba29609 100755
--- a/dev/create-release/create-release.sh
+++ b/dev/create-release/create-release.sh
@@ -39,7 +39,6 @@ RC_NAME=${RC_NAME:-rc2}
 M2_REPO=~/.m2/repository
 SPARK_REPO=$M2_REPO/org/apache/spark
 NEXUS_ROOT=https://repository.apache.org/service/local/staging
-NEXUS_UPLOAD=$NEXUS_ROOT/deploy/maven2
 NEXUS_PROFILE=d63f592e7eac0 # Profile for Spark staging uploads
 
 if [ -z "$JAVA_HOME" ]; then
@@ -64,19 +63,28 @@ if [[ ! "$@" =~ --package-only ]]; then
   # NOTE: This is done "eagerly" i.e. we don't check if we can succesfully build
   # or before we coin the release commit. This helps avoid races where
   # other people add commits to this branch while we are in the middle of building.
-  old="  <version>${RELEASE_VERSION}-SNAPSHOT<\/version>"
-  new="  <version>${RELEASE_VERSION}<\/version>"
-  find . -name pom.xml -o -name package.scala | grep -v dev | xargs -I {} sed -i \
-    -e "s/$old/$new/" {}
+  cur_ver="${RELEASE_VERSION}-SNAPSHOT"
+  rel_ver="${RELEASE_VERSION}"
+  next_ver="${NEXT_VERSION}-SNAPSHOT"
+
+  old="^\( \{2,4\}\)<version>${cur_ver}<\/version>$"
+  new="\1<version>${rel_ver}<\/version>"
+  find . -name pom.xml | grep -v dev | xargs -I {} sed -i \
+    -e "s/${old}/${new}/" {}
+  find . -name package.scala | grep -v dev | xargs -I {} sed -i \
+    -e "s/${old}/${new}/" {}
+
   git commit -a -m "Preparing Spark release $GIT_TAG"
   echo "Creating tag $GIT_TAG at the head of $GIT_BRANCH"
   git tag $GIT_TAG
 
-  old="  <version>${RELEASE_VERSION}<\/version>"
-  new="  <version>${NEXT_VERSION}-SNAPSHOT<\/version>"
-  find . -name pom.xml -o -name package.scala | grep -v dev | xargs -I {} sed -i \
+  old="^\( \{2,4\}\)<version>${rel_ver}<\/version>$"
+  new="\1<version>${next_ver}<\/version>"
+  find . -name pom.xml | grep -v dev | xargs -I {} sed -i \
     -e "s/$old/$new/" {}
-  git commit -a -m "Preparing development version ${NEXT_VERSION}-SNAPSHOT"
+  find . -name package.scala | grep -v dev | xargs -I {} sed -i \
+    -e "s/${old}/${new}/" {}
+  git commit -a -m "Preparing development version $next_ver"
   git push origin $GIT_TAG
   git push origin HEAD:$GIT_BRANCH
   git checkout -f $GIT_TAG 
@@ -118,12 +126,13 @@ if [[ ! "$@" =~ --package-only ]]; then
     gpg --print-md SHA1 $file > $file.sha1
   done
 
-  echo "Uplading files to $NEXUS_UPLOAD"
+  nexus_upload=$NEXUS_ROOT/deployByRepositoryId/$staged_repo_id
+  echo "Uplading files to $nexus_upload"
   for file in $(find . -type f)
   do
     # strip leading ./
     file_short=$(echo $file | sed -e "s/\.\///")
-    dest_url="$NEXUS_UPLOAD/org/apache/spark/$file_short"
+    dest_url="$nexus_upload/org/apache/spark/$file_short"
     echo "  Uploading $file_short"
     curl -u $ASF_USERNAME:$ASF_PASSWORD --upload-file $file_short $dest_url
   done
diff --git a/dev/create-release/generate-contributors.py b/dev/create-release/generate-contributors.py
index f4bf734081583..a3b78a3eac6d0 100755
--- a/dev/create-release/generate-contributors.py
+++ b/dev/create-release/generate-contributors.py
@@ -29,20 +29,6 @@
 START_COMMIT = os.environ.get("START_COMMIT", "37b100")
 END_COMMIT = os.environ.get("END_COMMIT", "3693ae")
 
-try:
-    from jira.client import JIRA
-except ImportError:
-    print "This tool requires the jira-python library"
-    print "Install using 'sudo pip install jira-python'"
-    sys.exit(-1)
-
-try:
-    import unidecode
-except ImportError:
-    print "This tool requires the unidecode library to decode obscure github usernames"
-    print "Install using 'sudo pip install unidecode'"
-    sys.exit(-1)
-
 # If commit range is not specified, prompt the user to provide it
 if not START_COMMIT or not END_COMMIT:
     print "A commit range is required to proceed."
@@ -107,13 +93,17 @@ def print_indented(_list):
     if reverts: print "Reverts (%d)" % len(reverts); print_indented(reverts)
     if nojiras: print "No JIRA (%d)" % len(nojiras); print_indented(nojiras)
     print "==================== Warning: the above commits will be ignored ==================\n"
-response = raw_input("%d commits left to process. Ok to proceed? [y/N] " % len(filtered_commits))
-if response.lower() != "y":
+response = raw_input("%d commits left to process. Ok to proceed? [Y/n] " % len(filtered_commits))
+if response.lower() != "y" and response:
     sys.exit("Ok, exiting.")
 
 # Keep track of warnings to tell the user at the end
 warnings = []
 
+# Mapping from the invalid author name to its associated JIRA issues
+# E.g. andrewor14 -> set("SPARK-2413", "SPARK-3551", "SPARK-3471")
+invalid_authors = {}
+
 # Populate a map that groups issues and components by author
 # It takes the form: Author name -> { Contribution type -> Spark components }
 # For instance,
@@ -130,13 +120,22 @@ def print_indented(_list):
 #
 author_info = {}
 jira_options = { "server": JIRA_API_BASE }
-jira = JIRA(jira_options)
+jira_client = JIRA(options = jira_options)
 print "\n=========================== Compiling contributor list ==========================="
 for commit in filtered_commits:
     commit_hash = re.findall("^[a-z0-9]+", commit)[0]
     issues = re.findall("SPARK-[0-9]+", commit.upper())
     author = get_author(commit_hash)
-    author = unidecode.unidecode(unicode(author, "UTF-8")) # guard against special characters
+    author = unidecode.unidecode(unicode(author, "UTF-8")).strip() # guard against special characters
+    # If the author name is invalid, keep track of it along
+    # with all associated issues so we can translate it later
+    if is_valid_author(author):
+        author = capitalize_author(author)
+    else:
+        if author not in invalid_authors:
+            invalid_authors[author] = set()
+        for issue in issues:
+            invalid_authors[author].add(issue)
     date = get_date(commit_hash)
     # Parse components from the commit message, if any
     commit_components = find_components(commit, commit_hash)
@@ -147,11 +146,11 @@ def populate(issue_type, components):
             author_info[author] = {}
         if issue_type not in author_info[author]:
             author_info[author][issue_type] = set()
-        for component in all_components:
+        for component in components:
             author_info[author][issue_type].add(component)
     # Find issues and components associated with this commit
     for issue in issues:
-        jira_issue = jira.issue(issue)
+        jira_issue = jira_client.issue(issue)
         jira_type = jira_issue.fields.issuetype.name
         jira_type = translate_issue_type(jira_type, issue, warnings)
         jira_components = [translate_component(c.name, commit_hash, warnings)\
@@ -168,7 +167,6 @@ def populate(issue_type, components):
 # Each line takes the format "Author name - semi-colon delimited contributions"
 # e.g. Andrew Or - Bug fixes in Windows, Core, and Web UI; improvements in Core
 # e.g. Tathagata Das - Bug fixes and new features in Streaming
-contributors_file_name = "contributors.txt"
 contributors_file = open(contributors_file_name, "w")
 authors = author_info.keys()
 authors.sort()
@@ -192,11 +190,23 @@ def populate(issue_type, components):
     # Do not use python's capitalize() on the whole string to preserve case
     assert contribution
     contribution = contribution[0].capitalize() + contribution[1:]
+    # If the author name is invalid, use an intermediate format that
+    # can be translated through translate-contributors.py later
+    # E.g. andrewor14/SPARK-3425/SPARK-1157/SPARK-6672
+    if author in invalid_authors and invalid_authors[author]:
+        author = author + "/" + "/".join(invalid_authors[author])
     line = "%s - %s" % (author, contribution)
     contributors_file.write(line + "\n")
 contributors_file.close()
 print "Contributors list is successfully written to %s!" % contributors_file_name
 
+# Prompt the user to translate author names if necessary
+if invalid_authors:
+    warnings.append("Found the following invalid authors:")
+    for a in invalid_authors:
+        warnings.append("\t%s" % a)
+    warnings.append("Please run './translate-contributors.py' to translate them.")
+
 # Log any warnings encountered in the process
 if warnings:
     print "\n============ Warnings encountered while creating the contributor list ============"
diff --git a/dev/create-release/releaseutils.py b/dev/create-release/releaseutils.py
index e56d7fa58fa2c..76a10c32886d4 100755
--- a/dev/create-release/releaseutils.py
+++ b/dev/create-release/releaseutils.py
@@ -21,6 +21,32 @@
 import re
 from subprocess import Popen, PIPE
 
+try:
+    from jira.client import JIRA
+    from jira.exceptions import JIRAError
+except ImportError:
+    print "This tool requires the jira-python library"
+    print "Install using 'sudo pip install jira-python'"
+    sys.exit(-1)
+
+try:
+    from github import Github
+    from github import GithubException
+except ImportError:
+    print "This tool requires the PyGithub library"
+    print "Install using 'sudo pip install PyGithub'"
+    sys.exit(-1)
+
+try:
+    import unidecode
+except ImportError:
+    print "This tool requires the unidecode library to decode obscure github usernames"
+    print "Install using 'sudo pip install unidecode'"
+    sys.exit(-1)
+
+# Contributors list file name
+contributors_file_name = "contributors.txt"
+
 # Utility functions run git commands (written with Git 1.8.5)
 def run_cmd(cmd): return Popen(cmd, stdout=PIPE).communicate()[0]
 def get_author(commit_hash):
@@ -46,7 +72,8 @@ def num_commits_in_range(start_hash, end_hash):
     "build": "build fixes",
     "improvement": "improvements",
     "new feature": "new features",
-    "documentation": "documentation"
+    "documentation": "documentation",
+    "test": "test"
 }
 
 # Maintain a mapping for translating component names when creating the release notes
@@ -122,3 +149,40 @@ def nice_join(str_list):
     else:
         return ", ".join(str_list[:-1]) + ", and " + str_list[-1]
 
+# Return the full name of the specified user on Github
+# If the user doesn't exist, return None
+def get_github_name(author, github_client):
+    if github_client:
+        try:
+            return github_client.get_user(author).name
+        except GithubException as e:
+            # If this is not a "not found" exception
+            if e.status != 404:
+                raise e
+    return None
+
+# Return the full name of the specified user on JIRA
+# If the user doesn't exist, return None
+def get_jira_name(author, jira_client):
+    if jira_client:
+        try:
+            return jira_client.user(author).displayName
+        except JIRAError as e:
+            # If this is not a "not found" exception
+            if e.status_code != 404:
+                raise e
+    return None
+
+# Return whether the given name is in the form <First Name><space><Last Name>
+def is_valid_author(author):
+    if not author: return False
+    author_words = len(author.split(" "))
+    return author_words == 2 or author_words == 3
+
+# Capitalize the first letter of each word in the given author name
+def capitalize_author(author):
+    if not author: return None
+    words = author.split(" ")
+    words = [w[0].capitalize() + w[1:] for w in words if w]
+    return " ".join(words)
+
diff --git a/dev/create-release/translate-contributors.py b/dev/create-release/translate-contributors.py
new file mode 100755
index 0000000000000..ef4625b003cb6
--- /dev/null
+++ b/dev/create-release/translate-contributors.py
@@ -0,0 +1,190 @@
+#!/usr/bin/env python
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements.  See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License.  You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script translates invalid authors in the contributors list generated
+# by generate-contributors.py. When the script encounters an author name that
+# is considered invalid, it searches Github and JIRA in an attempt to search
+# for replacements. This tool runs in two modes:
+#
+# (1) Interactive mode: For each invalid author name, this script presents
+# all candidate replacements to the user and awaits user response. In this
+# mode, the user may also input a custom name. This is the default.
+#
+# (2) Non-interactive mode: For each invalid author name, this script replaces
+# the name with the first valid candidate it can find. If there is none, it
+# uses the original name. This can be enabled through the --non-interactive flag.
+
+import os
+import sys
+
+from releaseutils import *
+
+# You must set the following before use!
+JIRA_API_BASE = os.environ.get("JIRA_API_BASE", "https://issues.apache.org/jira")
+JIRA_USERNAME = os.environ.get("JIRA_USERNAME", None)
+JIRA_PASSWORD = os.environ.get("JIRA_PASSWORD", None)
+if not JIRA_USERNAME or not JIRA_PASSWORD:
+    sys.exit("Both JIRA_USERNAME and JIRA_PASSWORD must be set")
+
+# Write new contributors list to <old_file_name>.new
+if not os.path.isfile(contributors_file_name):
+    print "Contributors file %s does not exist!" % contributors_file_name
+    print "Have you run ./generate-contributors.py yet?"
+    sys.exit(1)
+contributors_file = open(contributors_file_name, "r")
+new_contributors_file_name = contributors_file_name + ".new"
+new_contributors_file = open(new_contributors_file_name, "w")
+warnings = []
+
+# In non-interactive mode, this script will choose the first replacement that is valid
+INTERACTIVE_MODE = True
+if len(sys.argv) > 1:
+    options = set(sys.argv[1:])
+    if "--non-interactive" in options:
+        INTERACTIVE_MODE = False
+if INTERACTIVE_MODE:
+    print "Running in interactive mode. To disable this, provide the --non-interactive flag."
+
+# Setup Github and JIRA clients
+jira_options = { "server": JIRA_API_BASE }
+jira_client = JIRA(options = jira_options, basic_auth = (JIRA_USERNAME, JIRA_PASSWORD))
+github_client = Github()
+
+# Generate candidates for the given author. This should only be called if the given author
+# name does not represent a full name as this operation is somewhat expensive. Under the
+# hood, it makes several calls to the Github and JIRA API servers to find the candidates.
+#
+# This returns a list of (candidate name, source) 2-tuples. E.g.
+# [
+#   (NOT_FOUND, "No full name found for Github user andrewor14"),
+#   ("Andrew Or", "Full name of JIRA user andrewor14"),
+#   ("Andrew Orso", "Full name of SPARK-1444 assignee andrewor14"),
+#   ("Andrew Ordall", "Full name of SPARK-1663 assignee andrewor14"),
+#   (NOT_FOUND, "No assignee found for SPARK-1763")
+# ]
+NOT_FOUND = "Not found"
+def generate_candidates(author, issues):
+    candidates = []
+    # First check for full name of Github user
+    github_name = get_github_name(new_author, github_client)
+    if github_name:
+        candidates.append((github_name, "Full name of Github user %s" % new_author))
+    else:
+        candidates.append((NOT_FOUND, "No full name found for Github user %s" % new_author))
+    # Then do the same for JIRA user
+    jira_name = get_jira_name(new_author, jira_client)
+    if jira_name:
+        candidates.append((jira_name, "Full name of JIRA user %s" % new_author))
+    else:
+        candidates.append((NOT_FOUND, "No full name found for JIRA user %s" % new_author))
+    # Then do the same for the assignee of each of the associated JIRAs
+    # Note that a given issue may not have an assignee, or the assignee may not have a full name
+    for issue in issues:
+        jira_issue = jira_client.issue(issue)
+        jira_assignee = jira_issue.fields.assignee
+        if jira_assignee:
+            user_name = jira_assignee.name
+            display_name = jira_assignee.displayName
+            if display_name:
+                candidates.append((display_name, "Full name of %s assignee %s" % (issue, user_name)))
+            else:
+                candidates.append((NOT_FOUND, "No full name found for %s assignee %" % (issue, user_name)))
+        else:
+            candidates.append((NOT_FOUND, "No assignee found for %s" % issue))
+    # Guard against special characters in candidate names
+    # Note that the candidate name may already be in unicode (JIRA returns this)
+    for i, (candidate, source) in enumerate(candidates):
+        try:
+            candidate = unicode(candidate, "UTF-8")
+        except TypeError:
+            # already in unicode
+            pass
+        candidate = unidecode.unidecode(candidate).strip()
+        candidates[i] = (candidate, source)
+    return candidates
+
+# Translate each invalid author by searching for possible candidates from Github and JIRA
+# In interactive mode, this script presents the user with a list of choices and have the user
+# select from this list. Additionally, the user may also choose to enter a custom name.
+# In non-interactive mode, this script picks the first valid author name from the candidates
+# If no such name exists, the original name is used (without the JIRA numbers).
+print "\n========================== Translating contributor list =========================="
+for line in contributors_file:
+    author = line.split(" - ")[0]
+    print "Processing author %s" % author
+    if not author:
+        print "    ERROR: Expected the following format <author> - <contributions>"
+        print "    ERROR: Actual = %s" % line
+    if not is_valid_author(author):
+        new_author = author.split("/")[0]
+        issues = author.split("/")[1:]
+        candidates = generate_candidates(new_author, issues)
+        # Print out potential replacement candidates along with the sources, e.g.
+        #   [X] No full name found for Github user andrewor14
+        #   [0] Andrew Or - Full name of JIRA user andrewor14
+        #   [1] Andrew Orso - Full name of SPARK-1444 assignee andrewor14
+        #   [2] Andrew Ordall - Full name of SPARK-1663 assignee andrewor14
+        #   [X] No assignee found for SPARK-1763
+        #   [3] Custom
+        candidate_names = []
+        for candidate, source in candidates:
+            if candidate == NOT_FOUND:
+                print "    [X] %s" % source
+            else:
+                index = len(candidate_names)
+                candidate_names.append(candidate)
+                print "    [%d] %s - %s" % (index, candidate, source)
+        custom_index = len(candidate_names)
+        # In interactive mode, additionally provide "custom" option and await user response
+        if INTERACTIVE_MODE:
+            print "    [%d] Custom" % custom_index
+            response = raw_input("    Your choice: ")
+            while not response.isdigit() or int(response) > custom_index:
+                response = raw_input("    Please enter an integer between 0 and %d: " % custom_index)
+            response = int(response)
+            if response == custom_index:
+                new_author = raw_input("    Please type a custom name for this author: ")
+            else:
+                new_author = candidate_names[response]
+        # In non-interactive mode, just pick the first candidate
+        else:
+            valid_candidate_names = [name for name, _ in candidates\
+                if is_valid_author(name) and name != NOT_FOUND]
+            if valid_candidate_names:
+                new_author = valid_candidate_names[0]
+        # Finally, capitalize the author and replace the original one with it
+        # If the final replacement is still invalid, log a warning
+        if is_valid_author(new_author):
+            new_author = capitalize_author(new_author)
+        else:
+            warnings.append("Unable to find a valid name %s for author %s" % (new_author, author))
+        print "    * Replacing %s with %s" % (author, new_author)
+        line = line.replace(author, new_author)
+    new_contributors_file.write(line)
+print "==================================================================================\n"
+contributors_file.close()
+new_contributors_file.close()
+
+print "Translated contributors list successfully written to %s!" % new_contributors_file_name
+
+# Log any warnings encountered in the process
+if warnings:
+    print "\n========== Warnings encountered while translating the contributor list ==========="
+    for w in warnings: print w
+    print "Please manually correct these in the final contributors list at %s." % new_contributors_file_name
+    print "==================================================================================\n"
+
diff --git a/dev/merge_spark_pr.py b/dev/merge_spark_pr.py
index 02ac20984add9..dfa924d2aa0ba 100755
--- a/dev/merge_spark_pr.py
+++ b/dev/merge_spark_pr.py
@@ -214,15 +214,10 @@ def fix_version_from_branch(branch, versions):
         return filter(lambda x: x.name.startswith(branch_ver), versions)[-1]
 
 
-def resolve_jira(title, merge_branches, comment):
+def resolve_jira_issue(merge_branches, comment, default_jira_id=""):
     asf_jira = jira.client.JIRA({'server': JIRA_API_BASE},
                                 basic_auth=(JIRA_USERNAME, JIRA_PASSWORD))
 
-    default_jira_id = ""
-    search = re.findall("SPARK-[0-9]{4,5}", title)
-    if len(search) > 0:
-        default_jira_id = search[0]
-
     jira_id = raw_input("Enter a JIRA id [%s]: " % default_jira_id)
     if jira_id == "":
         jira_id = default_jira_id
@@ -280,6 +275,15 @@ def get_version_json(version_str):
     print "Succesfully resolved %s with fixVersions=%s!" % (jira_id, fix_versions)
 
 
+def resolve_jira_issues(title, merge_branches, comment):
+    jira_ids = re.findall("SPARK-[0-9]{4,5}", title)
+
+    if len(jira_ids) == 0:
+        resolve_jira_issue(merge_branches, comment)
+    for jira_id in jira_ids:
+        resolve_jira_issue(merge_branches, comment, jira_id)
+
+
 branches = get_json("%s/branches" % GITHUB_API_BASE)
 branch_names = filter(lambda x: x.startswith("branch-"), [x['name'] for x in branches])
 # Assumes branch names can be sorted lexicographically
@@ -338,7 +342,7 @@ def get_version_json(version_str):
     if JIRA_USERNAME and JIRA_PASSWORD:
         continue_maybe("Would you like to update an associated JIRA?")
         jira_comment = "Issue resolved by pull request %s\n[%s/%s]" % (pr_num, GITHUB_BASE, pr_num)
-        resolve_jira(title, merged_refs, jira_comment)
+        resolve_jira_issues(title, merged_refs, jira_comment)
     else:
         print "JIRA_USERNAME and JIRA_PASSWORD not set"
         print "Exiting without trying to close the associated JIRA."
diff --git a/docs/building-spark.md b/docs/building-spark.md
index 40a47410e683a..6cca2da8e86d2 100644
--- a/docs/building-spark.md
+++ b/docs/building-spark.md
@@ -118,6 +118,7 @@ mvn -Pyarn -Phadoop-2.4 -Dhadoop.version=2.4.0 -Phive -Phive-0.12.0 -Phive-thrif
 # Building for Scala 2.11
 To produce a Spark package compiled with Scala 2.11, use the `-Dscala-2.11` property:
 
+    dev/change-version-to-2.11.sh
     mvn -Pyarn -Phadoop-2.4 -Dscala-2.11 -DskipTests clean package
 
 Scala 2.11 support in Spark is experimental and does not support a few features.
diff --git a/docs/configuration.md b/docs/configuration.md
index 0b77f5ab645c9..4b4bbea564d3a 100644
--- a/docs/configuration.md
+++ b/docs/configuration.md
@@ -98,7 +98,7 @@ of the most common options to set are:
   <td>(none)</td>
   <td>
     The cluster manager to connect to. See the list of
-    <a href="scala-programming-guide.html#master-urls"> allowed master URL's</a>.
+    <a href="submitting-applications.html#master-urls"> allowed master URL's</a>.
   </td>
 </tr>
 <tr>
diff --git a/docs/ec2-scripts.md b/docs/ec2-scripts.md
index 66bf5f1a855ed..ed51d0abb3a45 100644
--- a/docs/ec2-scripts.md
+++ b/docs/ec2-scripts.md
@@ -85,6 +85,11 @@ another.
      specified version of Spark. The `<version>` can be a version number
      (e.g. "0.7.3") or a specific git hash. By default, a recent
      version will be used.
+-    `--spark-git-repo=<repository url>` will let you run a custom version of
+     Spark that is built from the given git repository. By default, the
+     [Apache Github mirror](https://github.com/apache/spark) will be used.
+     When using a custom Spark version, `--spark-version` must be set to git
+     commit hash, such as 317e114, instead of a version number.
 -    If one of your launches fails due to e.g. not having the right
 permissions on your private key file, you can run `launch` with the
 `--resume` option to restart the setup process on an existing cluster.
diff --git a/docs/img/ml-Pipeline.png b/docs/img/ml-Pipeline.png
new file mode 100644
index 0000000000000..607928906bedd
Binary files /dev/null and b/docs/img/ml-Pipeline.png differ
diff --git a/docs/img/ml-PipelineModel.png b/docs/img/ml-PipelineModel.png
new file mode 100644
index 0000000000000..9ebc16719d365
Binary files /dev/null and b/docs/img/ml-PipelineModel.png differ
diff --git a/docs/img/ml-Pipelines.pptx b/docs/img/ml-Pipelines.pptx
new file mode 100644
index 0000000000000..1f773376abc7a
Binary files /dev/null and b/docs/img/ml-Pipelines.pptx differ
diff --git a/docs/ml-guide.md b/docs/ml-guide.md
new file mode 100644
index 0000000000000..1c2e27341473b
--- /dev/null
+++ b/docs/ml-guide.md
@@ -0,0 +1,702 @@
+---
+layout: global
+title: Spark ML Programming Guide
+---
+
+Spark ML is Spark's new machine learning package.  It is currently an alpha component but is potentially a successor to [MLlib](mllib-guide.html). The `spark.ml` package aims to replace the old APIs with a cleaner, more uniform set of APIs which will help users create full machine learning pipelines.
+
+MLlib vs. Spark ML:
+
+* Users can use algorithms from either of the two packages, but APIs may differ.  Currently, `spark.ml` offers a subset of the algorithms from `spark.mllib`. Since Spark ML is an alpha component, its API may change in future releases.
+* Developers should contribute new algorithms to `spark.mllib` and can optionally contribute to `spark.ml`.  See below for more details.
+* Spark ML only has Scala and Java APIs, whereas MLlib also has a Python API.
+
+**Table of Contents**
+
+* This will become a table of contents (this text will be scraped).
+{:toc}
+
+# Main Concepts
+
+Spark ML standardizes APIs for machine learning algorithms to make it easier to combine multiple algorithms into a single pipeline, or workflow.  This section covers the key concepts introduced by the Spark ML API.
+
+* **[ML Dataset](ml-guide.html#ml-dataset)**: Spark ML uses the [`SchemaRDD`](api/scala/index.html#org.apache.spark.sql.SchemaRDD) from Spark SQL as a dataset which can hold a variety of data types.
+E.g., a dataset could have different columns storing text, feature vectors, true labels, and predictions.
+
+* **[`Transformer`](ml-guide.html#transformers)**: A `Transformer` is an algorithm which can transform one `SchemaRDD` into another `SchemaRDD`.
+E.g., an ML model is a `Transformer` which transforms an RDD with features into an RDD with predictions.
+
+* **[`Estimator`](ml-guide.html#estimators)**: An `Estimator` is an algorithm which can be fit on a `SchemaRDD` to produce a `Transformer`.
+E.g., a learning algorithm is an `Estimator` which trains on a dataset and produces a model.
+
+* **[`Pipeline`](ml-guide.html#pipeline)**: A `Pipeline` chains multiple `Transformer`s and `Estimator`s together to specify an ML workflow.
+
+* **[`Param`](ml-guide.html#parameters)**: All `Transformer`s and `Estimator`s now share a common API for specifying parameters.
+
+## ML Dataset
+
+Machine learning can be applied to a wide variety of data types, such as vectors, text, images, and structured data.
+Spark ML adopts the [`SchemaRDD`](api/scala/index.html#org.apache.spark.sql.SchemaRDD) from Spark SQL in order to support a variety of data types under a unified Dataset concept.
+
+`SchemaRDD` supports many basic and structured types; see the [Spark SQL datatype reference](sql-programming-guide.html#spark-sql-datatype-reference) for a list of supported types.
+In addition to the types listed in the Spark SQL guide, `SchemaRDD` can use ML [`Vector`](api/scala/index.html#org.apache.spark.mllib.linalg.Vector) types.
+
+A `SchemaRDD` can be created either implicitly or explicitly from a regular `RDD`.  See the code examples below and the [Spark SQL programming guide](sql-programming-guide.html) for examples.
+
+Columns in a `SchemaRDD` are named.  The code examples below use names such as "text," "features," and "label."
+
+## ML Algorithms
+
+### Transformers
+
+A [`Transformer`](api/scala/index.html#org.apache.spark.ml.Transformer) is an abstraction which includes feature transformers and learned models.  Technically, a `Transformer` implements a method `transform()` which converts one `SchemaRDD` into another, generally by appending one or more columns.
+For example:
+
+* A feature transformer might take a dataset, read a column (e.g., text), convert it into a new column (e.g., feature vectors), append the new column to the dataset, and output the updated dataset.
+* A learning model might take a dataset, read the column containing feature vectors, predict the label for each feature vector, append the labels as a new column, and output the updated dataset.
+
+### Estimators
+
+An [`Estimator`](api/scala/index.html#org.apache.spark.ml.Estimator) abstracts the concept of a learning algorithm or any algorithm which fits or trains on data.  Technically, an `Estimator` implements a method `fit()` which accepts a `SchemaRDD` and produces a `Transformer`.
+For example, a learning algorithm such as `LogisticRegression` is an `Estimator`, and calling `fit()` trains a `LogisticRegressionModel`, which is a `Transformer`.
+
+### Properties of ML Algorithms
+
+`Transformer`s and `Estimator`s are both stateless.  In the future, stateful algorithms may be supported via alternative concepts.
+
+Each instance of a `Transformer` or `Estimator` has a unique ID, which is useful in specifying parameters (discussed below).
+
+## Pipeline
+
+In machine learning, it is common to run a sequence of algorithms to process and learn from data.
+E.g., a simple text document processing workflow might include several stages:
+
+* Split each document's text into words.
+* Convert each document's words into a numerical feature vector.
+* Learn a prediction model using the feature vectors and labels.
+
+Spark ML represents such a workflow as a [`Pipeline`](api/scala/index.html#org.apache.spark.ml.Pipeline),
+which consists of a sequence of [`PipelineStage`s](api/scala/index.html#org.apache.spark.ml.PipelineStage) (`Transformer`s and `Estimator`s) to be run in a specific order.  We will use this simple workflow as a running example in this section.
+
+### How It Works
+
+A `Pipeline` is specified as a sequence of stages, and each stage is either a `Transformer` or an `Estimator`.
+These stages are run in order, and the input dataset is modified as it passes through each stage.
+For `Transformer` stages, the `transform()` method is called on the dataset.
+For `Estimator` stages, the `fit()` method is called to produce a `Transformer` (which becomes part of the `PipelineModel`, or fitted `Pipeline`), and that `Transformer`'s `transform()` method is called on the dataset.
+
+We illustrate this for the simple text document workflow.  The figure below is for the *training time* usage of a `Pipeline`.
+
+<p style="text-align: center;">
+  <img
+    src="img/ml-Pipeline.png"
+    title="Spark ML Pipeline Example"
+    alt="Spark ML Pipeline Example"
+    width="80%"
+  />
+</p>
+
+Above, the top row represents a `Pipeline` with three stages.
+The first two (`Tokenizer` and `HashingTF`) are `Transformer`s (blue), and the third (`LogisticRegression`) is an `Estimator` (red).
+The bottom row represents data flowing through the pipeline, where cylinders indicate `SchemaRDD`s.
+The `Pipeline.fit()` method is called on the original dataset which has raw text documents and labels.
+The `Tokenizer.transform()` method splits the raw text documents into words, adding a new column with words into the dataset.
+The `HashingTF.transform()` method converts the words column into feature vectors, adding a new column with those vectors to the dataset.
+Now, since `LogisticRegression` is an `Estimator`, the `Pipeline` first calls `LogisticRegression.fit()` to produce a `LogisticRegressionModel`.
+If the `Pipeline` had more stages, it would call the `LogisticRegressionModel`'s `transform()` method on the dataset before passing the dataset to the next stage.
+
+A `Pipeline` is an `Estimator`.
+Thus, after a `Pipeline`'s `fit()` method runs, it produces a `PipelineModel` which is a `Transformer`.  This `PipelineModel` is used at *test time*; the figure below illustrates this usage.
+
+<p style="text-align: center;">
+  <img
+    src="img/ml-PipelineModel.png"
+    title="Spark ML PipelineModel Example"
+    alt="Spark ML PipelineModel Example"
+    width="80%"
+  />
+</p>
+
+In the figure above, the `PipelineModel` has the same number of stages as the original `Pipeline`, but all `Estimator`s in the original `Pipeline` have become `Transformer`s.
+When the `PipelineModel`'s `transform()` method is called on a test dataset, the data are passed through the `Pipeline` in order.
+Each stage's `transform()` method updates the dataset and passes it to the next stage.
+
+`Pipeline`s and `PipelineModel`s help to ensure that training and test data go through identical feature processing steps.
+
+### Details
+
+*DAG `Pipeline`s*: A `Pipeline`'s stages are specified as an ordered array.  The examples given here are all for linear `Pipeline`s, i.e., `Pipeline`s in which each stage uses data produced by the previous stage.  It is possible to create non-linear `Pipeline`s as long as the data flow graph forms a Directed Acyclic Graph (DAG).  This graph is currently specified implicitly based on the input and output column names of each stage (generally specified as parameters).  If the `Pipeline` forms a DAG, then the stages must be specified in topological order.
+
+*Runtime checking*: Since `Pipeline`s can operate on datasets with varied types, they cannot use compile-time type checking.  `Pipeline`s and `PipelineModel`s instead do runtime checking before actually running the `Pipeline`.  This type checking is done using the dataset *schema*, a description of the data types of columns in the `SchemaRDD`.
+
+## Parameters
+
+Spark ML `Estimator`s and `Transformer`s use a uniform API for specifying parameters.
+
+A [`Param`](api/scala/index.html#org.apache.spark.ml.param.Param) is a named parameter with self-contained documentation.
+A [`ParamMap`](api/scala/index.html#org.apache.spark.ml.param.ParamMap) is a set of (parameter, value) pairs.
+
+There are two main ways to pass parameters to an algorithm:
+
+1. Set parameters for an instance.  E.g., if `lr` is an instance of `LogisticRegression`, one could call `lr.setMaxIter(10)` to make `lr.fit()` use at most 10 iterations.  This API resembles the API used in MLlib.
+2. Pass a `ParamMap` to `fit()` or `transform()`.  Any parameters in the `ParamMap` will override parameters previously specified via setter methods.
+
+Parameters belong to specific instances of `Estimator`s and `Transformer`s.
+For example, if we have two `LogisticRegression` instances `lr1` and `lr2`, then we can build a `ParamMap` with both `maxIter` parameters specified: `ParamMap(lr1.maxIter -> 10, lr2.maxIter -> 20)`.
+This is useful if there are two algorithms with the `maxIter` parameter in a `Pipeline`.
+
+# Code Examples
+
+This section gives code examples illustrating the functionality discussed above.
+There is not yet documentation for specific algorithms in Spark ML.  For more info, please refer to the [API Documentation](api/scala/index.html#org.apache.spark.ml.package).  Spark ML algorithms are currently wrappers for MLlib algorithms, and the [MLlib programming guide](mllib-guide.html) has details on specific algorithms.
+
+## Example: Estimator, Transformer, and Param
+
+This example covers the concepts of `Estimator`, `Transformer`, and `Param`.
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.ml.classification.LogisticRegression
+import org.apache.spark.ml.param.ParamMap
+import org.apache.spark.mllib.linalg.{Vector, Vectors}
+import org.apache.spark.mllib.regression.LabeledPoint
+import org.apache.spark.sql.{Row, SQLContext}
+
+val conf = new SparkConf().setAppName("SimpleParamsExample")
+val sc = new SparkContext(conf)
+val sqlContext = new SQLContext(sc)
+import sqlContext._
+
+// Prepare training data.
+// We use LabeledPoint, which is a case class.  Spark SQL can convert RDDs of case classes
+// into SchemaRDDs, where it uses the case class metadata to infer the schema.
+val training = sparkContext.parallelize(Seq(
+  LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)),
+  LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)),
+  LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)),
+  LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5))))
+
+// Create a LogisticRegression instance.  This instance is an Estimator.
+val lr = new LogisticRegression()
+// Print out the parameters, documentation, and any default values.
+println("LogisticRegression parameters:\n" + lr.explainParams() + "\n")
+
+// We may set parameters using setter methods.
+lr.setMaxIter(10)
+  .setRegParam(0.01)
+
+// Learn a LogisticRegression model.  This uses the parameters stored in lr.
+val model1 = lr.fit(training)
+// Since model1 is a Model (i.e., a Transformer produced by an Estimator),
+// we can view the parameters it used during fit().
+// This prints the parameter (name: value) pairs, where names are unique IDs for this
+// LogisticRegression instance.
+println("Model 1 was fit using parameters: " + model1.fittingParamMap)
+
+// We may alternatively specify parameters using a ParamMap,
+// which supports several methods for specifying parameters.
+val paramMap = ParamMap(lr.maxIter -> 20)
+paramMap.put(lr.maxIter, 30) // Specify 1 Param.  This overwrites the original maxIter.
+paramMap.put(lr.regParam -> 0.1, lr.threshold -> 0.5) // Specify multiple Params.
+
+// One can also combine ParamMaps.
+val paramMap2 = ParamMap(lr.scoreCol -> "probability") // Changes output column name.
+val paramMapCombined = paramMap ++ paramMap2
+
+// Now learn a new model using the paramMapCombined parameters.
+// paramMapCombined overrides all parameters set earlier via lr.set* methods.
+val model2 = lr.fit(training, paramMapCombined)
+println("Model 2 was fit using parameters: " + model2.fittingParamMap)
+
+// Prepare test documents.
+val test = sparkContext.parallelize(Seq(
+  LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)),
+  LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)),
+  LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5))))
+
+// Make predictions on test documents using the Transformer.transform() method.
+// LogisticRegression.transform will only use the 'features' column.
+// Note that model2.transform() outputs a 'probability' column instead of the usual 'score'
+// column since we renamed the lr.scoreCol parameter previously.
+model2.transform(test)
+  .select('features, 'label, 'probability, 'prediction)
+  .collect()
+  .foreach { case Row(features: Vector, label: Double, prob: Double, prediction: Double) =>
+    println("(" + features + ", " + label + ") -> prob=" + prob + ", prediction=" + prediction)
+  }
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import java.util.List;
+import com.google.common.collect.Lists;
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.ml.classification.LogisticRegressionModel;
+import org.apache.spark.ml.param.ParamMap;
+import org.apache.spark.ml.classification.LogisticRegression;
+import org.apache.spark.mllib.linalg.Vectors;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.sql.api.java.JavaSQLContext;
+import org.apache.spark.sql.api.java.JavaSchemaRDD;
+import org.apache.spark.sql.api.java.Row;
+
+SparkConf conf = new SparkConf().setAppName("JavaSimpleParamsExample");
+JavaSparkContext jsc = new JavaSparkContext(conf);
+JavaSQLContext jsql = new JavaSQLContext(jsc);
+
+// Prepare training data.
+// We use LabeledPoint, which is a case class.  Spark SQL can convert RDDs of case classes
+// into SchemaRDDs, where it uses the case class metadata to infer the schema.
+List<LabeledPoint> localTraining = Lists.newArrayList(
+  new LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)),
+  new LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)),
+  new LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)),
+  new LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5)));
+JavaSchemaRDD training = jsql.applySchema(jsc.parallelize(localTraining), LabeledPoint.class);
+
+// Create a LogisticRegression instance.  This instance is an Estimator.
+LogisticRegression lr = new LogisticRegression();
+// Print out the parameters, documentation, and any default values.
+System.out.println("LogisticRegression parameters:\n" + lr.explainParams() + "\n");
+
+// We may set parameters using setter methods.
+lr.setMaxIter(10)
+  .setRegParam(0.01);
+
+// Learn a LogisticRegression model.  This uses the parameters stored in lr.
+LogisticRegressionModel model1 = lr.fit(training);
+// Since model1 is a Model (i.e., a Transformer produced by an Estimator),
+// we can view the parameters it used during fit().
+// This prints the parameter (name: value) pairs, where names are unique IDs for this
+// LogisticRegression instance.
+System.out.println("Model 1 was fit using parameters: " + model1.fittingParamMap());
+
+// We may alternatively specify parameters using a ParamMap.
+ParamMap paramMap = new ParamMap();
+paramMap.put(lr.maxIter(), 20); // Specify 1 Param.
+paramMap.put(lr.maxIter(), 30); // This overwrites the original maxIter.
+paramMap.put(lr.regParam(), 0.1);
+
+// One can also combine ParamMaps.
+ParamMap paramMap2 = new ParamMap();
+paramMap2.put(lr.scoreCol(), "probability"); // Changes output column name.
+ParamMap paramMapCombined = paramMap.$plus$plus(paramMap2);
+
+// Now learn a new model using the paramMapCombined parameters.
+// paramMapCombined overrides all parameters set earlier via lr.set* methods.
+LogisticRegressionModel model2 = lr.fit(training, paramMapCombined);
+System.out.println("Model 2 was fit using parameters: " + model2.fittingParamMap());
+
+// Prepare test documents.
+List<LabeledPoint> localTest = Lists.newArrayList(
+    new LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)),
+    new LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)),
+    new LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5)));
+JavaSchemaRDD test = jsql.applySchema(jsc.parallelize(localTest), LabeledPoint.class);
+
+// Make predictions on test documents using the Transformer.transform() method.
+// LogisticRegression.transform will only use the 'features' column.
+// Note that model2.transform() outputs a 'probability' column instead of the usual 'score'
+// column since we renamed the lr.scoreCol parameter previously.
+model2.transform(test).registerAsTable("results");
+JavaSchemaRDD results =
+    jsql.sql("SELECT features, label, probability, prediction FROM results");
+for (Row r: results.collect()) {
+  System.out.println("(" + r.get(0) + ", " + r.get(1) + ") -> prob=" + r.get(2)
+      + ", prediction=" + r.get(3));
+}
+{% endhighlight %}
+</div>
+
+</div>
+
+## Example: Pipeline
+
+This example follows the simple text document `Pipeline` illustrated in the figures above.
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.ml.Pipeline
+import org.apache.spark.ml.classification.LogisticRegression
+import org.apache.spark.ml.feature.{HashingTF, Tokenizer}
+import org.apache.spark.sql.{Row, SQLContext}
+
+// Labeled and unlabeled instance types.
+// Spark SQL can infer schema from case classes.
+case class LabeledDocument(id: Long, text: String, label: Double)
+case class Document(id: Long, text: String)
+
+// Set up contexts.  Import implicit conversions to SchemaRDD from sqlContext.
+val conf = new SparkConf().setAppName("SimpleTextClassificationPipeline")
+val sc = new SparkContext(conf)
+val sqlContext = new SQLContext(sc)
+import sqlContext._
+
+// Prepare training documents, which are labeled.
+val training = sparkContext.parallelize(Seq(
+  LabeledDocument(0L, "a b c d e spark", 1.0),
+  LabeledDocument(1L, "b d", 0.0),
+  LabeledDocument(2L, "spark f g h", 1.0),
+  LabeledDocument(3L, "hadoop mapreduce", 0.0)))
+
+// Configure an ML pipeline, which consists of three stages: tokenizer, hashingTF, and lr.
+val tokenizer = new Tokenizer()
+  .setInputCol("text")
+  .setOutputCol("words")
+val hashingTF = new HashingTF()
+  .setNumFeatures(1000)
+  .setInputCol(tokenizer.getOutputCol)
+  .setOutputCol("features")
+val lr = new LogisticRegression()
+  .setMaxIter(10)
+  .setRegParam(0.01)
+val pipeline = new Pipeline()
+  .setStages(Array(tokenizer, hashingTF, lr))
+
+// Fit the pipeline to training documents.
+val model = pipeline.fit(training)
+
+// Prepare test documents, which are unlabeled.
+val test = sparkContext.parallelize(Seq(
+  Document(4L, "spark i j k"),
+  Document(5L, "l m n"),
+  Document(6L, "mapreduce spark"),
+  Document(7L, "apache hadoop")))
+
+// Make predictions on test documents.
+model.transform(test)
+  .select('id, 'text, 'score, 'prediction)
+  .collect()
+  .foreach { case Row(id: Long, text: String, score: Double, prediction: Double) =>
+    println("(" + id + ", " + text + ") --> score=" + score + ", prediction=" + prediction)
+  }
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import java.io.Serializable;
+import java.util.List;
+import com.google.common.collect.Lists;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.ml.Pipeline;
+import org.apache.spark.ml.PipelineModel;
+import org.apache.spark.ml.PipelineStage;
+import org.apache.spark.ml.classification.LogisticRegression;
+import org.apache.spark.ml.feature.HashingTF;
+import org.apache.spark.ml.feature.Tokenizer;
+import org.apache.spark.sql.api.java.JavaSQLContext;
+import org.apache.spark.sql.api.java.JavaSchemaRDD;
+import org.apache.spark.sql.api.java.Row;
+import org.apache.spark.SparkConf;
+
+// Labeled and unlabeled instance types.
+// Spark SQL can infer schema from Java Beans.
+public class Document implements Serializable {
+  private Long id;
+  private String text;
+
+  public Document(Long id, String text) {
+    this.id = id;
+    this.text = text;
+  }
+
+  public Long getId() { return this.id; }
+  public void setId(Long id) { this.id = id; }
+
+  public String getText() { return this.text; }
+  public void setText(String text) { this.text = text; }
+}
+
+public class LabeledDocument extends Document implements Serializable {
+  private Double label;
+
+  public LabeledDocument(Long id, String text, Double label) {
+    super(id, text);
+    this.label = label;
+  }
+
+  public Double getLabel() { return this.label; }
+  public void setLabel(Double label) { this.label = label; }
+}
+
+// Set up contexts.
+SparkConf conf = new SparkConf().setAppName("JavaSimpleTextClassificationPipeline");
+JavaSparkContext jsc = new JavaSparkContext(conf);
+JavaSQLContext jsql = new JavaSQLContext(jsc);
+
+// Prepare training documents, which are labeled.
+List<LabeledDocument> localTraining = Lists.newArrayList(
+  new LabeledDocument(0L, "a b c d e spark", 1.0),
+  new LabeledDocument(1L, "b d", 0.0),
+  new LabeledDocument(2L, "spark f g h", 1.0),
+  new LabeledDocument(3L, "hadoop mapreduce", 0.0));
+JavaSchemaRDD training =
+  jsql.applySchema(jsc.parallelize(localTraining), LabeledDocument.class);
+
+// Configure an ML pipeline, which consists of three stages: tokenizer, hashingTF, and lr.
+Tokenizer tokenizer = new Tokenizer()
+  .setInputCol("text")
+  .setOutputCol("words");
+HashingTF hashingTF = new HashingTF()
+  .setNumFeatures(1000)
+  .setInputCol(tokenizer.getOutputCol())
+  .setOutputCol("features");
+LogisticRegression lr = new LogisticRegression()
+  .setMaxIter(10)
+  .setRegParam(0.01);
+Pipeline pipeline = new Pipeline()
+  .setStages(new PipelineStage[] {tokenizer, hashingTF, lr});
+
+// Fit the pipeline to training documents.
+PipelineModel model = pipeline.fit(training);
+
+// Prepare test documents, which are unlabeled.
+List<Document> localTest = Lists.newArrayList(
+  new Document(4L, "spark i j k"),
+  new Document(5L, "l m n"),
+  new Document(6L, "mapreduce spark"),
+  new Document(7L, "apache hadoop"));
+JavaSchemaRDD test =
+  jsql.applySchema(jsc.parallelize(localTest), Document.class);
+
+// Make predictions on test documents.
+model.transform(test).registerAsTable("prediction");
+JavaSchemaRDD predictions = jsql.sql("SELECT id, text, score, prediction FROM prediction");
+for (Row r: predictions.collect()) {
+  System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> score=" + r.get(2)
+      + ", prediction=" + r.get(3));
+}
+{% endhighlight %}
+</div>
+
+</div>
+
+## Example: Model Selection via Cross-Validation
+
+An important task in ML is *model selection*, or using data to find the best model or parameters for a given task.  This is also called *tuning*.
+`Pipeline`s facilitate model selection by making it easy to tune an entire `Pipeline` at once, rather than tuning each element in the `Pipeline` separately.
+
+Currently, `spark.ml` supports model selection using the [`CrossValidator`](api/scala/index.html#org.apache.spark.ml.tuning.CrossValidator) class, which takes an `Estimator`, a set of `ParamMap`s, and an [`Evaluator`](api/scala/index.html#org.apache.spark.ml.Evaluator).
+`CrossValidator` begins by splitting the dataset into a set of *folds* which are used as separate training and test datasets; e.g., with `$k=3$` folds, `CrossValidator` will generate 3 (training, test) dataset pairs, each of which uses 2/3 of the data for training and 1/3 for testing.
+`CrossValidator` iterates through the set of `ParamMap`s. For each `ParamMap`, it trains the given `Estimator` and evaluates it using the given `Evaluator`.
+The `ParamMap` which produces the best evaluation metric (averaged over the `$k$` folds) is selected as the best model.
+`CrossValidator` finally fits the `Estimator` using the best `ParamMap` and the entire dataset.
+
+The following example demonstrates using `CrossValidator` to select from a grid of parameters.
+To help construct the parameter grid, we use the [`ParamGridBuilder`](api/scala/index.html#org.apache.spark.ml.tuning.ParamGridBuilder) utility.
+
+Note that cross-validation over a grid of parameters is expensive.
+E.g., in the example below, the parameter grid has 3 values for `hashingTF.numFeatures` and 2 values for `lr.regParam`, and `CrossValidator` uses 2 folds.  This multiplies out to `$(3 \times 2) \times 2 = 12$` different models being trained.
+In realistic settings, it can be common to try many more parameters and use more folds (`$k=3$` and `$k=10$` are common).
+In other words, using `CrossValidator` can be very expensive.
+However, it is also a well-established method for choosing parameters which is more statistically sound than heuristic hand-tuning.
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.SparkContext._
+import org.apache.spark.ml.Pipeline
+import org.apache.spark.ml.classification.LogisticRegression
+import org.apache.spark.ml.evaluation.BinaryClassificationEvaluator
+import org.apache.spark.ml.feature.{HashingTF, Tokenizer}
+import org.apache.spark.ml.tuning.{ParamGridBuilder, CrossValidator}
+import org.apache.spark.sql.{Row, SQLContext}
+
+val conf = new SparkConf().setAppName("CrossValidatorExample")
+val sc = new SparkContext(conf)
+val sqlContext = new SQLContext(sc)
+import sqlContext._
+
+// Prepare training documents, which are labeled.
+val training = sparkContext.parallelize(Seq(
+  LabeledDocument(0L, "a b c d e spark", 1.0),
+  LabeledDocument(1L, "b d", 0.0),
+  LabeledDocument(2L, "spark f g h", 1.0),
+  LabeledDocument(3L, "hadoop mapreduce", 0.0),
+  LabeledDocument(4L, "b spark who", 1.0),
+  LabeledDocument(5L, "g d a y", 0.0),
+  LabeledDocument(6L, "spark fly", 1.0),
+  LabeledDocument(7L, "was mapreduce", 0.0),
+  LabeledDocument(8L, "e spark program", 1.0),
+  LabeledDocument(9L, "a e c l", 0.0),
+  LabeledDocument(10L, "spark compile", 1.0),
+  LabeledDocument(11L, "hadoop software", 0.0)))
+
+// Configure an ML pipeline, which consists of three stages: tokenizer, hashingTF, and lr.
+val tokenizer = new Tokenizer()
+  .setInputCol("text")
+  .setOutputCol("words")
+val hashingTF = new HashingTF()
+  .setInputCol(tokenizer.getOutputCol)
+  .setOutputCol("features")
+val lr = new LogisticRegression()
+  .setMaxIter(10)
+val pipeline = new Pipeline()
+  .setStages(Array(tokenizer, hashingTF, lr))
+
+// We now treat the Pipeline as an Estimator, wrapping it in a CrossValidator instance.
+// This will allow us to jointly choose parameters for all Pipeline stages.
+// A CrossValidator requires an Estimator, a set of Estimator ParamMaps, and an Evaluator.
+val crossval = new CrossValidator()
+  .setEstimator(pipeline)
+  .setEvaluator(new BinaryClassificationEvaluator)
+// We use a ParamGridBuilder to construct a grid of parameters to search over.
+// With 3 values for hashingTF.numFeatures and 2 values for lr.regParam,
+// this grid will have 3 x 2 = 6 parameter settings for CrossValidator to choose from.
+val paramGrid = new ParamGridBuilder()
+  .addGrid(hashingTF.numFeatures, Array(10, 100, 1000))
+  .addGrid(lr.regParam, Array(0.1, 0.01))
+  .build()
+crossval.setEstimatorParamMaps(paramGrid)
+crossval.setNumFolds(2) // Use 3+ in practice
+
+// Run cross-validation, and choose the best set of parameters.
+val cvModel = crossval.fit(training)
+// Get the best LogisticRegression model (with the best set of parameters from paramGrid).
+val lrModel = cvModel.bestModel
+
+// Prepare test documents, which are unlabeled.
+val test = sparkContext.parallelize(Seq(
+  Document(4L, "spark i j k"),
+  Document(5L, "l m n"),
+  Document(6L, "mapreduce spark"),
+  Document(7L, "apache hadoop")))
+
+// Make predictions on test documents. cvModel uses the best model found (lrModel).
+cvModel.transform(test)
+  .select('id, 'text, 'score, 'prediction)
+  .collect()
+  .foreach { case Row(id: Long, text: String, score: Double, prediction: Double) =>
+  println("(" + id + ", " + text + ") --> score=" + score + ", prediction=" + prediction)
+}
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import java.util.List;
+import com.google.common.collect.Lists;
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.ml.Model;
+import org.apache.spark.ml.Pipeline;
+import org.apache.spark.ml.PipelineStage;
+import org.apache.spark.ml.classification.LogisticRegression;
+import org.apache.spark.ml.evaluation.BinaryClassificationEvaluator;
+import org.apache.spark.ml.feature.HashingTF;
+import org.apache.spark.ml.feature.Tokenizer;
+import org.apache.spark.ml.param.ParamMap;
+import org.apache.spark.ml.tuning.CrossValidator;
+import org.apache.spark.ml.tuning.CrossValidatorModel;
+import org.apache.spark.ml.tuning.ParamGridBuilder;
+import org.apache.spark.sql.api.java.JavaSQLContext;
+import org.apache.spark.sql.api.java.JavaSchemaRDD;
+import org.apache.spark.sql.api.java.Row;
+
+SparkConf conf = new SparkConf().setAppName("JavaCrossValidatorExample");
+JavaSparkContext jsc = new JavaSparkContext(conf);
+JavaSQLContext jsql = new JavaSQLContext(jsc);
+
+// Prepare training documents, which are labeled.
+List<LabeledDocument> localTraining = Lists.newArrayList(
+  new LabeledDocument(0L, "a b c d e spark", 1.0),
+  new LabeledDocument(1L, "b d", 0.0),
+  new LabeledDocument(2L, "spark f g h", 1.0),
+  new LabeledDocument(3L, "hadoop mapreduce", 0.0),
+  new LabeledDocument(4L, "b spark who", 1.0),
+  new LabeledDocument(5L, "g d a y", 0.0),
+  new LabeledDocument(6L, "spark fly", 1.0),
+  new LabeledDocument(7L, "was mapreduce", 0.0),
+  new LabeledDocument(8L, "e spark program", 1.0),
+  new LabeledDocument(9L, "a e c l", 0.0),
+  new LabeledDocument(10L, "spark compile", 1.0),
+  new LabeledDocument(11L, "hadoop software", 0.0));
+JavaSchemaRDD training =
+    jsql.applySchema(jsc.parallelize(localTraining), LabeledDocument.class);
+
+// Configure an ML pipeline, which consists of three stages: tokenizer, hashingTF, and lr.
+Tokenizer tokenizer = new Tokenizer()
+  .setInputCol("text")
+  .setOutputCol("words");
+HashingTF hashingTF = new HashingTF()
+  .setNumFeatures(1000)
+  .setInputCol(tokenizer.getOutputCol())
+  .setOutputCol("features");
+LogisticRegression lr = new LogisticRegression()
+  .setMaxIter(10)
+  .setRegParam(0.01);
+Pipeline pipeline = new Pipeline()
+  .setStages(new PipelineStage[] {tokenizer, hashingTF, lr});
+
+// We now treat the Pipeline as an Estimator, wrapping it in a CrossValidator instance.
+// This will allow us to jointly choose parameters for all Pipeline stages.
+// A CrossValidator requires an Estimator, a set of Estimator ParamMaps, and an Evaluator.
+CrossValidator crossval = new CrossValidator()
+    .setEstimator(pipeline)
+    .setEvaluator(new BinaryClassificationEvaluator());
+// We use a ParamGridBuilder to construct a grid of parameters to search over.
+// With 3 values for hashingTF.numFeatures and 2 values for lr.regParam,
+// this grid will have 3 x 2 = 6 parameter settings for CrossValidator to choose from.
+ParamMap[] paramGrid = new ParamGridBuilder()
+    .addGrid(hashingTF.numFeatures(), new int[]{10, 100, 1000})
+    .addGrid(lr.regParam(), new double[]{0.1, 0.01})
+    .build();
+crossval.setEstimatorParamMaps(paramGrid);
+crossval.setNumFolds(2); // Use 3+ in practice
+
+// Run cross-validation, and choose the best set of parameters.
+CrossValidatorModel cvModel = crossval.fit(training);
+// Get the best LogisticRegression model (with the best set of parameters from paramGrid).
+Model lrModel = cvModel.bestModel();
+
+// Prepare test documents, which are unlabeled.
+List<Document> localTest = Lists.newArrayList(
+  new Document(4L, "spark i j k"),
+  new Document(5L, "l m n"),
+  new Document(6L, "mapreduce spark"),
+  new Document(7L, "apache hadoop"));
+JavaSchemaRDD test = jsql.applySchema(jsc.parallelize(localTest), Document.class);
+
+// Make predictions on test documents. cvModel uses the best model found (lrModel).
+cvModel.transform(test).registerAsTable("prediction");
+JavaSchemaRDD predictions = jsql.sql("SELECT id, text, score, prediction FROM prediction");
+for (Row r: predictions.collect()) {
+  System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> score=" + r.get(2)
+      + ", prediction=" + r.get(3));
+}
+{% endhighlight %}
+</div>
+
+</div>
+
+# Dependencies
+
+Spark ML currently depends on MLlib and has the same dependencies.
+Please see the [MLlib Dependencies guide](mllib-guide.html#Dependencies) for more info.
+
+Spark ML also depends upon Spark SQL, but the relevant parts of Spark SQL do not bring additional dependencies.
+
+# Developers
+
+**Development plan**
+
+If all goes well, `spark.ml` will become the primary ML package at the time of the Spark 1.3 release.  Initially, simple wrappers will be used to port algorithms to `spark.ml`, but eventually, code will be moved to `spark.ml` and `spark.mllib` will be deprecated.
+
+**Advice to developers**
+
+During the next development cycle, new algorithms should be contributed to `spark.mllib`, but we welcome patches sent to either package.  If an algorithm is best expressed using the new API (e.g., feature transformers), we may ask for developers to use the new `spark.ml` API.
+Wrappers for old and new algorithms can be contributed to `spark.ml`.
+
+Users will be able to use algorithms from either of the two packages.  The main difficulty will be the differences in APIs between the two packages.
+
diff --git a/docs/mllib-decision-tree.md b/docs/mllib-decision-tree.md
index 12a6afbeea829..fc8e732251a30 100644
--- a/docs/mllib-decision-tree.md
+++ b/docs/mllib-decision-tree.md
@@ -11,7 +11,7 @@ displayTitle: <a href="mllib-guide.html">MLlib</a> - Decision Tree
 and their ensembles are popular methods for the machine learning tasks of
 classification and regression. Decision trees are widely used since they are easy to interpret,
 handle categorical features, extend to the multiclass classification setting, do not require
-feature scaling and are able to capture nonlinearities and feature interactions. Tree ensemble
+feature scaling, and are able to capture non-linearities and feature interactions. Tree ensemble
 algorithms such as random forests and boosting are among the top performers for classification and
 regression tasks.
 
@@ -19,6 +19,8 @@ MLlib supports decision trees for binary and multiclass classification and for r
 using both continuous and categorical features. The implementation partitions data by rows,
 allowing distributed training with millions of instances.
 
+Ensembles of trees (Random Forests and Gradient-Boosted Trees) are described in the [Ensembles guide](mllib-ensembles.html).
+
 ## Basic algorithm
 
 The decision tree is a greedy algorithm that performs a recursive binary partitioning of the feature
@@ -42,18 +44,18 @@ impurity measure for regression (variance).
     <tr>
       <td>Gini impurity</td>
 	  <td>Classification</td>
-	  <td>$\sum_{i=1}^{M} f_i(1-f_i)$</td><td>$f_i$ is the frequency of label $i$ at a node and $M$ is the number of unique labels.</td>
+	  <td>$\sum_{i=1}^{C} f_i(1-f_i)$</td><td>$f_i$ is the frequency of label $i$ at a node and $C$ is the number of unique labels.</td>
     </tr>
     <tr>
       <td>Entropy</td>
 	  <td>Classification</td>
-	  <td>$\sum_{i=1}^{M} -f_ilog(f_i)$</td><td>$f_i$ is the frequency of label $i$ at a node and $M$ is the number of unique labels.</td>
+	  <td>$\sum_{i=1}^{C} -f_ilog(f_i)$</td><td>$f_i$ is the frequency of label $i$ at a node and $C$ is the number of unique labels.</td>
     </tr>
     <tr>
       <td>Variance</td>
 	  <td>Regression</td>
-     <td>$\frac{1}{n} \sum_{i=1}^{N} (x_i - \mu)^2$</td><td>$y_i$ is label for an instance,
-	  $N$ is the number of instances and $\mu$ is the mean given by $\frac{1}{N} \sum_{i=1}^n x_i$.</td>
+     <td>$\frac{1}{N} \sum_{i=1}^{N} (x_i - \mu)^2$</td><td>$y_i$ is label for an instance,
+	  $N$ is the number of instances and $\mu$ is the mean given by $\frac{1}{N} \sum_{i=1}^N x_i$.</td>
     </tr>
   </tbody>
 </table>
@@ -103,36 +105,73 @@ and the resulting `$M-1$` split candidates are considered.
 
 ### Stopping rule
 
-The recursive tree construction is stopped at a node when one of the two conditions is met:
+The recursive tree construction is stopped at a node when one of the following conditions is met:
 
 1. The node depth is equal to the `maxDepth` training parameter.
-2. No split candidate leads to an information gain at the node.
+2. No split candidate leads to an information gain greater than `minInfoGain`.
+3. No split candidate produces child nodes which each have at least `minInstancesPerNode` training instances.
+
+## Usage tips
+
+We include a few guidelines for using decision trees by discussing the various parameters.
+The parameters are listed below roughly in order of descending importance.  New users should mainly consider the "Problem specification parameters" section and the `maxDepth` parameter.
+
+### Problem specification parameters
+
+These parameters describe the problem you want to solve and your dataset.
+They should be specified and do not require tuning.
+
+* **`algo`**: `Classification` or `Regression`
+
+* **`numClasses`**: Number of classes (for `Classification` only)
+
+* **`categoricalFeaturesInfo`**: Specifies which features are categorical and how many categorical values each of those features can take.  This is given as a map from feature indices to feature arity (number of categories).  Any features not in this map are treated as continuous.
+  * E.g., `Map(0 -> 2, 4 -> 10)` specifies that feature `0` is binary (taking values `0` or `1`) and that feature `4` has 10 categories (values `{0, 1, ..., 9}`).  Note that feature indices are 0-based: features `0` and `4` are the 1st and 5th elements of an instance's feature vector.
+  * Note that you do not have to specify `categoricalFeaturesInfo`.  The algorithm will still run and may get reasonable results.  However, performance should be better if categorical features are properly designated.
+
+### Stopping criteria
+
+These parameters determine when the tree stops building (adding new nodes).
+When tuning these parameters, be careful to validate on held-out test data to avoid overfitting.
+
+* **`maxDepth`**: Maximum depth of a tree.  Deeper trees are more expressive (potentially allowing higher accuracy), but they are also more costly to train and are more likely to overfit.
+
+* **`minInstancesPerNode`**: For a node to be split further, each of its children must receive at least this number of training instances.  This is commonly used with [RandomForest](api/scala/index.html#org.apache.spark.mllib.tree.RandomForest) since those are often trained deeper than individual trees.
+
+* **`minInfoGain`**: For a node to be split further, the split must improve at least this much (in terms of information gain).
+
+### Tunable parameters
 
-## Implementation details
+These parameters may be tuned.  Be careful to validate on held-out test data when tuning in order to avoid overfitting.
 
-### Max memory requirements
+* **`maxBins`**: Number of bins used when discretizing continuous features.
+  * Increasing `maxBins` allows the algorithm to consider more split candidates and make fine-grained split decisions.  However, it also increases computation and communication.
+  * Note that the `maxBins` parameter must be at least the maximum number of categories `$M$` for any categorical feature.
 
-For faster processing, the decision tree algorithm performs simultaneous histogram computations for
-all nodes at each level of the tree. This could lead to high memory requirements at deeper levels
-of the tree, potentially leading to memory overflow errors. To alleviate this problem, a `maxMemoryInMB`
-training parameter specifies the maximum amount of memory at the workers (twice as much at the
-master) to be allocated to the histogram computation. The default value is conservatively chosen to
-be 256 MB to allow the decision algorithm to work in most scenarios. Once the memory requirements
-for a level-wise computation cross the `maxMemoryInMB` threshold, the node training tasks at each
-subsequent level are split into smaller tasks.
+* **`maxMemoryInMB`**: Amount of memory to be used for collecting sufficient statistics.
+  * The default value is conservatively chosen to be 256 MB to allow the decision algorithm to work in most scenarios.  Increasing `maxMemoryInMB` can lead to faster training (if the memory is available) by allowing fewer passes over the data.  However, there may be decreasing returns as `maxMemoryInMB` grows since the amount of communication on each iteration can be proportional to `maxMemoryInMB`.
+  * *Implementation details*: For faster processing, the decision tree algorithm collects statistics about groups of nodes to split (rather than 1 node at a time).  The number of nodes which can be handled in one group is determined by the memory requirements (which vary per features).  The `maxMemoryInMB` parameter specifies the memory limit in terms of megabytes which each worker can use for these statistics.
 
-Note that, if you have a large amount of memory, increasing `maxMemoryInMB` can lead to faster
-training by requiring fewer passes over the data.
+* **`subsamplingRate`**: Fraction of the training data used for learning the decision tree.  This parameter is most relevant for training ensembles of trees (using [`RandomForest`](api/scala/index.html#org.apache.spark.mllib.tree.RandomForest) and [`GradientBoostedTrees`](api/scala/index.html#org.apache.spark.mllib.tree.GradientBoostedTrees)), where it can be useful to subsample the original data.  For training a single decision tree, this parameter is less useful since the number of training instances is generally not the main constraint.
 
-### Binning feature values
+* **`impurity`**: Impurity measure (discussed above) used to choose between candidate splits.  This measure must match the `algo` parameter.
 
-Increasing `maxBins` allows the algorithm to consider more split candidates and make fine-grained
-split decisions.  However, it also increases computation and communication.
+### Caching and checkpointing
 
-Note that the `maxBins` parameter must be at least the maximum number of categories `$M$` for
-any categorical feature.
+MLlib 1.2 adds several features for scaling up to larger (deeper) trees and tree ensembles.  When `maxDepth` is set to be large, it can be useful to turn on node ID caching and checkpointing.  These parameters are also useful for [RandomForest](api/scala/index.html#org.apache.spark.mllib.tree.RandomForest) when `numTrees` is set to be large.
 
-### Scaling
+* **`useNodeIdCache`**: If this is set to true, the algorithm will avoid passing the current model (tree or trees) to executors on each iteration.
+  * This can be useful with deep trees (speeding up computation on workers) and for large Random Forests (reducing communication on each iteration).
+  * *Implementation details*: By default, the algorithm communicates the current model to executors so that executors can match training instances with tree nodes.  When this setting is turned on, then the algorithm will instead cache this information.
+
+Node ID caching generates a sequence of RDDs (1 per iteration).  This long lineage can cause performance problems, but checkpointing intermediate RDDs can alleviate those problems.
+Note that checkpointing is only applicable when `useNodeIdCache` is set to true.
+
+* **`checkpointDir`**: Directory for checkpointing node ID cache RDDs.
+
+* **`checkpointInterval`**: Frequency for checkpointing node ID cache RDDs.  Setting this too low will cause extra overhead from writing to HDFS; setting this too high can cause problems if executors fail and the RDD needs to be recomputed.
+
+## Scaling
 
 Computation scales approximately linearly in the number of training instances,
 in the number of features, and in the `maxBins` parameter.
@@ -148,7 +187,7 @@ The example below demonstrates how to load a
 [LIBSVM data file](http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/),
 parse it as an RDD of `LabeledPoint` and then
 perform classification using a decision tree with Gini impurity as an impurity measure and a
-maximum tree depth of 5. The training error is calculated to measure the algorithm accuracy.
+maximum tree depth of 5. The test error is calculated to measure the algorithm accuracy.
 
 <div class="codetabs">
 
@@ -158,8 +197,10 @@ import org.apache.spark.mllib.tree.DecisionTree
 import org.apache.spark.mllib.util.MLUtils
 
 // Load and parse the data file.
-// Cache the data since we will use it again to compute training error.
-val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt").cache()
+val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt")
+// Split the data into training and test sets (30% held out for testing)
+val splits = data.randomSplit(Array(0.7, 0.3))
+val (trainingData, testData) = (splits(0), splits(1))
 
 // Train a DecisionTree model.
 //  Empty categoricalFeaturesInfo indicates all features are continuous.
@@ -169,17 +210,17 @@ val impurity = "gini"
 val maxDepth = 5
 val maxBins = 32
 
-val model = DecisionTree.trainClassifier(data, numClasses, categoricalFeaturesInfo, impurity,
-  maxDepth, maxBins)
+val model = DecisionTree.trainClassifier(trainingData, numClasses, categoricalFeaturesInfo,
+  impurity, maxDepth, maxBins)
 
-// Evaluate model on training instances and compute training error
-val labelAndPreds = data.map { point =>
+// Evaluate model on test instances and compute test error
+val labelAndPreds = testData.map { point =>
   val prediction = model.predict(point.features)
   (point.label, prediction)
 }
-val trainErr = labelAndPreds.filter(r => r._1 != r._2).count.toDouble / data.count
-println("Training Error = " + trainErr)
-println("Learned classification tree model:\n" + model)
+val testErr = labelAndPreds.filter(r => r._1 != r._2).count.toDouble / testData.count()
+println("Test Error = " + testErr)
+println("Learned classification tree model:\n" + model.toDebugString)
 {% endhighlight %}
 </div>
 
@@ -187,7 +228,6 @@ println("Learned classification tree model:\n" + model)
 {% highlight java %}
 import java.util.HashMap;
 import scala.Tuple2;
-import org.apache.spark.api.java.function.Function2;
 import org.apache.spark.api.java.JavaPairRDD;
 import org.apache.spark.api.java.JavaRDD;
 import org.apache.spark.api.java.JavaSparkContext;
@@ -203,37 +243,42 @@ SparkConf sparkConf = new SparkConf().setAppName("JavaDecisionTree");
 JavaSparkContext sc = new JavaSparkContext(sparkConf);
 
 // Load and parse the data file.
-// Cache the data since we will use it again to compute training error.
 String datapath = "data/mllib/sample_libsvm_data.txt";
-JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD().cache();
+JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+// Split the data into training and test sets (30% held out for testing)
+JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+JavaRDD<LabeledPoint> trainingData = splits[0];
+JavaRDD<LabeledPoint> testData = splits[1];
 
 // Set parameters.
 //  Empty categoricalFeaturesInfo indicates all features are continuous.
 Integer numClasses = 2;
-HashMap<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+Map<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
 String impurity = "gini";
 Integer maxDepth = 5;
 Integer maxBins = 32;
 
 // Train a DecisionTree model for classification.
-final DecisionTreeModel model = DecisionTree.trainClassifier(data, numClasses,
+final DecisionTreeModel model = DecisionTree.trainClassifier(trainingData, numClasses,
   categoricalFeaturesInfo, impurity, maxDepth, maxBins);
 
-// Evaluate model on training instances and compute training error
+// Evaluate model on test instances and compute test error
 JavaPairRDD<Double, Double> predictionAndLabel =
-  data.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
-    @Override public Tuple2<Double, Double> call(LabeledPoint p) {
+  testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+    @Override
+    public Tuple2<Double, Double> call(LabeledPoint p) {
       return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
     }
   });
-Double trainErr =
+Double testErr =
   1.0 * predictionAndLabel.filter(new Function<Tuple2<Double, Double>, Boolean>() {
-    @Override public Boolean call(Tuple2<Double, Double> pl) {
+    @Override
+    public Boolean call(Tuple2<Double, Double> pl) {
       return !pl._1().equals(pl._2());
     }
-  }).count() / data.count();
-System.out.println("Training error: " + trainErr);
-System.out.println("Learned classification tree model:\n" + model);
+  }).count() / testData.count();
+System.out.println("Test Error: " + testErr);
+System.out.println("Learned classification tree model:\n" + model.toDebugString());
 {% endhighlight %}
 </div>
 
@@ -244,26 +289,23 @@ from pyspark.mllib.tree import DecisionTree
 from pyspark.mllib.util import MLUtils
 
 # Load and parse the data file into an RDD of LabeledPoint.
-# Cache the data since we will use it again to compute training error.
-data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt').cache()
+data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt')
+# Split the data into training and test sets (30% held out for testing)
+(trainingData, testData) = data.randomSplit([0.7, 0.3])
 
 # Train a DecisionTree model.
 #  Empty categoricalFeaturesInfo indicates all features are continuous.
-model = DecisionTree.trainClassifier(data, numClasses=2, categoricalFeaturesInfo={},
+model = DecisionTree.trainClassifier(trainingData, numClasses=2, categoricalFeaturesInfo={},
                                      impurity='gini', maxDepth=5, maxBins=32)
 
-# Evaluate model on training instances and compute training error
-predictions = model.predict(data.map(lambda x: x.features))
-labelsAndPredictions = data.map(lambda lp: lp.label).zip(predictions)
-trainErr = labelsAndPredictions.filter(lambda (v, p): v != p).count() / float(data.count())
-print('Training Error = ' + str(trainErr))
+# Evaluate model on test instances and compute test error
+predictions = model.predict(testData.map(lambda x: x.features))
+labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
+testErr = labelsAndPredictions.filter(lambda (v, p): v != p).count() / float(testData.count())
+print('Test Error = ' + str(testErr))
 print('Learned classification tree model:')
-print(model)
+print(model.toDebugString())
 {% endhighlight %}
-
-Note: When making predictions for a dataset, it is more efficient to do batch prediction rather
-than separately calling `predict` on each data point.  This is because the Python code makes calls
-to an underlying `DecisionTree` model in Scala.
 </div>
 
 </div>
@@ -285,8 +327,10 @@ import org.apache.spark.mllib.tree.DecisionTree
 import org.apache.spark.mllib.util.MLUtils
 
 // Load and parse the data file.
-// Cache the data since we will use it again to compute training error.
-val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt").cache()
+val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt")
+// Split the data into training and test sets (30% held out for testing)
+val splits = data.randomSplit(Array(0.7, 0.3))
+val (trainingData, testData) = (splits(0), splits(1))
 
 // Train a DecisionTree model.
 //  Empty categoricalFeaturesInfo indicates all features are continuous.
@@ -295,17 +339,17 @@ val impurity = "variance"
 val maxDepth = 5
 val maxBins = 32
 
-val model = DecisionTree.trainRegressor(data, categoricalFeaturesInfo, impurity,
+val model = DecisionTree.trainRegressor(trainingData, categoricalFeaturesInfo, impurity,
   maxDepth, maxBins)
 
-// Evaluate model on training instances and compute training error
-val labelsAndPredictions = data.map { point =>
+// Evaluate model on test instances and compute test error
+val labelsAndPredictions = testData.map { point =>
   val prediction = model.predict(point.features)
   (point.label, prediction)
 }
-val trainMSE = labelsAndPredictions.map{ case(v, p) => math.pow((v - p), 2)}.mean()
-println("Training Mean Squared Error = " + trainMSE)
-println("Learned regression tree model:\n" + model)
+val testMSE = labelsAndPredictions.map{ case(v, p) => math.pow((v - p), 2)}.mean()
+println("Test Mean Squared Error = " + testMSE)
+println("Learned regression tree model:\n" + model.toDebugString)
 {% endhighlight %}
 </div>
 
@@ -325,45 +369,51 @@ import org.apache.spark.mllib.tree.model.DecisionTreeModel;
 import org.apache.spark.mllib.util.MLUtils;
 import org.apache.spark.SparkConf;
 
-// Load and parse the data file.
-// Cache the data since we will use it again to compute training error.
-String datapath = "data/mllib/sample_libsvm_data.txt";
-JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD().cache();
-
 SparkConf sparkConf = new SparkConf().setAppName("JavaDecisionTree");
 JavaSparkContext sc = new JavaSparkContext(sparkConf);
 
+// Load and parse the data file.
+String datapath = "data/mllib/sample_libsvm_data.txt";
+JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+// Split the data into training and test sets (30% held out for testing)
+JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+JavaRDD<LabeledPoint> trainingData = splits[0];
+JavaRDD<LabeledPoint> testData = splits[1];
+
 // Set parameters.
 //  Empty categoricalFeaturesInfo indicates all features are continuous.
-HashMap<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+Map<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
 String impurity = "variance";
 Integer maxDepth = 5;
 Integer maxBins = 32;
 
 // Train a DecisionTree model.
-final DecisionTreeModel model = DecisionTree.trainRegressor(data,
+final DecisionTreeModel model = DecisionTree.trainRegressor(trainingData,
   categoricalFeaturesInfo, impurity, maxDepth, maxBins);
 
-// Evaluate model on training instances and compute training error
+// Evaluate model on test instances and compute test error
 JavaPairRDD<Double, Double> predictionAndLabel =
-  data.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
-    @Override public Tuple2<Double, Double> call(LabeledPoint p) {
+  testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+    @Override
+    public Tuple2<Double, Double> call(LabeledPoint p) {
       return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
     }
   });
-Double trainMSE =
+Double testMSE =
   predictionAndLabel.map(new Function<Tuple2<Double, Double>, Double>() {
-    @Override public Double call(Tuple2<Double, Double> pl) {
+    @Override
+    public Double call(Tuple2<Double, Double> pl) {
       Double diff = pl._1() - pl._2();
       return diff * diff;
     }
   }).reduce(new Function2<Double, Double, Double>() {
-    @Override public Double call(Double a, Double b) {
+    @Override
+    public Double call(Double a, Double b) {
       return a + b;
     }
   }) / data.count();
-System.out.println("Training Mean Squared Error: " + trainMSE);
-System.out.println("Learned regression tree model:\n" + model);
+System.out.println("Test Mean Squared Error: " + testMSE);
+System.out.println("Learned regression tree model:\n" + model.toDebugString());
 {% endhighlight %}
 </div>
 
@@ -374,26 +424,23 @@ from pyspark.mllib.tree import DecisionTree
 from pyspark.mllib.util import MLUtils
 
 # Load and parse the data file into an RDD of LabeledPoint.
-# Cache the data since we will use it again to compute training error.
-data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt').cache()
+data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt')
+# Split the data into training and test sets (30% held out for testing)
+(trainingData, testData) = data.randomSplit([0.7, 0.3])
 
 # Train a DecisionTree model.
 #  Empty categoricalFeaturesInfo indicates all features are continuous.
-model = DecisionTree.trainRegressor(data, categoricalFeaturesInfo={},
+model = DecisionTree.trainRegressor(trainingData, categoricalFeaturesInfo={},
                                     impurity='variance', maxDepth=5, maxBins=32)
 
-# Evaluate model on training instances and compute training error
-predictions = model.predict(data.map(lambda x: x.features))
-labelsAndPredictions = data.map(lambda lp: lp.label).zip(predictions)
-trainMSE = labelsAndPredictions.map(lambda (v, p): (v - p) * (v - p)).sum() / float(data.count())
-print('Training Mean Squared Error = ' + str(trainMSE))
+# Evaluate model on test instances and compute test error
+predictions = model.predict(testData.map(lambda x: x.features))
+labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
+testMSE = labelsAndPredictions.map(lambda (v, p): (v - p) * (v - p)).sum() / float(testData.count())
+print('Test Mean Squared Error = ' + str(testMSE))
 print('Learned regression tree model:')
-print(model)
+print(model.toDebugString())
 {% endhighlight %}
-
-Note: When making predictions for a dataset, it is more efficient to do batch prediction rather
-than separately calling `predict` on each data point.  This is because the Python code makes calls
-to an underlying `DecisionTree` model in Scala.
 </div>
 
 </div>
diff --git a/docs/mllib-ensembles.md b/docs/mllib-ensembles.md
new file mode 100644
index 0000000000000..23ede04b62d5b
--- /dev/null
+++ b/docs/mllib-ensembles.md
@@ -0,0 +1,653 @@
+---
+layout: global
+title: Ensembles - MLlib
+displayTitle: <a href="mllib-guide.html">MLlib</a> - Ensembles
+---
+
+* Table of contents
+{:toc}
+
+An [ensemble method](http://en.wikipedia.org/wiki/Ensemble_learning)
+is a learning algorithm which creates a model composed of a set of other base models.
+MLlib supports two major ensemble algorithms: [`GradientBoostedTrees`](api/scala/index.html#org.apache.spark.mllib.tree.GradientBosotedTrees) and [`RandomForest`](api/scala/index.html#org.apache.spark.mllib.tree.RandomForest).
+Both use [decision trees](mllib-decision-tree.html) as their base models.
+
+## Gradient-Boosted Trees vs. Random Forests
+
+Both [Gradient-Boosted Trees (GBTs)](mllib-ensembles.html#Gradient-Boosted-Trees-(GBTS)) and [Random Forests](mllib-ensembles.html#Random-Forests) are algorithms for learning ensembles of trees, but the training processes are different.  There are several practical trade-offs:
+
+ * GBTs train one tree at a time, so they can take longer to train than random forests.  Random Forests can train multiple trees in parallel.
+   * On the other hand, it is often reasonable to use smaller (shallower) trees with GBTs than with Random Forests, and training smaller trees takes less time.
+ * Random Forests can be less prone to overfitting.  Training more trees in a Random Forest reduces the likelihood of overfitting, but training more trees with GBTs increases the likelihood of overfitting.  (In statistical language, Random Forests reduce variance by using more trees, whereas GBTs reduce bias by using more trees.)
+ * Random Forests can be easier to tune since performance improves monotonically with the number of trees (whereas performance can start to decrease for GBTs if the number of trees grows too large).
+
+In short, both algorithms can be effective, and the choice should be based on the particular dataset.
+
+## Random Forests
+
+[Random forests](http://en.wikipedia.org/wiki/Random_forest)
+are ensembles of [decision trees](mllib-decision-tree.html).
+Random forests are one of the most successful machine learning models for classification and
+regression.  They combine many decision trees in order to reduce the risk of overfitting.
+Like decision trees, random forests handle categorical features,
+extend to the multiclass classification setting, do not require
+feature scaling, and are able to capture non-linearities and feature interactions.
+
+MLlib supports random forests for binary and multiclass classification and for regression,
+using both continuous and categorical features.
+MLlib implements random forests using the existing [decision tree](mllib-decision-tree.html)
+implementation.  Please see the decision tree guide for more information on trees.
+
+### Basic algorithm
+
+Random forests train a set of decision trees separately, so the training can be done in parallel.
+The algorithm injects randomness into the training process so that each decision tree is a bit
+different.  Combining the predictions from each tree reduces the variance of the predictions,
+improving the performance on test data.
+
+#### Training
+
+The randomness injected into the training process includes:
+
+* Subsampling the original dataset on each iteration to get a different training set (a.k.a. bootstrapping).
+* Considering different random subsets of features to split on at each tree node.
+
+Apart from these randomizations, decision tree training is done in the same way as for individual decision trees.
+
+#### Prediction
+
+To make a prediction on a new instance, a random forest must aggregate the predictions from its set of decision trees.  This aggregation is done differently for classification and regression.
+
+*Classification*: Majority vote. Each tree's prediction is counted as a vote for one class.  The label is predicted to be the class which receives the most votes.
+
+*Regression*: Averaging. Each tree predicts a real value.  The label is predicted to be the average of the tree predictions.
+
+### Usage tips
+
+We include a few guidelines for using random forests by discussing the various parameters.
+We omit some decision tree parameters since those are covered in the [decision tree guide](mllib-decision-tree.html).
+
+The first two parameters we mention are the most important, and tuning them can often improve performance:
+
+* **`numTrees`**: Number of trees in the forest.
+  * Increasing the number of trees will decrease the variance in predictions, improving the model's test-time accuracy.
+  * Training time increases roughly linearly in the number of trees.
+
+* **`maxDepth`**: Maximum depth of each tree in the forest.
+  * Increasing the depth makes the model more expressive and powerful.  However, deep trees take longer to train and are also more prone to overfitting.
+  * In general, it is acceptable to train deeper trees when using random forests than when using a single decision tree.  One tree is more likely to overfit than a random forest (because of the variance reduction from averaging multiple trees in the forest).
+
+The next two parameters generally do not require tuning.  However, they can be tuned to speed up training.
+
+* **`subsamplingRate`**: This parameter specifies the size of the dataset used for training each tree in the forest, as a fraction of the size of the original dataset.  The default (1.0) is recommended, but decreasing this fraction can speed up training.
+
+* **`featureSubsetStrategy`**: Number of features to use as candidates for splitting at each tree node.  The number is specified as a fraction or function of the total number of features.  Decreasing this number will speed up training, but can sometimes impact performance if too low.
+
+### Examples
+
+#### Classification
+
+The example below demonstrates how to load a
+[LIBSVM data file](http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/),
+parse it as an RDD of `LabeledPoint` and then
+perform classification using a Random Forest.
+The test error is calculated to measure the algorithm accuracy.
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.mllib.tree.RandomForest
+import org.apache.spark.mllib.util.MLUtils
+
+// Load and parse the data file.
+val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt")
+// Split the data into training and test sets (30% held out for testing)
+val splits = data.randomSplit(Array(0.7, 0.3))
+val (trainingData, testData) = (splits(0), splits(1))
+
+// Train a RandomForest model.
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+val numClasses = 2
+val categoricalFeaturesInfo = Map[Int, Int]()
+val numTrees = 3 // Use more in practice.
+val featureSubsetStrategy = "auto" // Let the algorithm choose.
+val impurity = "gini"
+val maxDepth = 4
+val maxBins = 32
+
+val model = RandomForest.trainClassifier(trainingData, numClasses, categoricalFeaturesInfo,
+  numTrees, featureSubsetStrategy, impurity, maxDepth, maxBins)
+
+// Evaluate model on test instances and compute test error
+val labelAndPreds = testData.map { point =>
+  val prediction = model.predict(point.features)
+  (point.label, prediction)
+}
+val testErr = labelAndPreds.filter(r => r._1 != r._2).count.toDouble / testData.count()
+println("Test Error = " + testErr)
+println("Learned classification forest model:\n" + model.toDebugString)
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import scala.Tuple2;
+import java.util.HashMap;
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaPairRDD;
+import org.apache.spark.api.java.JavaRDD;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.api.java.function.Function;
+import org.apache.spark.api.java.function.PairFunction;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.mllib.tree.RandomForest;
+import org.apache.spark.mllib.tree.model.RandomForestModel;
+import org.apache.spark.mllib.util.MLUtils;
+
+SparkConf sparkConf = new SparkConf().setAppName("JavaRandomForestClassification");
+JavaSparkContext sc = new JavaSparkContext(sparkConf);
+
+// Load and parse the data file.
+String datapath = "data/mllib/sample_libsvm_data.txt";
+JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+// Split the data into training and test sets (30% held out for testing)
+JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+JavaRDD<LabeledPoint> trainingData = splits[0];
+JavaRDD<LabeledPoint> testData = splits[1];
+
+// Train a RandomForest model.
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+Integer numClasses = 2;
+HashMap<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+Integer numTrees = 3; // Use more in practice.
+String featureSubsetStrategy = "auto"; // Let the algorithm choose.
+String impurity = "gini";
+Integer maxDepth = 5;
+Integer maxBins = 32;
+Integer seed = 12345;
+
+final RandomForestModel model = RandomForest.trainClassifier(trainingData, numClasses,
+  categoricalFeaturesInfo, numTrees, featureSubsetStrategy, impurity, maxDepth, maxBins,
+  seed);
+
+// Evaluate model on test instances and compute test error
+JavaPairRDD<Double, Double> predictionAndLabel =
+  testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+    @Override
+    public Tuple2<Double, Double> call(LabeledPoint p) {
+      return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
+    }
+  });
+Double testErr =
+  1.0 * predictionAndLabel.filter(new Function<Tuple2<Double, Double>, Boolean>() {
+    @Override
+    public Boolean call(Tuple2<Double, Double> pl) {
+      return !pl._1().equals(pl._2());
+    }
+  }).count() / testData.count();
+System.out.println("Test Error: " + testErr);
+System.out.println("Learned classification forest model:\n" + model.toDebugString());
+{% endhighlight %}
+</div>
+
+<div data-lang="python">
+{% highlight python %}
+from pyspark.mllib.tree import RandomForest
+from pyspark.mllib.util import MLUtils
+
+# Load and parse the data file into an RDD of LabeledPoint.
+data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt')
+# Split the data into training and test sets (30% held out for testing)
+(trainingData, testData) = data.randomSplit([0.7, 0.3])
+
+# Train a RandomForest model.
+#  Empty categoricalFeaturesInfo indicates all features are continuous.
+#  Note: Use larger numTrees in practice.
+#  Setting featureSubsetStrategy="auto" lets the algorithm choose.
+model = RandomForest.trainClassifier(trainingData, numClasses=2, categoricalFeaturesInfo={},
+                                     numTrees=3, featureSubsetStrategy="auto",
+                                     impurity='gini', maxDepth=4, maxBins=32)
+
+# Evaluate model on test instances and compute test error
+predictions = model.predict(testData.map(lambda x: x.features))
+labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
+testErr = labelsAndPredictions.filter(lambda (v, p): v != p).count() / float(testData.count())
+print('Test Error = ' + str(testErr))
+print('Learned classification forest model:')
+print(model.toDebugString())
+{% endhighlight %}
+</div>
+
+</div>
+
+#### Regression
+
+The example below demonstrates how to load a
+[LIBSVM data file](http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/),
+parse it as an RDD of `LabeledPoint` and then
+perform regression using a Random Forest.
+The Mean Squared Error (MSE) is computed at the end to evaluate
+[goodness of fit](http://en.wikipedia.org/wiki/Goodness_of_fit).
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.mllib.tree.RandomForest
+import org.apache.spark.mllib.util.MLUtils
+
+// Load and parse the data file.
+val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt")
+// Split the data into training and test sets (30% held out for testing)
+val splits = data.randomSplit(Array(0.7, 0.3))
+val (trainingData, testData) = (splits(0), splits(1))
+
+// Train a RandomForest model.
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+val numClasses = 2
+val categoricalFeaturesInfo = Map[Int, Int]()
+val numTrees = 3 // Use more in practice.
+val featureSubsetStrategy = "auto" // Let the algorithm choose.
+val impurity = "variance"
+val maxDepth = 4
+val maxBins = 32
+
+val model = RandomForest.trainRegressor(trainingData, categoricalFeaturesInfo,
+  numTrees, featureSubsetStrategy, impurity, maxDepth, maxBins)
+
+// Evaluate model on test instances and compute test error
+val labelsAndPredictions = testData.map { point =>
+  val prediction = model.predict(point.features)
+  (point.label, prediction)
+}
+val testMSE = labelsAndPredictions.map{ case(v, p) => math.pow((v - p), 2)}.mean()
+println("Test Mean Squared Error = " + testMSE)
+println("Learned regression forest model:\n" + model.toDebugString)
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import java.util.HashMap;
+import scala.Tuple2;
+import org.apache.spark.api.java.function.Function2;
+import org.apache.spark.api.java.JavaPairRDD;
+import org.apache.spark.api.java.JavaRDD;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.api.java.function.Function;
+import org.apache.spark.api.java.function.PairFunction;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.mllib.tree.RandomForest;
+import org.apache.spark.mllib.tree.model.RandomForestModel;
+import org.apache.spark.mllib.util.MLUtils;
+import org.apache.spark.SparkConf;
+
+SparkConf sparkConf = new SparkConf().setAppName("JavaRandomForest");
+JavaSparkContext sc = new JavaSparkContext(sparkConf);
+
+// Load and parse the data file.
+String datapath = "data/mllib/sample_libsvm_data.txt";
+JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+// Split the data into training and test sets (30% held out for testing)
+JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+JavaRDD<LabeledPoint> trainingData = splits[0];
+JavaRDD<LabeledPoint> testData = splits[1];
+
+// Set parameters.
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+Map<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+String impurity = "variance";
+Integer maxDepth = 4;
+Integer maxBins = 32;
+
+// Train a RandomForest model.
+final RandomForestModel model = RandomForest.trainRegressor(trainingData,
+  categoricalFeaturesInfo, impurity, maxDepth, maxBins);
+
+// Evaluate model on test instances and compute test error
+JavaPairRDD<Double, Double> predictionAndLabel =
+  testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+    @Override
+    public Tuple2<Double, Double> call(LabeledPoint p) {
+      return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
+    }
+  });
+Double testMSE =
+  predictionAndLabel.map(new Function<Tuple2<Double, Double>, Double>() {
+    @Override
+    public Double call(Tuple2<Double, Double> pl) {
+      Double diff = pl._1() - pl._2();
+      return diff * diff;
+    }
+  }).reduce(new Function2<Double, Double, Double>() {
+    @Override
+    public Double call(Double a, Double b) {
+      return a + b;
+    }
+  }) / testData.count();
+System.out.println("Test Mean Squared Error: " + testMSE);
+System.out.println("Learned regression forest model:\n" + model.toDebugString());
+{% endhighlight %}
+</div>
+
+<div data-lang="python">
+{% highlight python %}
+from pyspark.mllib.tree import RandomForest
+from pyspark.mllib.util import MLUtils
+
+# Load and parse the data file into an RDD of LabeledPoint.
+data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt')
+# Split the data into training and test sets (30% held out for testing)
+(trainingData, testData) = data.randomSplit([0.7, 0.3])
+
+# Train a RandomForest model.
+#  Empty categoricalFeaturesInfo indicates all features are continuous.
+#  Note: Use larger numTrees in practice.
+#  Setting featureSubsetStrategy="auto" lets the algorithm choose.
+model = RandomForest.trainRegressor(trainingData, categoricalFeaturesInfo={},
+                                    numTrees=3, featureSubsetStrategy="auto",
+                                    impurity='variance', maxDepth=4, maxBins=32)
+
+# Evaluate model on test instances and compute test error
+predictions = model.predict(testData.map(lambda x: x.features))
+labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
+testMSE = labelsAndPredictions.map(lambda (v, p): (v - p) * (v - p)).sum() / float(testData.count())
+print('Test Mean Squared Error = ' + str(testMSE))
+print('Learned regression forest model:')
+print(model.toDebugString())
+{% endhighlight %}
+</div>
+
+</div>
+
+## Gradient-Boosted Trees (GBTs)
+
+[Gradient-Boosted Trees (GBTs)](http://en.wikipedia.org/wiki/Gradient_boosting)
+are ensembles of [decision trees](mllib-decision-tree.html).
+GBTs iteratively train decision trees in order to minimize a loss function.
+Like decision trees, GBTs handle categorical features,
+extend to the multiclass classification setting, do not require
+feature scaling, and are able to capture non-linearities and feature interactions.
+
+MLlib supports GBTs for binary classification and for regression,
+using both continuous and categorical features.
+MLlib implements GBTs using the existing [decision tree](mllib-decision-tree.html) implementation.  Please see the decision tree guide for more information on trees.
+
+*Note*: GBTs do not yet support multiclass classification.  For multiclass problems, please use
+[decision trees](mllib-decision-tree.html) or [Random Forests](mllib-ensembles.html#Random-Forest).
+
+### Basic algorithm
+
+Gradient boosting iteratively trains a sequence of decision trees.
+On each iteration, the algorithm uses the current ensemble to predict the label of each training instance and then compares the prediction with the true label.  The dataset is re-labeled to put more emphasis on training instances with poor predictions.  Thus, in the next iteration, the decision tree will help correct for previous mistakes.
+
+The specific mechanism for re-labeling instances is defined by a loss function (discussed below).  With each iteration, GBTs further reduce this loss function on the training data.
+
+#### Losses
+
+The table below lists the losses currently supported by GBTs in MLlib.
+Note that each loss is applicable to one of classification or regression, not both.
+
+Notation: $N$ = number of instances. $y_i$ = label of instance $i$.  $x_i$ = features of instance $i$.  $F(x_i)$ = model's predicted label for instance $i$.
+
+<table class="table">
+  <thead>
+    <tr><th>Loss</th><th>Task</th><th>Formula</th><th>Description</th></tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>Log Loss</td>
+	  <td>Classification</td>
+	  <td>$2 \sum_{i=1}^{N} \log(1+\exp(-2 y_i F(x_i)))$</td><td>Twice binomial negative log likelihood.</td>
+    </tr>
+    <tr>
+      <td>Squared Error</td>
+	  <td>Regression</td>
+	  <td>$\sum_{i=1}^{N} (y_i - F(x_i))^2$</td><td>Also called L2 loss.  Default loss for regression tasks.</td>
+    </tr>
+    <tr>
+      <td>Absolute Error</td>
+	  <td>Regression</td>
+     <td>$\sum_{i=1}^{N} |y_i - F(x_i)|$</td><td>Also called L1 loss.  Can be more robust to outliers than Squared Error.</td>
+    </tr>
+  </tbody>
+</table>
+
+### Usage tips
+
+We include a few guidelines for using GBTs by discussing the various parameters.
+We omit some decision tree parameters since those are covered in the [decision tree guide](mllib-decision-tree.html).
+
+* **`loss`**: See the section above for information on losses and their applicability to tasks (classification vs. regression).  Different losses can give significantly different results, depending on the dataset.
+
+* **`numIterations`**: This sets the number of trees in the ensemble.  Each iteration produces one tree.  Increasing this number makes the model more expressive, improving training data accuracy.  However, test-time accuracy may suffer if this is too large.
+
+* **`learningRate`**: This parameter should not need to be tuned.  If the algorithm behavior seems unstable, decreasing this value may improve stability.
+
+* **`algo`**: The algorithm or task (classification vs. regression) is set using the tree [Strategy] parameter.
+
+
+### Examples
+
+GBTs currently have APIs in Scala and Java.  Examples in both languages are shown below.
+
+#### Classification
+
+The example below demonstrates how to load a
+[LIBSVM data file](http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/),
+parse it as an RDD of `LabeledPoint` and then
+perform classification using Gradient-Boosted Trees with log loss.
+The test error is calculated to measure the algorithm accuracy.
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.mllib.tree.GradientBoostedTrees
+import org.apache.spark.mllib.tree.configuration.BoostingStrategy
+import org.apache.spark.mllib.util.MLUtils
+
+// Load and parse the data file.
+val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt")
+// Split the data into training and test sets (30% held out for testing)
+val splits = data.randomSplit(Array(0.7, 0.3))
+val (trainingData, testData) = (splits(0), splits(1))
+
+// Train a GradientBoostedTrees model.
+//  The defaultParams for Classification use LogLoss by default.
+val boostingStrategy = BoostingStrategy.defaultParams("Classification")
+boostingStrategy.numIterations = 3 // Note: Use more iterations in practice.
+boostingStrategy.treeStrategy.numClassesForClassification = 2
+boostingStrategy.treeStrategy.maxDepth = 5
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+boostingStrategy.treeStrategy.categoricalFeaturesInfo = Map[Int, Int]()
+
+val model = GradientBoostedTrees.train(trainingData, boostingStrategy)
+
+// Evaluate model on test instances and compute test error
+val labelAndPreds = testData.map { point =>
+  val prediction = model.predict(point.features)
+  (point.label, prediction)
+}
+val testErr = labelAndPreds.filter(r => r._1 != r._2).count.toDouble / testData.count()
+println("Test Error = " + testErr)
+println("Learned classification GBT model:\n" + model.toDebugString)
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import scala.Tuple2;
+import java.util.HashMap;
+import java.util.Map;
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaPairRDD;
+import org.apache.spark.api.java.JavaRDD;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.api.java.function.Function;
+import org.apache.spark.api.java.function.PairFunction;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.mllib.tree.GradientBoostedTrees;
+import org.apache.spark.mllib.tree.configuration.BoostingStrategy;
+import org.apache.spark.mllib.tree.model.GradientBoostedTreesModel;
+import org.apache.spark.mllib.util.MLUtils;
+
+SparkConf sparkConf = new SparkConf().setAppName("JavaGradientBoostedTrees");
+JavaSparkContext sc = new JavaSparkContext(sparkConf);
+
+// Load and parse the data file.
+String datapath = "data/mllib/sample_libsvm_data.txt";
+JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+// Split the data into training and test sets (30% held out for testing)
+JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+JavaRDD<LabeledPoint> trainingData = splits[0];
+JavaRDD<LabeledPoint> testData = splits[1];
+
+// Train a GradientBoostedTrees model.
+//  The defaultParams for Classification use LogLoss by default.
+BoostingStrategy boostingStrategy = BoostingStrategy.defaultParams("Classification");
+boostingStrategy.setNumIterations(3); // Note: Use more iterations in practice.
+boostingStrategy.getTreeStrategy().setNumClassesForClassification(2);
+boostingStrategy.getTreeStrategy().setMaxDepth(5);
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+Map<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+boostingStrategy.treeStrategy().setCategoricalFeaturesInfo(categoricalFeaturesInfo);
+
+final GradientBoostedTreesModel model =
+  GradientBoostedTrees.train(trainingData, boostingStrategy);
+
+// Evaluate model on test instances and compute test error
+JavaPairRDD<Double, Double> predictionAndLabel =
+  testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+    @Override
+    public Tuple2<Double, Double> call(LabeledPoint p) {
+      return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
+    }
+  });
+Double testErr =
+  1.0 * predictionAndLabel.filter(new Function<Tuple2<Double, Double>, Boolean>() {
+    @Override
+    public Boolean call(Tuple2<Double, Double> pl) {
+      return !pl._1().equals(pl._2());
+    }
+  }).count() / testData.count();
+System.out.println("Test Error: " + testErr);
+System.out.println("Learned classification GBT model:\n" + model.toDebugString());
+{% endhighlight %}
+</div>
+
+</div>
+
+#### Regression
+
+The example below demonstrates how to load a
+[LIBSVM data file](http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/),
+parse it as an RDD of `LabeledPoint` and then
+perform regression using Gradient-Boosted Trees with Squared Error as the loss.
+The Mean Squared Error (MSE) is computed at the end to evaluate
+[goodness of fit](http://en.wikipedia.org/wiki/Goodness_of_fit).
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.mllib.tree.GradientBoostedTrees
+import org.apache.spark.mllib.tree.configuration.BoostingStrategy
+import org.apache.spark.mllib.util.MLUtils
+
+// Load and parse the data file.
+val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt")
+// Split the data into training and test sets (30% held out for testing)
+val splits = data.randomSplit(Array(0.7, 0.3))
+val (trainingData, testData) = (splits(0), splits(1))
+
+// Train a GradientBoostedTrees model.
+//  The defaultParams for Regression use SquaredError by default.
+val boostingStrategy = BoostingStrategy.defaultParams("Regression")
+boostingStrategy.numIterations = 3 // Note: Use more iterations in practice.
+boostingStrategy.treeStrategy.maxDepth = 5
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+boostingStrategy.treeStrategy.categoricalFeaturesInfo = Map[Int, Int]()
+
+val model = GradientBoostedTrees.train(trainingData, boostingStrategy)
+
+// Evaluate model on test instances and compute test error
+val labelsAndPredictions = testData.map { point =>
+  val prediction = model.predict(point.features)
+  (point.label, prediction)
+}
+val testMSE = labelsAndPredictions.map{ case(v, p) => math.pow((v - p), 2)}.mean()
+println("Test Mean Squared Error = " + testMSE)
+println("Learned regression GBT model:\n" + model.toDebugString)
+{% endhighlight %}
+</div>
+
+<div data-lang="java">
+{% highlight java %}
+import scala.Tuple2;
+import java.util.HashMap;
+import java.util.Map;
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.function.Function2;
+import org.apache.spark.api.java.JavaPairRDD;
+import org.apache.spark.api.java.JavaRDD;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.api.java.function.Function;
+import org.apache.spark.api.java.function.PairFunction;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.mllib.tree.GradientBoostedTrees;
+import org.apache.spark.mllib.tree.configuration.BoostingStrategy;
+import org.apache.spark.mllib.tree.model.GradientBoostedTreesModel;
+import org.apache.spark.mllib.util.MLUtils;
+
+SparkConf sparkConf = new SparkConf().setAppName("JavaGradientBoostedTrees");
+JavaSparkContext sc = new JavaSparkContext(sparkConf);
+
+// Load and parse the data file.
+String datapath = "data/mllib/sample_libsvm_data.txt";
+JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+// Split the data into training and test sets (30% held out for testing)
+JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+JavaRDD<LabeledPoint> trainingData = splits[0];
+JavaRDD<LabeledPoint> testData = splits[1];
+
+// Train a GradientBoostedTrees model.
+//  The defaultParams for Regression use SquaredError by default.
+BoostingStrategy boostingStrategy = BoostingStrategy.defaultParams("Regression");
+boostingStrategy.setNumIterations(3); // Note: Use more iterations in practice.
+boostingStrategy.getTreeStrategy().setMaxDepth(5);
+//  Empty categoricalFeaturesInfo indicates all features are continuous.
+Map<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+boostingStrategy.treeStrategy().setCategoricalFeaturesInfo(categoricalFeaturesInfo);
+
+final GradientBoostedTreesModel model =
+  GradientBoostedTrees.train(trainingData, boostingStrategy);
+
+// Evaluate model on test instances and compute test error
+JavaPairRDD<Double, Double> predictionAndLabel =
+  testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+    @Override
+    public Tuple2<Double, Double> call(LabeledPoint p) {
+      return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
+    }
+  });
+Double testMSE =
+  predictionAndLabel.map(new Function<Tuple2<Double, Double>, Double>() {
+    @Override
+    public Double call(Tuple2<Double, Double> pl) {
+      Double diff = pl._1() - pl._2();
+      return diff * diff;
+    }
+  }).reduce(new Function2<Double, Double, Double>() {
+    @Override
+    public Double call(Double a, Double b) {
+      return a + b;
+    }
+  }) / data.count();
+System.out.println("Test Mean Squared Error: " + testMSE);
+System.out.println("Learned regression GBT model:\n" + model.toDebugString());
+{% endhighlight %}
+</div>
+
+</div>
diff --git a/docs/mllib-guide.md b/docs/mllib-guide.md
index 94fc98ce4fabe..efd7dda310712 100644
--- a/docs/mllib-guide.md
+++ b/docs/mllib-guide.md
@@ -1,6 +1,6 @@
 ---
 layout: global
-title: Machine Learning Library (MLlib)
+title: Machine Learning Library (MLlib) Programming Guide
 ---
 
 MLlib is Spark's scalable machine learning library consisting of common learning algorithms and utilities,
@@ -16,8 +16,9 @@ filtering, dimensionality reduction, as well as underlying optimization primitiv
   * random data generation  
 * [Classification and regression](mllib-classification-regression.html)
   * [linear models (SVMs, logistic regression, linear regression)](mllib-linear-methods.html)
-  * [decision trees](mllib-decision-tree.html)
   * [naive Bayes](mllib-naive-bayes.html)
+  * [decision trees](mllib-decision-tree.html)
+  * [ensembles of trees](mllib-ensembles.html) (Random Forests and Gradient-Boosted Trees)
 * [Collaborative filtering](mllib-collaborative-filtering.html)
   * alternating least squares (ALS)
 * [Clustering](mllib-clustering.html)
@@ -34,6 +35,17 @@ MLlib is under active development.
 The APIs marked `Experimental`/`DeveloperApi` may change in future releases, 
 and the migration guide below will explain all changes between releases.
 
+# spark.ml: The New ML Package
+
+Spark 1.2 includes a new machine learning package called `spark.ml`, currently an alpha component but potentially a successor to `spark.mllib`.  The `spark.ml` package aims to replace the old APIs with a cleaner, more uniform set of APIs which will help users create full machine learning pipelines.
+
+See the **[spark.ml programming guide](ml-guide.html)** for more information on this package.
+
+Users can use algorithms from either of the two packages, but APIs may differ.  Currently, `spark.ml` offers a subset of the algorithms from `spark.mllib`.
+
+Developers should contribute new algorithms to `spark.mllib` and can optionally contribute to `spark.ml`.
+See the `spark.ml` programming guide linked above for more details.
+
 # Dependencies
 
 MLlib uses the linear algebra package [Breeze](http://www.scalanlp.org/),
@@ -60,6 +72,32 @@ To use MLlib in Python, you will need [NumPy](http://www.numpy.org) version 1.4
 
 # Migration Guide
 
+## From 1.1 to 1.2
+
+The only API changes in MLlib v1.2 are in
+[`DecisionTree`](api/scala/index.html#org.apache.spark.mllib.tree.DecisionTree),
+which continues to be an experimental API in MLlib 1.2:
+
+1. *(Breaking change)* The Scala API for classification takes a named argument specifying the number
+of classes.  In MLlib v1.1, this argument was called `numClasses` in Python and
+`numClassesForClassification` in Scala.  In MLlib v1.2, the names are both set to `numClasses`.
+This `numClasses` parameter is specified either via
+[`Strategy`](api/scala/index.html#org.apache.spark.mllib.tree.configuration.Strategy)
+or via [`DecisionTree`](api/scala/index.html#org.apache.spark.mllib.tree.DecisionTree)
+static `trainClassifier` and `trainRegressor` methods.
+
+2. *(Breaking change)* The API for
+[`Node`](api/scala/index.html#org.apache.spark.mllib.tree.model.Node) has changed.
+This should generally not affect user code, unless the user manually constructs decision trees
+(instead of using the `trainClassifier` or `trainRegressor` methods).
+The tree `Node` now includes more information, including the probability of the predicted label
+(for classification).
+
+3. Printing methods' output has changed.  The `toString` (Scala/Java) and `__repr__` (Python) methods used to print the full model; they now print a summary.  For the full model, use `toDebugString`.
+
+Examples in the Spark distribution and examples in the
+[Decision Trees Guide](mllib-decision-tree.html#examples) have been updated accordingly.
+
 ## From 1.0 to 1.1
 
 The only API changes in MLlib v1.1 are in
diff --git a/docs/mllib-linear-methods.md b/docs/mllib-linear-methods.md
index bc914a1899801..44b7f67c57734 100644
--- a/docs/mllib-linear-methods.md
+++ b/docs/mllib-linear-methods.md
@@ -110,12 +110,16 @@ However, L1 regularization can help promote sparsity in weights leading to small
 It is not recommended to train models without any regularization,
 especially when the number of training examples is small.
 
+### Optimization
+
+Under the hood, linear methods use convex optimization methods to optimize the objective functions.  MLlib uses two methods, SGD and L-BFGS, described in the [optimization section](mllib-optimization.html).  Currently, most algorithm APIs support Stochastic Gradient Descent (SGD), and a few support L-BFGS. Refer to [this optimization section](mllib-optimization.html#Choosing-an-Optimization-Method) for guidelines on choosing between optimization methods.
+
 ## Binary classification
 
 [Binary classification](http://en.wikipedia.org/wiki/Binary_classification)
 aims to divide items into two categories: positive and negative.  MLlib
-supports two linear methods for binary classification: linear support vector
-machines (SVMs) and logistic regression. For both methods, MLlib supports
+supports two linear methods for binary classification: linear Support Vector
+Machines (SVMs) and logistic regression. For both methods, MLlib supports
 L1 and L2 regularized variants. The training data set is represented by an RDD
 of [LabeledPoint](mllib-data-types.html) in MLlib.  Note that, in the
 mathematical formulation in this guide, a training label $y$ is denoted as
@@ -123,7 +127,7 @@ either $+1$ (positive) or $-1$ (negative), which is convenient for the
 formulation.  *However*, the negative label is represented by $0$ in MLlib
 instead of $-1$, to be consistent with multiclass labeling.
 
-### Linear support vector machines (SVMs)
+### Linear Support Vector Machines (SVMs)
 
 The [linear SVM](http://en.wikipedia.org/wiki/Support_vector_machine#Linear_SVM)
 is a standard method for large-scale classification tasks. It is a linear method as described above in equation `$\eqref{eq:regPrimal}$`, with the loss function in the formulation given by the hinge loss:
diff --git a/docs/mllib-optimization.md b/docs/mllib-optimization.md
index 45141c235be90..4d101afca2c97 100644
--- a/docs/mllib-optimization.md
+++ b/docs/mllib-optimization.md
@@ -138,6 +138,12 @@ vertical scalability issue (the number of training features) when computing the
 explicitly in Newton's method. As a result, L-BFGS often achieves rapider convergence compared with 
 other first-order optimization. 
 
+### Choosing an Optimization Method
+
+[Linear methods](mllib-linear-methods.html) use optimization internally, and some linear methods in MLlib support both SGD and L-BFGS.
+Different optimization methods can have different convergence guarantees depending on the properties of the objective function, and we cannot cover the literature here.
+In general, when L-BFGS is available, we recommend using it instead of SGD since L-BFGS tends to converge faster (in fewer iterations).
+
 ## Implementation in MLlib
 
 ### Gradient descent and stochastic gradient descent
@@ -168,10 +174,7 @@ descent. All updaters in MLlib use a step size at the t-th step equal to
 * `regParam` is the regularization parameter when using L1 or L2 regularization.
 * `miniBatchFraction` is the fraction of the total data that is sampled in 
 each iteration, to compute the gradient direction.
-
-Available algorithms for gradient descent:
-
-* [GradientDescent](api/scala/index.html#org.apache.spark.mllib.optimization.GradientDescent)
+  * Sampling still requires a pass over the entire RDD, so decreasing `miniBatchFraction` may not speed up optimization much.  Users will see the greatest speedup when the gradient is expensive to compute, for only the chosen samples are used for computing the gradient.
 
 ### L-BFGS
 L-BFGS is currently only a low-level optimization primitive in `MLlib`. If you want to use L-BFGS in various 
@@ -359,13 +362,15 @@ public class LBFGSExample {
 {% endhighlight %}
 </div>
 </div>
-#### Developer's note
+
+## Developer's notes
+
 Since the Hessian is constructed approximately from previous gradient evaluations, 
 the objective function can not be changed during the optimization process. 
 As a result, Stochastic L-BFGS will not work naively by just using miniBatch; 
 therefore, we don't provide this until we have better understanding.
 
-* `Updater` is a class originally designed for gradient decent which computes 
+`Updater` is a class originally designed for gradient decent which computes 
 the actual gradient descent step. However, we're able to take the gradient and 
 loss of objective function of regularization for L-BFGS by ignoring the part of logic
 only for gradient decent such as adaptive step size stuff. We will refactorize
diff --git a/docs/programming-guide.md b/docs/programming-guide.md
index c60de6e970531..5e0d5c15d7069 100644
--- a/docs/programming-guide.md
+++ b/docs/programming-guide.md
@@ -934,6 +934,12 @@ for details.
   <td> Reshuffle the data in the RDD randomly to create either more or fewer partitions and balance it across them.
     This always shuffles all data over the network. </td>
 </tr>
+<tr>
+  <td> <b>repartitionAndSortWithinPartitions</b>(<i>partitioner</i>) </td>
+  <td> Repartition the RDD according to the given partitioner and, within each resulting partition,
+  sort records by their keys. This is more efficient than calling <code>repartition</code> and then sorting within 
+  each partition because it can push the sorting down into the shuffle machinery. </td>
+</tr>
 </table>
 
 ### Actions
@@ -1177,7 +1183,7 @@ Accumulators are variables that are only "added" to through an associative opera
 therefore be efficiently supported in parallel. They can be used to implement counters (as in
 MapReduce) or sums. Spark natively supports accumulators of numeric types, and programmers
 can add support for new types. If accumulators are created with a name, they will be
-displayed in Spark's UI. This can can be useful for understanding the progress of 
+displayed in Spark's UI. This can be useful for understanding the progress of 
 running stages (NOTE: this is not yet supported in Python).
 
 An accumulator is created from an initial value `v` by calling `SparkContext.accumulator(v)`. Tasks
diff --git a/docs/running-on-yarn.md b/docs/running-on-yarn.md
index dfe2db4b3fce8..16897dbb65311 100644
--- a/docs/running-on-yarn.md
+++ b/docs/running-on-yarn.md
@@ -30,7 +30,7 @@ Most of the configs are the same for Spark on YARN as for other deployment modes
 </tr>
 <tr>
   <td><code>spark.yarn.submit.file.replication</code></td>
-  <td>3</td>
+  <td>The default HDFS replication (usually 3)</td>
   <td>
     HDFS replication level for the files uploaded into HDFS for the application. These include things like the Spark jar, the app jar, and any distributed cache files/archives.
   </td>
@@ -91,6 +91,13 @@ Most of the configs are the same for Spark on YARN as for other deployment modes
     The amount of off heap memory (in megabytes) to be allocated per driver. This is memory that accounts for things like VM overheads, interned strings, other native overheads, etc. This tends to grow with the container size (typically 6-10%).
   </td>
 </tr>
+<tr>
+  <td><code>spark.yarn.queue</code></td>
+  <td>default</td>
+  <td>
+    The name of the YARN queue to which the application is submitted.
+  </td>
+</tr>
 <tr>
   <td><code>spark.yarn.jar</code></td>
   <td>(none)</td>
diff --git a/docs/spark-standalone.md b/docs/spark-standalone.md
index a3028aa86dc45..ae7b81d5bb71f 100644
--- a/docs/spark-standalone.md
+++ b/docs/spark-standalone.md
@@ -10,7 +10,7 @@ In addition to running on the Mesos or YARN cluster managers, Spark also provide
 
 # Installing Spark Standalone to a Cluster
 
-To install Spark Standalone mode, you simply place a compiled version of Spark on each node on the cluster. You can obtain pre-built versions of Spark with each release or [build it yourself](index.html#building).
+To install Spark Standalone mode, you simply place a compiled version of Spark on each node on the cluster. You can obtain pre-built versions of Spark with each release or [build it yourself](building-spark.html).
 
 # Starting a Cluster Manually
 
@@ -34,8 +34,12 @@ Finally, the following configuration options can be passed to the master and wor
 <table class="table">
   <tr><th style="width:21%">Argument</th><th>Meaning</th></tr>
   <tr>
-    <td><code>-i IP</code>, <code>--ip IP</code></td>
-    <td>IP address or DNS name to listen on</td>
+    <td><code>-h HOST</code>, <code>--host HOST</code></td>
+    <td>Hostname to listen on</td>
+  </tr>
+  <tr>
+    <td><code>-i HOST</code>, <code>--ip HOST</code></td>
+    <td>Hostname to listen on (deprecated, use -h or --host)</td>
   </tr>
   <tr>
     <td><code>-p PORT</code>, <code>--port PORT</code></td>
@@ -57,6 +61,10 @@ Finally, the following configuration options can be passed to the master and wor
     <td><code>-d DIR</code>, <code>--work-dir DIR</code></td>
     <td>Directory to use for scratch space and job output logs (default: SPARK_HOME/work); only on worker</td>
   </tr>
+  <tr>
+    <td><code>--properties-file FILE</code></td>
+    <td>Path to a custom Spark properties file to load (default: conf/spark-defaults.conf)</td>
+  </tr>
 </table>
 
 
diff --git a/docs/sql-programming-guide.md b/docs/sql-programming-guide.md
index 5500da83b2b66..be284fbe217a5 100644
--- a/docs/sql-programming-guide.md
+++ b/docs/sql-programming-guide.md
@@ -146,7 +146,7 @@ describes the various methods for loading data into a SchemaRDD.
 
 Spark SQL supports two different methods for converting existing RDDs into SchemaRDDs.  The first
 method uses reflection to infer the schema of an RDD that contains specific types of objects.  This
-reflection based approach leads to more concise code and works well when you already know the schema 
+reflection based approach leads to more concise code and works well when you already know the schema
 while writing your Spark application.
 
 The second method for creating SchemaRDDs is through a programmatic interface that allows you to
@@ -278,7 +278,7 @@ performed on JSON files.
 from pyspark.sql import SQLContext, Row
 sqlContext = SQLContext(sc)
 
-# Load a text file and convert each line to a dictionary.
+# Load a text file and convert each line to a Row.
 lines = sc.textFile("examples/src/main/resources/people.txt")
 parts = lines.map(lambda l: l.split(","))
 people = parts.map(lambda p: Row(name=p[0], age=int(p[1])))
@@ -566,7 +566,7 @@ for teenName in teenNames.collect():
 
 ### Configuration
 
-Configuration of Parquet can be done using the `setConf` method on SQLContext or by running 
+Configuration of Parquet can be done using the `setConf` method on SQLContext or by running
 `SET key=value` commands using SQL.
 
 <table class="table">
@@ -575,8 +575,8 @@ Configuration of Parquet can be done using the `setConf` method on SQLContext or
   <td><code>spark.sql.parquet.binaryAsString</code></td>
   <td>false</td>
   <td>
-    Some other Parquet-producing systems, in particular Impala and older versions of Spark SQL, do 
-    not differentiate between binary data and strings when writing out the Parquet schema.  This 
+    Some other Parquet-producing systems, in particular Impala and older versions of Spark SQL, do
+    not differentiate between binary data and strings when writing out the Parquet schema.  This
     flag tells Spark SQL to interpret binary data as a string to provide compatibility with these systems.
   </td>
 </tr>
@@ -591,10 +591,20 @@ Configuration of Parquet can be done using the `setConf` method on SQLContext or
   <td><code>spark.sql.parquet.compression.codec</code></td>
   <td>gzip</td>
   <td>
-    Sets the compression codec use when writing Parquet files. Acceptable values include: 
+    Sets the compression codec use when writing Parquet files. Acceptable values include:
     uncompressed, snappy, gzip, lzo.
   </td>
 </tr>
+<tr>
+  <td><code>spark.sql.parquet.filterPushdown</code></td>
+  <td>false</td>
+  <td>
+    Turn on Parquet filter pushdown optimization. This feature is turned off by default because of a known
+    bug in Paruet 1.6.0rc3 (<a href="https://issues.apache.org/jira/browse/PARQUET-136">PARQUET-136</a>).
+    However, if your table doesn't contain any nullable string or binary columns, it's still safe to turn
+    this feature on.
+  </td>
+</tr>
 <tr>
   <td><code>spark.sql.hive.convertMetastoreParquet</code></td>
   <td>true</td>
@@ -900,7 +910,6 @@ export HIVE_SERVER2_THRIFT_BIND_HOST=<listening-host>
 ./sbin/start-thriftserver.sh \
   --master <master-uri> \
   ...
-```
 {% endhighlight %}
 
 or system properties:
@@ -911,7 +920,6 @@ or system properties:
   --hiveconf hive.server2.thrift.bind.host=<listening-host> \
   --master <master-uri>
   ...
-```
 {% endhighlight %}
 
 Now you can use beeline to test the Thrift JDBC/ODBC server:
@@ -947,7 +955,7 @@ options.
 
 ## Migration Guide for Shark User
 
-### Scheduling 
+### Scheduling
 To set a [Fair Scheduler](job-scheduling.html#fair-scheduler-pools) pool for a JDBC client session,
 users can set the `spark.sql.thriftserver.scheduler.pool` variable:
 
@@ -994,7 +1002,7 @@ Several caching related features are not supported yet:
 ## Compatibility with Apache Hive
 
 Spark SQL is designed to be compatible with the Hive Metastore, SerDes and UDFs.  Currently Spark
-SQL is based on Hive 0.12.0.
+SQL is based on Hive 0.12.0 and 0.13.1.
 
 #### Deploying in Existing Hive Warehouses
 
@@ -1033,6 +1041,7 @@ Spark SQL supports the vast majority of Hive features, such as:
 * Sampling
 * Explain
 * Partitioned tables
+* View
 * All Hive DDL Functions, including:
   * `CREATE TABLE`
   * `CREATE TABLE AS SELECT`
@@ -1048,6 +1057,7 @@ Spark SQL supports the vast majority of Hive features, such as:
   * `STRING`
   * `BINARY`
   * `TIMESTAMP`
+  * `DATE`
   * `ARRAY<>`
   * `MAP<>`
   * `STRUCT<>`
@@ -1148,6 +1158,7 @@ evaluated by the SQL execution engine.  A full list of the functions supported c
 * Datetime type
     - `TimestampType`: Represents values comprising values of fields year, month, day,
     hour, minute, and second.
+    - `DateType`: Represents values comprising values of fields year, month, day.
 * Complex types
     - `ArrayType(elementType, containsNull)`: Represents values comprising a sequence of
     elements with the type of `elementType`. `containsNull` is used to indicate if
@@ -1255,6 +1266,13 @@ import  org.apache.spark.sql._
   TimestampType
   </td>
 </tr>
+<tr>
+  <td> <b>DateType</b> </td>
+  <td> java.sql.Date </td>
+  <td>
+  DateType
+  </td>
+</tr>
 <tr>
   <td> <b>ArrayType</b> </td>
   <td> scala.collection.Seq </td>
@@ -1381,6 +1399,13 @@ please use factory methods provided in
   DataType.TimestampType
   </td>
 </tr>
+<tr>
+  <td> <b>DateType</b> </td>
+  <td> java.sql.Date </td>
+  <td>
+  DataType.DateType
+  </td>
+</tr>
 <tr>
   <td> <b>ArrayType</b> </td>
   <td> java.util.List </td>
@@ -1528,6 +1553,13 @@ from pyspark.sql import *
   TimestampType()
   </td>
 </tr>
+<tr>
+  <td> <b>DateType</b> </td>
+  <td> datetime.date </td>
+  <td>
+  DateType()
+  </td>
+</tr>
 <tr>
   <td> <b>ArrayType</b> </td>
   <td> list, tuple, or array </td>
diff --git a/docs/streaming-programming-guide.md b/docs/streaming-programming-guide.md
index 44a1f3ad7560b..5ebe834a32d31 100644
--- a/docs/streaming-programming-guide.md
+++ b/docs/streaming-programming-guide.md
@@ -1081,7 +1081,7 @@ Some of the common mistakes to avoid are as follows.
 
 - Often writing data to external system requires creating a connection object
 (e.g. TCP connection to a remote server) and using it to send data to a remote system.
-For this purpose, a developer may inadvertantly try creating a connection object at
+For this purpose, a developer may inadvertently try creating a connection object at
 the Spark driver, but try to use it in a Spark worker to save records in the RDDs.
 For example (in Scala),
 
diff --git a/docs/tuning.md b/docs/tuning.md
index 9b5c9adac6a4f..0e2447dd46394 100644
--- a/docs/tuning.md
+++ b/docs/tuning.md
@@ -143,8 +143,7 @@ the space allocated to the RDD cache to mitigate this.
 **Measuring the Impact of GC**
 
 The first step in GC tuning is to collect statistics on how frequently garbage collection occurs and the amount of
-time spent GC. This can be done by adding `-verbose:gc -XX:+PrintGCDetails -XX:+PrintGCTimeStamps` to your
-`SPARK_JAVA_OPTS` environment variable. Next time your Spark job is run, you will see messages printed in the worker's logs
+time spent GC. This can be done by adding `-verbose:gc -XX:+PrintGCDetails -XX:+PrintGCTimeStamps` to the Java options.  (See the [configuration guide](configuration.html#Dynamically-Loading-Spark-Properties) for info on passing Java options to Spark jobs.)  Next time your Spark job is run, you will see messages printed in the worker's logs
 each time a garbage collection occurs. Note these logs will be on your cluster's worker nodes (in the `stdout` files in
 their work directories), *not* on your driver program.
 
diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py
index 742c7765e728e..5f9e484212635 100755
--- a/ec2/spark_ec2.py
+++ b/ec2/spark_ec2.py
@@ -33,6 +33,7 @@
 import time
 import urllib2
 import warnings
+from datetime import datetime
 from optparse import OptionParser
 from sys import stderr
 import boto
@@ -503,9 +504,9 @@ def get_existing_cluster(conn, opts, cluster_name, die_on_error=True):
         active = [i for i in res.instances if is_active(i)]
         for inst in active:
             group_names = [g.name for g in inst.groups]
-            if group_names == [cluster_name + "-master"]:
+            if (cluster_name + "-master") in group_names:
                 master_nodes.append(inst)
-            elif group_names == [cluster_name + "-slaves"]:
+            elif (cluster_name + "-slaves") in group_names:
                 slave_nodes.append(inst)
     if any((master_nodes, slave_nodes)):
         print "Found %d master(s), %d slaves" % (len(master_nodes), len(slave_nodes))
@@ -589,7 +590,9 @@ def setup_spark_cluster(master, opts):
 
 
 def is_ssh_available(host, opts):
-    "Checks if SSH is available on the host."
+    """
+    Check if SSH is available on a host.
+    """
     try:
         with open(os.devnull, 'w') as devnull:
             ret = subprocess.check_call(
@@ -604,6 +607,9 @@ def is_ssh_available(host, opts):
 
 
 def is_cluster_ssh_available(cluster_instances, opts):
+    """
+    Check if SSH is available on all the instances in a cluster.
+    """
     for i in cluster_instances:
         if not is_ssh_available(host=i.ip_address, opts=opts):
             return False
@@ -611,8 +617,10 @@ def is_cluster_ssh_available(cluster_instances, opts):
         return True
 
 
-def wait_for_cluster_state(cluster_instances, cluster_state, opts):
+def wait_for_cluster_state(conn, opts, cluster_instances, cluster_state):
     """
+    Wait for all the instances in the cluster to reach a designated state.
+
     cluster_instances: a list of boto.ec2.instance.Instance
     cluster_state: a string representing the desired state of all the instances in the cluster
            value can be 'ssh-ready' or a valid value from boto.ec2.instance.InstanceState such as
@@ -620,20 +628,27 @@ def wait_for_cluster_state(cluster_instances, cluster_state, opts):
            (would be nice to replace this with a proper enum: http://stackoverflow.com/a/1695250)
     """
     sys.stdout.write(
-        "Waiting for all instances in cluster to enter '{s}' state.".format(s=cluster_state)
+        "Waiting for cluster to enter '{s}' state.".format(s=cluster_state)
     )
     sys.stdout.flush()
 
+    start_time = datetime.now()
+
     num_attempts = 0
+    conn = ec2.connect_to_region(opts.region)
 
     while True:
-        time.sleep(3 * num_attempts)
+        time.sleep(5 * num_attempts)  # seconds
 
         for i in cluster_instances:
-            s = i.update()  # capture output to suppress print to screen in newer versions of boto
+            i.update()
+
+        statuses = conn.get_all_instance_status(instance_ids=[i.id for i in cluster_instances])
 
         if cluster_state == 'ssh-ready':
             if all(i.state == 'running' for i in cluster_instances) and \
+               all(s.system_status.status == 'ok' for s in statuses) and \
+               all(s.instance_status.status == 'ok' for s in statuses) and \
                is_cluster_ssh_available(cluster_instances, opts):
                 break
         else:
@@ -647,6 +662,12 @@ def wait_for_cluster_state(cluster_instances, cluster_state, opts):
 
     sys.stdout.write("\n")
 
+    end_time = datetime.now()
+    print "Cluster is now in '{s}' state. Waited {t} seconds.".format(
+        s=cluster_state,
+        t=(end_time - start_time).seconds
+    )
+
 
 # Get number of local disks available for a given EC2 instance type.
 def get_num_disks(instance_type):
@@ -895,7 +916,7 @@ def real_main():
         # See: https://docs.python.org/3.5/whatsnew/2.7.html
         warnings.warn(
             "This option is deprecated and has no effect. "
-            "spark-ec2 automatically waits as long as necessary for clusters to startup.",
+            "spark-ec2 automatically waits as long as necessary for clusters to start up.",
             DeprecationWarning
         )
 
@@ -922,9 +943,10 @@ def real_main():
         else:
             (master_nodes, slave_nodes) = launch_cluster(conn, opts, cluster_name)
         wait_for_cluster_state(
+            conn=conn,
+            opts=opts,
             cluster_instances=(master_nodes + slave_nodes),
-            cluster_state='ssh-ready',
-            opts=opts
+            cluster_state='ssh-ready'
         )
         setup_cluster(conn, master_nodes, slave_nodes, opts, True)
 
@@ -951,9 +973,10 @@ def real_main():
                 print "Deleting security groups (this will take some time)..."
                 group_names = [cluster_name + "-master", cluster_name + "-slaves"]
                 wait_for_cluster_state(
+                    conn=conn,
+                    opts=opts,
                     cluster_instances=(master_nodes + slave_nodes),
-                    cluster_state='terminated',
-                    opts=opts
+                    cluster_state='terminated'
                 )
                 attempt = 1
                 while attempt <= 3:
@@ -1055,9 +1078,10 @@ def real_main():
             if inst.state not in ["shutting-down", "terminated"]:
                 inst.start()
         wait_for_cluster_state(
+            conn=conn,
+            opts=opts,
             cluster_instances=(master_nodes + slave_nodes),
-            cluster_state='ssh-ready',
-            opts=opts
+            cluster_state='ssh-ready'
         )
         setup_cluster(conn, master_nodes, slave_nodes, opts, False)
 
diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java
new file mode 100644
index 0000000000000..f4b4f8d8c7b2f
--- /dev/null
+++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaCrossValidatorExample.java
@@ -0,0 +1,126 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.examples.ml;
+
+import java.util.List;
+
+import com.google.common.collect.Lists;
+
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.ml.Pipeline;
+import org.apache.spark.ml.PipelineStage;
+import org.apache.spark.ml.classification.LogisticRegression;
+import org.apache.spark.ml.evaluation.BinaryClassificationEvaluator;
+import org.apache.spark.ml.feature.HashingTF;
+import org.apache.spark.ml.feature.Tokenizer;
+import org.apache.spark.ml.param.ParamMap;
+import org.apache.spark.ml.tuning.CrossValidator;
+import org.apache.spark.ml.tuning.CrossValidatorModel;
+import org.apache.spark.ml.tuning.ParamGridBuilder;
+import org.apache.spark.sql.api.java.JavaSQLContext;
+import org.apache.spark.sql.api.java.JavaSchemaRDD;
+import org.apache.spark.sql.api.java.Row;
+
+/**
+ * A simple example demonstrating model selection using CrossValidator.
+ * This example also demonstrates how Pipelines are Estimators.
+ *
+ * This example uses the Java bean classes {@link org.apache.spark.examples.ml.LabeledDocument} and
+ * {@link org.apache.spark.examples.ml.Document} defined in the Scala example
+ * {@link org.apache.spark.examples.ml.SimpleTextClassificationPipeline}.
+ *
+ * Run with
+ * <pre>
+ * bin/run-example ml.JavaCrossValidatorExample
+ * </pre>
+ */
+public class JavaCrossValidatorExample {
+
+  public static void main(String[] args) {
+    SparkConf conf = new SparkConf().setAppName("JavaCrossValidatorExample");
+    JavaSparkContext jsc = new JavaSparkContext(conf);
+    JavaSQLContext jsql = new JavaSQLContext(jsc);
+
+    // Prepare training documents, which are labeled.
+    List<LabeledDocument> localTraining = Lists.newArrayList(
+      new LabeledDocument(0L, "a b c d e spark", 1.0),
+      new LabeledDocument(1L, "b d", 0.0),
+      new LabeledDocument(2L, "spark f g h", 1.0),
+      new LabeledDocument(3L, "hadoop mapreduce", 0.0),
+      new LabeledDocument(4L, "b spark who", 1.0),
+      new LabeledDocument(5L, "g d a y", 0.0),
+      new LabeledDocument(6L, "spark fly", 1.0),
+      new LabeledDocument(7L, "was mapreduce", 0.0),
+      new LabeledDocument(8L, "e spark program", 1.0),
+      new LabeledDocument(9L, "a e c l", 0.0),
+      new LabeledDocument(10L, "spark compile", 1.0),
+      new LabeledDocument(11L, "hadoop software", 0.0));
+    JavaSchemaRDD training =
+        jsql.applySchema(jsc.parallelize(localTraining), LabeledDocument.class);
+
+    // Configure an ML pipeline, which consists of three stages: tokenizer, hashingTF, and lr.
+    Tokenizer tokenizer = new Tokenizer()
+      .setInputCol("text")
+      .setOutputCol("words");
+    HashingTF hashingTF = new HashingTF()
+      .setNumFeatures(1000)
+      .setInputCol(tokenizer.getOutputCol())
+      .setOutputCol("features");
+    LogisticRegression lr = new LogisticRegression()
+      .setMaxIter(10)
+      .setRegParam(0.01);
+    Pipeline pipeline = new Pipeline()
+      .setStages(new PipelineStage[] {tokenizer, hashingTF, lr});
+
+    // We now treat the Pipeline as an Estimator, wrapping it in a CrossValidator instance.
+    // This will allow us to jointly choose parameters for all Pipeline stages.
+    // A CrossValidator requires an Estimator, a set of Estimator ParamMaps, and an Evaluator.
+    CrossValidator crossval = new CrossValidator()
+        .setEstimator(pipeline)
+        .setEvaluator(new BinaryClassificationEvaluator());
+    // We use a ParamGridBuilder to construct a grid of parameters to search over.
+    // With 3 values for hashingTF.numFeatures and 2 values for lr.regParam,
+    // this grid will have 3 x 2 = 6 parameter settings for CrossValidator to choose from.
+    ParamMap[] paramGrid = new ParamGridBuilder()
+        .addGrid(hashingTF.numFeatures(), new int[]{10, 100, 1000})
+        .addGrid(lr.regParam(), new double[]{0.1, 0.01})
+        .build();
+    crossval.setEstimatorParamMaps(paramGrid);
+    crossval.setNumFolds(2); // Use 3+ in practice
+
+    // Run cross-validation, and choose the best set of parameters.
+    CrossValidatorModel cvModel = crossval.fit(training);
+
+    // Prepare test documents, which are unlabeled.
+    List<Document> localTest = Lists.newArrayList(
+      new Document(4L, "spark i j k"),
+      new Document(5L, "l m n"),
+      new Document(6L, "mapreduce spark"),
+      new Document(7L, "apache hadoop"));
+    JavaSchemaRDD test = jsql.applySchema(jsc.parallelize(localTest), Document.class);
+
+    // Make predictions on test documents. cvModel uses the best model found (lrModel).
+    cvModel.transform(test).registerAsTable("prediction");
+    JavaSchemaRDD predictions = jsql.sql("SELECT id, text, score, prediction FROM prediction");
+    for (Row r: predictions.collect()) {
+      System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> score=" + r.get(2)
+          + ", prediction=" + r.get(3));
+    }
+  }
+}
diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java
new file mode 100644
index 0000000000000..e25b271777ed4
--- /dev/null
+++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleParamsExample.java
@@ -0,0 +1,111 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.examples.ml;
+
+import java.util.List;
+
+import com.google.common.collect.Lists;
+
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.ml.classification.LogisticRegressionModel;
+import org.apache.spark.ml.param.ParamMap;
+import org.apache.spark.ml.classification.LogisticRegression;
+import org.apache.spark.mllib.linalg.Vectors;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.sql.api.java.JavaSQLContext;
+import org.apache.spark.sql.api.java.JavaSchemaRDD;
+import org.apache.spark.sql.api.java.Row;
+
+/**
+ * A simple example demonstrating ways to specify parameters for Estimators and Transformers.
+ * Run with
+ * {{{
+ * bin/run-example ml.JavaSimpleParamsExample
+ * }}}
+ */
+public class JavaSimpleParamsExample {
+
+  public static void main(String[] args) {
+    SparkConf conf = new SparkConf().setAppName("JavaSimpleParamsExample");
+    JavaSparkContext jsc = new JavaSparkContext(conf);
+    JavaSQLContext jsql = new JavaSQLContext(jsc);
+
+    // Prepare training data.
+    // We use LabeledPoint, which is a JavaBean.  Spark SQL can convert RDDs of JavaBeans
+    // into SchemaRDDs, where it uses the bean metadata to infer the schema.
+    List<LabeledPoint> localTraining = Lists.newArrayList(
+      new LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)),
+      new LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)),
+      new LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)),
+      new LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5)));
+    JavaSchemaRDD training = jsql.applySchema(jsc.parallelize(localTraining), LabeledPoint.class);
+
+    // Create a LogisticRegression instance.  This instance is an Estimator.
+    LogisticRegression lr = new LogisticRegression();
+    // Print out the parameters, documentation, and any default values.
+    System.out.println("LogisticRegression parameters:\n" + lr.explainParams() + "\n");
+
+    // We may set parameters using setter methods.
+    lr.setMaxIter(10)
+      .setRegParam(0.01);
+
+    // Learn a LogisticRegression model.  This uses the parameters stored in lr.
+    LogisticRegressionModel model1 = lr.fit(training);
+    // Since model1 is a Model (i.e., a Transformer produced by an Estimator),
+    // we can view the parameters it used during fit().
+    // This prints the parameter (name: value) pairs, where names are unique IDs for this
+    // LogisticRegression instance.
+    System.out.println("Model 1 was fit using parameters: " + model1.fittingParamMap());
+
+    // We may alternatively specify parameters using a ParamMap.
+    ParamMap paramMap = new ParamMap();
+    paramMap.put(lr.maxIter().w(20)); // Specify 1 Param.
+    paramMap.put(lr.maxIter(), 30); // This overwrites the original maxIter.
+    paramMap.put(lr.regParam().w(0.1), lr.threshold().w(0.55)); // Specify multiple Params.
+
+    // One can also combine ParamMaps.
+    ParamMap paramMap2 = new ParamMap();
+    paramMap2.put(lr.scoreCol().w("probability")); // Change output column name
+    ParamMap paramMapCombined = paramMap.$plus$plus(paramMap2);
+
+    // Now learn a new model using the paramMapCombined parameters.
+    // paramMapCombined overrides all parameters set earlier via lr.set* methods.
+    LogisticRegressionModel model2 = lr.fit(training, paramMapCombined);
+    System.out.println("Model 2 was fit using parameters: " + model2.fittingParamMap());
+
+    // Prepare test documents.
+    List<LabeledPoint> localTest = Lists.newArrayList(
+        new LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)),
+        new LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)),
+        new LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5)));
+    JavaSchemaRDD test = jsql.applySchema(jsc.parallelize(localTest), LabeledPoint.class);
+
+    // Make predictions on test documents using the Transformer.transform() method.
+    // LogisticRegression.transform will only use the 'features' column.
+    // Note that model2.transform() outputs a 'probability' column instead of the usual 'score'
+    // column since we renamed the lr.scoreCol parameter previously.
+    model2.transform(test).registerAsTable("results");
+    JavaSchemaRDD results =
+        jsql.sql("SELECT features, label, probability, prediction FROM results");
+    for (Row r: results.collect()) {
+      System.out.println("(" + r.get(0) + ", " + r.get(1) + ") -> prob=" + r.get(2)
+          + ", prediction=" + r.get(3));
+    }
+  }
+}
diff --git a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java
index 22ba68d8c354c..54f18014e4b2f 100644
--- a/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java
+++ b/examples/src/main/java/org/apache/spark/examples/ml/JavaSimpleTextClassificationPipeline.java
@@ -80,14 +80,14 @@ public static void main(String[] args) {
       new Document(5L, "l m n"),
       new Document(6L, "mapreduce spark"),
       new Document(7L, "apache hadoop"));
-    JavaSchemaRDD test =
-      jsql.applySchema(jsc.parallelize(localTest), Document.class);
+    JavaSchemaRDD test = jsql.applySchema(jsc.parallelize(localTest), Document.class);
 
     // Make predictions on test documents.
     model.transform(test).registerAsTable("prediction");
     JavaSchemaRDD predictions = jsql.sql("SELECT id, text, score, prediction FROM prediction");
     for (Row r: predictions.collect()) {
-      System.out.println(r);
+      System.out.println("(" + r.get(0) + ", " + r.get(1) + ") --> score=" + r.get(2)
+          + ", prediction=" + r.get(3));
     }
   }
 }
diff --git a/examples/src/main/java/org/apache/spark/examples/mllib/JavaGradientBoostedTreesRunner.java b/examples/src/main/java/org/apache/spark/examples/mllib/JavaGradientBoostedTreesRunner.java
index 4a5ac404ea5ea..a1844d5d07ad4 100644
--- a/examples/src/main/java/org/apache/spark/examples/mllib/JavaGradientBoostedTreesRunner.java
+++ b/examples/src/main/java/org/apache/spark/examples/mllib/JavaGradientBoostedTreesRunner.java
@@ -73,7 +73,7 @@ public static void main(String[] args) {
           return p.label();
         }
       }).countByValue().size();
-      boostingStrategy.treeStrategy().setNumClassesForClassification(numClasses);
+      boostingStrategy.treeStrategy().setNumClasses(numClasses);
 
       // Train a GradientBoosting model for classification.
       final GradientBoostedTreesModel model = GradientBoostedTrees.train(data, boostingStrategy);
diff --git a/examples/src/main/java/org/apache/spark/examples/mllib/JavaRandomForestExample.java b/examples/src/main/java/org/apache/spark/examples/mllib/JavaRandomForestExample.java
new file mode 100644
index 0000000000000..89a4e092a5af7
--- /dev/null
+++ b/examples/src/main/java/org/apache/spark/examples/mllib/JavaRandomForestExample.java
@@ -0,0 +1,139 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.examples.mllib;
+
+import scala.Tuple2;
+
+import java.util.HashMap;
+
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.java.JavaPairRDD;
+import org.apache.spark.api.java.JavaRDD;
+import org.apache.spark.api.java.JavaSparkContext;
+import org.apache.spark.api.java.function.Function;
+import org.apache.spark.api.java.function.Function2;
+import org.apache.spark.api.java.function.PairFunction;
+import org.apache.spark.mllib.regression.LabeledPoint;
+import org.apache.spark.mllib.tree.RandomForest;
+import org.apache.spark.mllib.tree.model.RandomForestModel;
+import org.apache.spark.mllib.util.MLUtils;
+
+public final class JavaRandomForestExample {
+
+  /**
+   * Note: This example illustrates binary classification.
+   * For information on multiclass classification, please refer to the JavaDecisionTree.java
+   * example.
+   */
+  private static void testClassification(JavaRDD<LabeledPoint> trainingData,
+                                         JavaRDD<LabeledPoint> testData) {
+    // Train a RandomForest model.
+    //  Empty categoricalFeaturesInfo indicates all features are continuous.
+    Integer numClasses = 2;
+    HashMap<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+    Integer numTrees = 3; // Use more in practice.
+    String featureSubsetStrategy = "auto"; // Let the algorithm choose.
+    String impurity = "gini";
+    Integer maxDepth = 4;
+    Integer maxBins = 32;
+    Integer seed = 12345;
+
+    final RandomForestModel model = RandomForest.trainClassifier(trainingData, numClasses,
+        categoricalFeaturesInfo, numTrees, featureSubsetStrategy, impurity, maxDepth, maxBins,
+        seed);
+
+    // Evaluate model on test instances and compute test error
+    JavaPairRDD<Double, Double> predictionAndLabel =
+        testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+          @Override
+          public Tuple2<Double, Double> call(LabeledPoint p) {
+            return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
+          }
+        });
+    Double testErr =
+        1.0 * predictionAndLabel.filter(new Function<Tuple2<Double, Double>, Boolean>() {
+          @Override
+          public Boolean call(Tuple2<Double, Double> pl) {
+            return !pl._1().equals(pl._2());
+          }
+        }).count() / testData.count();
+    System.out.println("Test Error: " + testErr);
+    System.out.println("Learned classification forest model:\n" + model.toDebugString());
+  }
+
+  private static void testRegression(JavaRDD<LabeledPoint> trainingData,
+                                     JavaRDD<LabeledPoint> testData) {
+    // Train a RandomForest model.
+    //  Empty categoricalFeaturesInfo indicates all features are continuous.
+    HashMap<Integer, Integer> categoricalFeaturesInfo = new HashMap<Integer, Integer>();
+    Integer numTrees = 3; // Use more in practice.
+    String featureSubsetStrategy = "auto"; // Let the algorithm choose.
+    String impurity = "variance";
+    Integer maxDepth = 4;
+    Integer maxBins = 32;
+    Integer seed = 12345;
+
+    final RandomForestModel model = RandomForest.trainRegressor(trainingData,
+        categoricalFeaturesInfo, numTrees, featureSubsetStrategy, impurity, maxDepth, maxBins,
+        seed);
+
+    // Evaluate model on test instances and compute test error
+    JavaPairRDD<Double, Double> predictionAndLabel =
+        testData.mapToPair(new PairFunction<LabeledPoint, Double, Double>() {
+          @Override
+          public Tuple2<Double, Double> call(LabeledPoint p) {
+            return new Tuple2<Double, Double>(model.predict(p.features()), p.label());
+          }
+        });
+    Double testMSE =
+        predictionAndLabel.map(new Function<Tuple2<Double, Double>, Double>() {
+          @Override
+          public Double call(Tuple2<Double, Double> pl) {
+            Double diff = pl._1() - pl._2();
+            return diff * diff;
+          }
+        }).reduce(new Function2<Double, Double, Double>() {
+          @Override
+          public Double call(Double a, Double b) {
+            return a + b;
+          }
+        }) / testData.count();
+    System.out.println("Test Mean Squared Error: " + testMSE);
+    System.out.println("Learned regression forest model:\n" + model.toDebugString());
+  }
+
+  public static void main(String[] args) {
+    SparkConf sparkConf = new SparkConf().setAppName("JavaRandomForestExample");
+    JavaSparkContext sc = new JavaSparkContext(sparkConf);
+
+    // Load and parse the data file.
+    String datapath = "data/mllib/sample_libsvm_data.txt";
+    JavaRDD<LabeledPoint> data = MLUtils.loadLibSVMFile(sc.sc(), datapath).toJavaRDD();
+    // Split the data into training and test sets (30% held out for testing)
+    JavaRDD<LabeledPoint>[] splits = data.randomSplit(new double[]{0.7, 0.3});
+    JavaRDD<LabeledPoint> trainingData = splits[0];
+    JavaRDD<LabeledPoint> testData = splits[1];
+
+    System.out.println("\nRunning example of classification using RandomForest\n");
+    testClassification(trainingData, testData);
+
+    System.out.println("\nRunning example of regression using RandomForest\n");
+    testRegression(trainingData, testData);
+    sc.stop();
+  }
+}
diff --git a/examples/src/main/python/mllib/decision_tree_runner.py b/examples/src/main/python/mllib/decision_tree_runner.py
index 61ea4e06ecf3a..fccabd841b139 100755
--- a/examples/src/main/python/mllib/decision_tree_runner.py
+++ b/examples/src/main/python/mllib/decision_tree_runner.py
@@ -106,8 +106,7 @@ def reindexClassLabels(data):
 
 def usage():
     print >> sys.stderr, \
-        "Usage: decision_tree_runner [libsvm format data filepath]\n" + \
-        " Note: This only supports binary classification."
+        "Usage: decision_tree_runner [libsvm format data filepath]"
     exit(1)
 
 
@@ -127,16 +126,20 @@ def usage():
 
     # Re-index class labels if needed.
     (reindexedData, origToNewLabels) = reindexClassLabels(points)
+    numClasses = len(origToNewLabels)
 
     # Train a classifier.
     categoricalFeaturesInfo = {}  # no categorical features
-    model = DecisionTree.trainClassifier(reindexedData, numClasses=2,
+    model = DecisionTree.trainClassifier(reindexedData, numClasses=numClasses,
                                          categoricalFeaturesInfo=categoricalFeaturesInfo)
     # Print learned tree and stats.
     print "Trained DecisionTree for classification:"
-    print "  Model numNodes: %d\n" % model.numNodes()
-    print "  Model depth: %d\n" % model.depth()
-    print "  Training accuracy: %g\n" % getAccuracy(model, reindexedData)
-    print model
+    print "  Model numNodes: %d" % model.numNodes()
+    print "  Model depth: %d" % model.depth()
+    print "  Training accuracy: %g" % getAccuracy(model, reindexedData)
+    if model.numNodes() < 20:
+        print model.toDebugString()
+    else:
+        print model
 
     sc.stop()
diff --git a/examples/src/main/python/mllib/random_forest_example.py b/examples/src/main/python/mllib/random_forest_example.py
new file mode 100755
index 0000000000000..d3c24f7664329
--- /dev/null
+++ b/examples/src/main/python/mllib/random_forest_example.py
@@ -0,0 +1,89 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements.  See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License.  You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+"""
+Random Forest classification and regression using MLlib.
+
+Note: This example illustrates binary classification.
+      For information on multiclass classification, please refer to the decision_tree_runner.py
+      example.
+"""
+
+import sys
+
+from pyspark.context import SparkContext
+from pyspark.mllib.tree import RandomForest
+from pyspark.mllib.util import MLUtils
+
+
+def testClassification(trainingData, testData):
+    # Train a RandomForest model.
+    #  Empty categoricalFeaturesInfo indicates all features are continuous.
+    #  Note: Use larger numTrees in practice.
+    #  Setting featureSubsetStrategy="auto" lets the algorithm choose.
+    model = RandomForest.trainClassifier(trainingData, numClasses=2,
+                                         categoricalFeaturesInfo={},
+                                         numTrees=3, featureSubsetStrategy="auto",
+                                         impurity='gini', maxDepth=4, maxBins=32)
+
+    # Evaluate model on test instances and compute test error
+    predictions = model.predict(testData.map(lambda x: x.features))
+    labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
+    testErr = labelsAndPredictions.filter(lambda (v, p): v != p).count()\
+        / float(testData.count())
+    print('Test Error = ' + str(testErr))
+    print('Learned classification forest model:')
+    print(model.toDebugString())
+
+
+def testRegression(trainingData, testData):
+    # Train a RandomForest model.
+    #  Empty categoricalFeaturesInfo indicates all features are continuous.
+    #  Note: Use larger numTrees in practice.
+    #  Setting featureSubsetStrategy="auto" lets the algorithm choose.
+    model = RandomForest.trainRegressor(trainingData, categoricalFeaturesInfo={},
+                                        numTrees=3, featureSubsetStrategy="auto",
+                                        impurity='variance', maxDepth=4, maxBins=32)
+
+    # Evaluate model on test instances and compute test error
+    predictions = model.predict(testData.map(lambda x: x.features))
+    labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
+    testMSE = labelsAndPredictions.map(lambda (v, p): (v - p) * (v - p)).sum()\
+        / float(testData.count())
+    print('Test Mean Squared Error = ' + str(testMSE))
+    print('Learned regression forest model:')
+    print(model.toDebugString())
+
+
+if __name__ == "__main__":
+    if len(sys.argv) > 1:
+        print >> sys.stderr, "Usage: random_forest_example"
+        exit(1)
+    sc = SparkContext(appName="PythonRandomForestExample")
+
+    # Load and parse the data file into an RDD of LabeledPoint.
+    data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt')
+    # Split the data into training and test sets (30% held out for testing)
+    (trainingData, testData) = data.randomSplit([0.7, 0.3])
+
+    print('\nRunning example of classification using RandomForest\n')
+    testClassification(trainingData, testData)
+
+    print('\nRunning example of regression using RandomForest\n')
+    testRegression(trainingData, testData)
+
+    sc.stop()
diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala
new file mode 100644
index 0000000000000..ce6bc066bd70d
--- /dev/null
+++ b/examples/src/main/scala/org/apache/spark/examples/ml/CrossValidatorExample.scala
@@ -0,0 +1,110 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.examples.ml
+
+import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.SparkContext._
+import org.apache.spark.ml.Pipeline
+import org.apache.spark.ml.classification.LogisticRegression
+import org.apache.spark.ml.evaluation.BinaryClassificationEvaluator
+import org.apache.spark.ml.feature.{HashingTF, Tokenizer}
+import org.apache.spark.ml.tuning.{ParamGridBuilder, CrossValidator}
+import org.apache.spark.sql.{Row, SQLContext}
+
+/**
+ * A simple example demonstrating model selection using CrossValidator.
+ * This example also demonstrates how Pipelines are Estimators.
+ *
+ * This example uses the [[LabeledDocument]] and [[Document]] case classes from
+ * [[SimpleTextClassificationPipeline]].
+ *
+ * Run with
+ * {{{
+ * bin/run-example ml.CrossValidatorExample
+ * }}}
+ */
+object CrossValidatorExample {
+
+  def main(args: Array[String]) {
+    val conf = new SparkConf().setAppName("CrossValidatorExample")
+    val sc = new SparkContext(conf)
+    val sqlContext = new SQLContext(sc)
+    import sqlContext._
+
+    // Prepare training documents, which are labeled.
+    val training = sparkContext.parallelize(Seq(
+      LabeledDocument(0L, "a b c d e spark", 1.0),
+      LabeledDocument(1L, "b d", 0.0),
+      LabeledDocument(2L, "spark f g h", 1.0),
+      LabeledDocument(3L, "hadoop mapreduce", 0.0),
+      LabeledDocument(4L, "b spark who", 1.0),
+      LabeledDocument(5L, "g d a y", 0.0),
+      LabeledDocument(6L, "spark fly", 1.0),
+      LabeledDocument(7L, "was mapreduce", 0.0),
+      LabeledDocument(8L, "e spark program", 1.0),
+      LabeledDocument(9L, "a e c l", 0.0),
+      LabeledDocument(10L, "spark compile", 1.0),
+      LabeledDocument(11L, "hadoop software", 0.0)))
+
+    // Configure an ML pipeline, which consists of three stages: tokenizer, hashingTF, and lr.
+    val tokenizer = new Tokenizer()
+      .setInputCol("text")
+      .setOutputCol("words")
+    val hashingTF = new HashingTF()
+      .setInputCol(tokenizer.getOutputCol)
+      .setOutputCol("features")
+    val lr = new LogisticRegression()
+      .setMaxIter(10)
+    val pipeline = new Pipeline()
+      .setStages(Array(tokenizer, hashingTF, lr))
+
+    // We now treat the Pipeline as an Estimator, wrapping it in a CrossValidator instance.
+    // This will allow us to jointly choose parameters for all Pipeline stages.
+    // A CrossValidator requires an Estimator, a set of Estimator ParamMaps, and an Evaluator.
+    val crossval = new CrossValidator()
+      .setEstimator(pipeline)
+      .setEvaluator(new BinaryClassificationEvaluator)
+    // We use a ParamGridBuilder to construct a grid of parameters to search over.
+    // With 3 values for hashingTF.numFeatures and 2 values for lr.regParam,
+    // this grid will have 3 x 2 = 6 parameter settings for CrossValidator to choose from.
+    val paramGrid = new ParamGridBuilder()
+      .addGrid(hashingTF.numFeatures, Array(10, 100, 1000))
+      .addGrid(lr.regParam, Array(0.1, 0.01))
+      .build()
+    crossval.setEstimatorParamMaps(paramGrid)
+    crossval.setNumFolds(2) // Use 3+ in practice
+
+    // Run cross-validation, and choose the best set of parameters.
+    val cvModel = crossval.fit(training)
+
+    // Prepare test documents, which are unlabeled.
+    val test = sparkContext.parallelize(Seq(
+      Document(4L, "spark i j k"),
+      Document(5L, "l m n"),
+      Document(6L, "mapreduce spark"),
+      Document(7L, "apache hadoop")))
+
+    // Make predictions on test documents. cvModel uses the best model found (lrModel).
+    cvModel.transform(test)
+      .select('id, 'text, 'score, 'prediction)
+      .collect()
+      .foreach { case Row(id: Long, text: String, score: Double, prediction: Double) =>
+      println("(" + id + ", " + text + ") --> score=" + score + ", prediction=" + prediction)
+    }
+  }
+}
diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala
new file mode 100644
index 0000000000000..44d5b084c269a
--- /dev/null
+++ b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleParamsExample.scala
@@ -0,0 +1,101 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.examples.ml
+
+import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.SparkContext._
+import org.apache.spark.ml.classification.LogisticRegression
+import org.apache.spark.ml.param.ParamMap
+import org.apache.spark.mllib.linalg.{Vector, Vectors}
+import org.apache.spark.mllib.regression.LabeledPoint
+import org.apache.spark.sql.{Row, SQLContext}
+
+/**
+ * A simple example demonstrating ways to specify parameters for Estimators and Transformers.
+ * Run with
+ * {{{
+ * bin/run-example ml.SimpleParamsExample
+ * }}}
+ */
+object SimpleParamsExample {
+
+  def main(args: Array[String]) {
+    val conf = new SparkConf().setAppName("SimpleParamsExample")
+    val sc = new SparkContext(conf)
+    val sqlContext = new SQLContext(sc)
+    import sqlContext._
+
+    // Prepare training data.
+    // We use LabeledPoint, which is a case class.  Spark SQL can convert RDDs of Java Beans
+    // into SchemaRDDs, where it uses the bean metadata to infer the schema.
+    val training = sparkContext.parallelize(Seq(
+      LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)),
+      LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)),
+      LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)),
+      LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5))))
+
+    // Create a LogisticRegression instance.  This instance is an Estimator.
+    val lr = new LogisticRegression()
+    // Print out the parameters, documentation, and any default values.
+    println("LogisticRegression parameters:\n" + lr.explainParams() + "\n")
+
+    // We may set parameters using setter methods.
+    lr.setMaxIter(10)
+      .setRegParam(0.01)
+
+    // Learn a LogisticRegression model.  This uses the parameters stored in lr.
+    val model1 = lr.fit(training)
+    // Since model1 is a Model (i.e., a Transformer produced by an Estimator),
+    // we can view the parameters it used during fit().
+    // This prints the parameter (name: value) pairs, where names are unique IDs for this
+    // LogisticRegression instance.
+    println("Model 1 was fit using parameters: " + model1.fittingParamMap)
+
+    // We may alternatively specify parameters using a ParamMap,
+    // which supports several methods for specifying parameters.
+    val paramMap = ParamMap(lr.maxIter -> 20)
+    paramMap.put(lr.maxIter, 30) // Specify 1 Param.  This overwrites the original maxIter.
+    paramMap.put(lr.regParam -> 0.1, lr.threshold -> 0.55) // Specify multiple Params.
+
+    // One can also combine ParamMaps.
+    val paramMap2 = ParamMap(lr.scoreCol -> "probability") // Change output column name
+    val paramMapCombined = paramMap ++ paramMap2
+
+    // Now learn a new model using the paramMapCombined parameters.
+    // paramMapCombined overrides all parameters set earlier via lr.set* methods.
+    val model2 = lr.fit(training, paramMapCombined)
+    println("Model 2 was fit using parameters: " + model2.fittingParamMap)
+
+    // Prepare test documents.
+    val test = sparkContext.parallelize(Seq(
+      LabeledPoint(1.0, Vectors.dense(-1.0, 1.5, 1.3)),
+      LabeledPoint(0.0, Vectors.dense(3.0, 2.0, -0.1)),
+      LabeledPoint(1.0, Vectors.dense(0.0, 2.2, -1.5))))
+
+    // Make predictions on test documents using the Transformer.transform() method.
+    // LogisticRegression.transform will only use the 'features' column.
+    // Note that model2.transform() outputs a 'probability' column instead of the usual 'score'
+    // column since we renamed the lr.scoreCol parameter previously.
+    model2.transform(test)
+      .select('features, 'label, 'probability, 'prediction)
+      .collect()
+      .foreach { case Row(features: Vector, label: Double, prob: Double, prediction: Double) =>
+        println("(" + features + ", " + label + ") -> prob=" + prob + ", prediction=" + prediction)
+      }
+  }
+}
diff --git a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala
index ee7897d9062d9..92895a05e479a 100644
--- a/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/ml/SimpleTextClassificationPipeline.scala
@@ -20,10 +20,11 @@ package org.apache.spark.examples.ml
 import scala.beans.BeanInfo
 
 import org.apache.spark.{SparkConf, SparkContext}
+import org.apache.spark.SparkContext._
 import org.apache.spark.ml.Pipeline
 import org.apache.spark.ml.classification.LogisticRegression
 import org.apache.spark.ml.feature.{HashingTF, Tokenizer}
-import org.apache.spark.sql.SQLContext
+import org.apache.spark.sql.{Row, SQLContext}
 
 @BeanInfo
 case class LabeledDocument(id: Long, text: String, label: Double)
@@ -81,6 +82,8 @@ object SimpleTextClassificationPipeline {
     model.transform(test)
       .select('id, 'text, 'score, 'prediction)
       .collect()
-      .foreach(println)
+      .foreach { case Row(id: Long, text: String, score: Double, prediction: Double) =>
+        println("(" + id + ", " + text + ") --> score=" + score + ", prediction=" + prediction)
+      }
   }
 }
diff --git a/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala b/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala
index 98f9d1689c8e7..205d80dd02682 100644
--- a/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/mllib/DecisionTreeRunner.scala
@@ -17,6 +17,8 @@
 
 package org.apache.spark.examples.mllib
 
+import scala.language.reflectiveCalls
+
 import scopt.OptionParser
 
 import org.apache.spark.{SparkConf, SparkContext}
@@ -276,7 +278,7 @@ object DecisionTreeRunner {
           impurity = impurityCalculator,
           maxDepth = params.maxDepth,
           maxBins = params.maxBins,
-          numClassesForClassification = numClasses,
+          numClasses = numClasses,
           minInstancesPerNode = params.minInstancesPerNode,
           minInfoGain = params.minInfoGain,
           useNodeIdCache = params.useNodeIdCache,
diff --git a/examples/src/main/scala/org/apache/spark/examples/mllib/GradientBoostedTreesRunner.scala b/examples/src/main/scala/org/apache/spark/examples/mllib/GradientBoostedTreesRunner.scala
index 1def8b45a230c..431ead8c0c165 100644
--- a/examples/src/main/scala/org/apache/spark/examples/mllib/GradientBoostedTreesRunner.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/mllib/GradientBoostedTreesRunner.scala
@@ -103,7 +103,7 @@ object GradientBoostedTreesRunner {
       params.dataFormat, params.testInput, Algo.withName(params.algo), params.fracTest)
 
     val boostingStrategy = BoostingStrategy.defaultParams(params.algo)
-    boostingStrategy.treeStrategy.numClassesForClassification = numClasses
+    boostingStrategy.treeStrategy.numClasses = numClasses
     boostingStrategy.numIterations = params.numIterations
     boostingStrategy.treeStrategy.maxDepth = params.maxDepth
 
diff --git a/examples/src/main/scala/org/apache/spark/examples/mllib/StreamingKMeans.scala b/examples/src/main/scala/org/apache/spark/examples/mllib/StreamingKMeansExample.scala
similarity index 90%
rename from examples/src/main/scala/org/apache/spark/examples/mllib/StreamingKMeans.scala
rename to examples/src/main/scala/org/apache/spark/examples/mllib/StreamingKMeansExample.scala
index 33e5760aed997..8bb12d2ee9ed2 100644
--- a/examples/src/main/scala/org/apache/spark/examples/mllib/StreamingKMeans.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/mllib/StreamingKMeansExample.scala
@@ -17,10 +17,10 @@
 
 package org.apache.spark.examples.mllib
 
+import org.apache.spark.SparkConf
+import org.apache.spark.mllib.clustering.StreamingKMeans
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.mllib.clustering.StreamingKMeans
-import org.apache.spark.SparkConf
 import org.apache.spark.streaming.{Seconds, StreamingContext}
 
 /**
@@ -36,28 +36,28 @@ import org.apache.spark.streaming.{Seconds, StreamingContext}
  * `(y,[x1,x2,x3,...,xn])`
  * Where y is some identifier. n must be the same for train and test.
  *
- * Usage: StreamingKmeans <trainingDir> <testDir> <batchDuration> <numClusters> <numDimensions>
+ * Usage:
+ *   StreamingKMeansExample <trainingDir> <testDir> <batchDuration> <numClusters> <numDimensions>
  *
  * To run on your local machine using the two directories `trainingDir` and `testDir`,
  * with updates every 5 seconds, 2 dimensions per data point, and 3 clusters, call:
- *    $ bin/run-example \
- *        org.apache.spark.examples.mllib.StreamingKMeans trainingDir testDir 5 3 2
+ *    $ bin/run-example mllib.StreamingKMeansExample trainingDir testDir 5 3 2
  *
  * As you add text files to `trainingDir` the clusters will continuously update.
  * Anytime you add text files to `testDir`, you'll see predicted labels using the current model.
  *
  */
-object StreamingKMeans {
+object StreamingKMeansExample {
 
   def main(args: Array[String]) {
     if (args.length != 5) {
       System.err.println(
-        "Usage: StreamingKMeans " +
+        "Usage: StreamingKMeansExample " +
           "<trainingDir> <testDir> <batchDuration> <numClusters> <numDimensions>")
       System.exit(1)
     }
 
-    val conf = new SparkConf().setMaster("local").setAppName("StreamingLinearRegression")
+    val conf = new SparkConf().setMaster("local").setAppName("StreamingKMeansExample")
     val ssc = new StreamingContext(conf, Seconds(args(2).toLong))
 
     val trainingData = ssc.textFileStream(args(0)).map(Vectors.parse)
diff --git a/examples/src/main/scala/org/apache/spark/examples/sql/hive/HiveFromSpark.scala b/examples/src/main/scala/org/apache/spark/examples/sql/hive/HiveFromSpark.scala
index 227acc117502d..138923c4d7f2f 100644
--- a/examples/src/main/scala/org/apache/spark/examples/sql/hive/HiveFromSpark.scala
+++ b/examples/src/main/scala/org/apache/spark/examples/sql/hive/HiveFromSpark.scala
@@ -29,9 +29,10 @@ object HiveFromSpark {
     val sc = new SparkContext(sparkConf)
     val path = s"${System.getenv("SPARK_HOME")}/examples/src/main/resources/kv1.txt"
 
-    // A local hive context creates an instance of the Hive Metastore in process, storing 
-    // the warehouse data in the current directory.  This location can be overridden by
-    // specifying a second parameter to the constructor.
+    // A hive context adds support for finding tables in the MetaStore and writing queries
+    // using HiveQL. Users who do not have an existing Hive deployment can still create a
+    // HiveContext. When not configured by the hive-site.xml, the context automatically
+    // creates metastore_db and warehouse in the current directory.
     val hiveContext = new HiveContext(sc)
     import hiveContext._
 
diff --git a/external/mqtt/pom.xml b/external/mqtt/pom.xml
index 703806735b3ff..9025915f4447e 100644
--- a/external/mqtt/pom.xml
+++ b/external/mqtt/pom.xml
@@ -43,8 +43,8 @@
     </dependency>
     <dependency>
       <groupId>org.eclipse.paho</groupId>
-      <artifactId>mqtt-client</artifactId>
-       <version>0.4.0</version>
+      <artifactId>org.eclipse.paho.client.mqttv3</artifactId>
+      <version>1.0.1</version>
     </dependency>
     <dependency>
       <groupId>org.scalatest</groupId>
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/Graph.scala b/graphx/src/main/scala/org/apache/spark/graphx/Graph.scala
index 637791543514c..23538b71562de 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/Graph.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/Graph.scala
@@ -96,6 +96,14 @@ abstract class Graph[VD: ClassTag, ED: ClassTag] protected () extends Serializab
    */
   def cache(): Graph[VD, ED]
 
+  /**
+   * Mark this Graph for checkpointing. It will be saved to a file inside the checkpoint
+   * directory set with SparkContext.setCheckpointDir() and all references to its parent
+   * RDDs will be removed. It is strongly recommended that this Graph is persisted in
+   * memory, otherwise saving it on a file will require recomputation.
+   */
+  def checkpoint(): Unit
+
   /**
    * Uncaches only the vertices of this graph, leaving the edges alone. This is useful in iterative
    * algorithms that modify the vertex attributes but reuse the edges. This method can be used to
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/VertexRDD.scala b/graphx/src/main/scala/org/apache/spark/graphx/VertexRDD.scala
index 1db3df03c8052..09ae3f9f6c09b 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/VertexRDD.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/VertexRDD.scala
@@ -68,7 +68,7 @@ abstract class VertexRDD[VD](
    * Provides the `RDD[(VertexId, VD)]` equivalent output.
    */
   override def compute(part: Partition, context: TaskContext): Iterator[(VertexId, VD)] = {
-    firstParent[ShippableVertexPartition[VD]].iterator(part, context).next.iterator
+    firstParent[ShippableVertexPartition[VD]].iterator(part, context).next().iterator
   }
 
   /**
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/impl/EdgeRDDImpl.scala b/graphx/src/main/scala/org/apache/spark/graphx/impl/EdgeRDDImpl.scala
index a8169613b4fd2..897c7ee12a436 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/impl/EdgeRDDImpl.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/impl/EdgeRDDImpl.scala
@@ -26,7 +26,7 @@ import org.apache.spark.storage.StorageLevel
 import org.apache.spark.graphx._
 
 class EdgeRDDImpl[ED: ClassTag, VD: ClassTag] private[graphx] (
-    override val partitionsRDD: RDD[(PartitionID, EdgePartition[ED, VD])],
+    @transient override val partitionsRDD: RDD[(PartitionID, EdgePartition[ED, VD])],
     val targetStorageLevel: StorageLevel = StorageLevel.MEMORY_ONLY)
   extends EdgeRDD[ED](partitionsRDD.context, List(new OneToOneDependency(partitionsRDD))) {
 
@@ -70,6 +70,10 @@ class EdgeRDDImpl[ED: ClassTag, VD: ClassTag] private[graphx] (
     this
   }
 
+  override def checkpoint() = {
+    partitionsRDD.checkpoint()
+  }
+    
   /** The number of edges in the RDD. */
   override def count(): Long = {
     partitionsRDD.map(_._2.size.toLong).reduce(_ + _)
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/impl/GraphImpl.scala b/graphx/src/main/scala/org/apache/spark/graphx/impl/GraphImpl.scala
index 0eae2a673874a..a617d84aea9d4 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/impl/GraphImpl.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/impl/GraphImpl.scala
@@ -65,6 +65,11 @@ class GraphImpl[VD: ClassTag, ED: ClassTag] protected (
     this
   }
 
+  override def checkpoint(): Unit = {
+    vertices.checkpoint()
+    replicatedVertexView.edges.checkpoint()
+  }
+
   override def unpersistVertices(blocking: Boolean = true): Graph[VD, ED] = {
     vertices.unpersist(blocking)
     // TODO: unpersist the replicated vertices in `replicatedVertexView` but leave the edges alone
diff --git a/graphx/src/main/scala/org/apache/spark/graphx/impl/VertexRDDImpl.scala b/graphx/src/main/scala/org/apache/spark/graphx/impl/VertexRDDImpl.scala
index d92a55a189298..9732c5b00c6d9 100644
--- a/graphx/src/main/scala/org/apache/spark/graphx/impl/VertexRDDImpl.scala
+++ b/graphx/src/main/scala/org/apache/spark/graphx/impl/VertexRDDImpl.scala
@@ -27,7 +27,7 @@ import org.apache.spark.storage.StorageLevel
 import org.apache.spark.graphx._
 
 class VertexRDDImpl[VD] private[graphx] (
-    val partitionsRDD: RDD[ShippableVertexPartition[VD]],
+    @transient val partitionsRDD: RDD[ShippableVertexPartition[VD]],
     val targetStorageLevel: StorageLevel = StorageLevel.MEMORY_ONLY)
   (implicit override protected val vdTag: ClassTag[VD])
   extends VertexRDD[VD](partitionsRDD.context, List(new OneToOneDependency(partitionsRDD))) {
@@ -71,6 +71,10 @@ class VertexRDDImpl[VD] private[graphx] (
     this
   }
 
+  override def checkpoint() = {
+    partitionsRDD.checkpoint()
+  }
+    
   /** The number of vertices in the RDD. */
   override def count(): Long = {
     partitionsRDD.map(_.size).reduce(_ + _)
diff --git a/graphx/src/test/scala/org/apache/spark/graphx/GraphSuite.scala b/graphx/src/test/scala/org/apache/spark/graphx/GraphSuite.scala
index a05d1ddb21295..9da0064104fb6 100644
--- a/graphx/src/test/scala/org/apache/spark/graphx/GraphSuite.scala
+++ b/graphx/src/test/scala/org/apache/spark/graphx/GraphSuite.scala
@@ -19,6 +19,8 @@ package org.apache.spark.graphx
 
 import org.scalatest.FunSuite
 
+import com.google.common.io.Files
+
 import org.apache.spark.SparkContext
 import org.apache.spark.graphx.Graph._
 import org.apache.spark.graphx.PartitionStrategy._
@@ -365,4 +367,23 @@ class GraphSuite extends FunSuite with LocalSparkContext {
     }
   }
 
+  test("checkpoint") {
+    val checkpointDir = Files.createTempDir()
+    checkpointDir.deleteOnExit()
+    withSpark { sc =>
+      sc.setCheckpointDir(checkpointDir.getAbsolutePath)
+      val ring = (0L to 100L).zip((1L to 99L) :+ 0L).map { case (a, b) => Edge(a, b, 1)}
+      val rdd = sc.parallelize(ring)
+      val graph = Graph.fromEdges(rdd, 1.0F)
+      graph.checkpoint()
+      graph.edges.map(_.attr).count()
+      graph.vertices.map(_._2).count()
+
+      val edgesDependencies = graph.edges.partitionsRDD.dependencies
+      val verticesDependencies = graph.vertices.partitionsRDD.dependencies
+      assert(edgesDependencies.forall(_.rdd.isInstanceOf[CheckpointRDD[_]]))
+      assert(verticesDependencies.forall(_.rdd.isInstanceOf[CheckpointRDD[_]]))
+    }
+  }
+
 }
diff --git a/make-distribution.sh b/make-distribution.sh
index 7c0fb8992a155..45c99e42e5a5b 100755
--- a/make-distribution.sh
+++ b/make-distribution.sh
@@ -201,6 +201,9 @@ if [ -e "$FWDIR"/CHANGES.txt ]; then
   cp "$FWDIR/CHANGES.txt" "$DISTDIR"
 fi
 
+# Copy data files
+cp -r "$FWDIR/data" "$DISTDIR"
+
 # Copy other things
 mkdir "$DISTDIR"/conf
 cp "$FWDIR"/conf/*.template "$DISTDIR"/conf
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala b/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
index e545df1e37b9c..081a574beea5d 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala
@@ -162,11 +162,15 @@ class PipelineModel private[ml] (
   }
 
   override def transform(dataset: SchemaRDD, paramMap: ParamMap): SchemaRDD = {
-    transformSchema(dataset.schema, paramMap, logging = true)
-    stages.foldLeft(dataset)((cur, transformer) => transformer.transform(cur, paramMap))
+    // Precedence of ParamMaps: paramMap > this.paramMap > fittingParamMap
+    val map = (fittingParamMap ++ this.paramMap) ++ paramMap
+    transformSchema(dataset.schema, map, logging = true)
+    stages.foldLeft(dataset)((cur, transformer) => transformer.transform(cur, map))
   }
 
   private[ml] override def transformSchema(schema: StructType, paramMap: ParamMap): StructType = {
-    stages.foldLeft(schema)((cur, transformer) => transformer.transformSchema(cur, paramMap))
+    // Precedence of ParamMaps: paramMap > this.paramMap > fittingParamMap
+    val map = (fittingParamMap ++ this.paramMap) ++ paramMap
+    stages.foldLeft(schema)((cur, transformer) => transformer.transformSchema(cur, map))
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala b/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala
index 490e6609ad311..23fbd228d01cb 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/Transformer.scala
@@ -18,16 +18,14 @@
 package org.apache.spark.ml
 
 import scala.annotation.varargs
-import scala.reflect.runtime.universe.TypeTag
 
 import org.apache.spark.Logging
 import org.apache.spark.annotation.AlphaComponent
 import org.apache.spark.ml.param._
 import org.apache.spark.sql.SchemaRDD
 import org.apache.spark.sql.api.java.JavaSchemaRDD
-import org.apache.spark.sql.catalyst.ScalaReflection
 import org.apache.spark.sql.catalyst.analysis.Star
-import org.apache.spark.sql.catalyst.dsl._
+import org.apache.spark.sql.catalyst.expressions.ScalaUdf
 import org.apache.spark.sql.catalyst.types._
 
 /**
@@ -86,7 +84,7 @@ abstract class Transformer extends PipelineStage with Params {
  * Abstract class for transformers that take one input column, apply transformation, and output the
  * result as a new column.
  */
-private[ml] abstract class UnaryTransformer[IN, OUT: TypeTag, T <: UnaryTransformer[IN, OUT, T]]
+private[ml] abstract class UnaryTransformer[IN, OUT, T <: UnaryTransformer[IN, OUT, T]]
   extends Transformer with HasInputCol with HasOutputCol with Logging {
 
   def setInputCol(value: String): T = set(inputCol, value).asInstanceOf[T]
@@ -99,6 +97,11 @@ private[ml] abstract class UnaryTransformer[IN, OUT: TypeTag, T <: UnaryTransfor
    */
   protected def createTransformFunc(paramMap: ParamMap): IN => OUT
 
+  /**
+   * Returns the data type of the output column.
+   */
+  protected def outputDataType: DataType
+
   /**
    * Validates the input type. Throw an exception if it is invalid.
    */
@@ -111,9 +114,8 @@ private[ml] abstract class UnaryTransformer[IN, OUT: TypeTag, T <: UnaryTransfor
     if (schema.fieldNames.contains(map(outputCol))) {
       throw new IllegalArgumentException(s"Output column ${map(outputCol)} already exists.")
     }
-    val output = ScalaReflection.schemaFor[OUT]
     val outputFields = schema.fields :+
-      StructField(map(outputCol), output.dataType, output.nullable)
+      StructField(map(outputCol), outputDataType, !outputDataType.isPrimitive)
     StructType(outputFields)
   }
 
@@ -121,7 +123,7 @@ private[ml] abstract class UnaryTransformer[IN, OUT: TypeTag, T <: UnaryTransfor
     transformSchema(dataset.schema, paramMap, logging = true)
     import dataset.sqlContext._
     val map = this.paramMap ++ paramMap
-    val udf = this.createTransformFunc(map)
-    dataset.select(Star(None), udf.call(map(inputCol).attr) as map(outputCol))
+    val udf = ScalaUdf(this.createTransformFunc(map), outputDataType, Seq(map(inputCol).attr))
+    dataset.select(Star(None), udf as map(outputCol))
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
index b98b1755a3584..e0bfb1e484a2e 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
@@ -21,7 +21,8 @@ import org.apache.spark.annotation.AlphaComponent
 import org.apache.spark.ml.UnaryTransformer
 import org.apache.spark.ml.param.{IntParam, ParamMap}
 import org.apache.spark.mllib.feature
-import org.apache.spark.mllib.linalg.Vector
+import org.apache.spark.mllib.linalg.{VectorUDT, Vector}
+import org.apache.spark.sql.catalyst.types.DataType
 
 /**
  * :: AlphaComponent ::
@@ -39,4 +40,6 @@ class HashingTF extends UnaryTransformer[Iterable[_], Vector, HashingTF] {
     val hashingTF = new feature.HashingTF(paramMap(numFeatures))
     hashingTF.transform
   }
+
+  override protected def outputDataType: DataType = new VectorUDT()
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala
index 0a6599b64c011..9352f40f372d3 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/feature/Tokenizer.scala
@@ -20,7 +20,7 @@ package org.apache.spark.ml.feature
 import org.apache.spark.annotation.AlphaComponent
 import org.apache.spark.ml.UnaryTransformer
 import org.apache.spark.ml.param.ParamMap
-import org.apache.spark.sql.{DataType, StringType}
+import org.apache.spark.sql.{DataType, StringType, ArrayType}
 
 /**
  * :: AlphaComponent ::
@@ -36,4 +36,6 @@ class Tokenizer extends UnaryTransformer[String, Seq[String], Tokenizer] {
   protected override def validateInputType(inputType: DataType): Unit = {
     require(inputType == StringType, s"Input type must be string type but got $inputType.")
   }
+
+  override protected def outputDataType: DataType = new ArrayType(StringType, false)
 }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/param/params.scala b/mllib/src/main/scala/org/apache/spark/ml/param/params.scala
index 8fd46aef4b99d..04f9cfb1bfc2f 100644
--- a/mllib/src/main/scala/org/apache/spark/ml/param/params.scala
+++ b/mllib/src/main/scala/org/apache/spark/ml/param/params.scala
@@ -17,13 +17,12 @@
 
 package org.apache.spark.ml.param
 
-import java.lang.reflect.Modifier
-
-import org.apache.spark.annotation.AlphaComponent
-
 import scala.annotation.varargs
 import scala.collection.mutable
 
+import java.lang.reflect.Modifier
+
+import org.apache.spark.annotation.AlphaComponent
 import org.apache.spark.ml.Identifiable
 
 /**
@@ -222,6 +221,7 @@ class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any]) exten
   /**
    * Puts a list of param pairs (overwrites if the input params exists).
    */
+  @varargs
   def put(paramPairs: ParamPair[_]*): this.type = {
     paramPairs.foreach { p =>
       put(p.param.asInstanceOf[Param[Any]], p.value)
@@ -282,6 +282,7 @@ class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any]) exten
    * where the latter overwrites this if there exists conflicts.
    */
   def ++(other: ParamMap): ParamMap = {
+    // TODO: Provide a better method name for Java users.
     new ParamMap(this.map ++ other.map)
   }
 
@@ -290,6 +291,7 @@ class ParamMap private[ml] (private val map: mutable.Map[Param[Any], Any]) exten
    * Adds all parameters from the input param map into this param map.
    */
   def ++=(other: ParamMap): this.type = {
+    // TODO: Provide a better method name for Java users.
     this.map ++= other.map
     this
   }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala b/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala
index 9f20cd5d00dcd..c4e5fd8e461fc 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala
@@ -477,7 +477,7 @@ class PythonMLLibAPI extends Serializable {
       algo = algo,
       impurity = impurity,
       maxDepth = maxDepth,
-      numClassesForClassification = numClasses,
+      numClasses = numClasses,
       maxBins = maxBins,
       categoricalFeaturesInfo = categoricalFeaturesInfo.asScala.toMap,
       minInstancesPerNode = minInstancesPerNode,
@@ -513,7 +513,7 @@ class PythonMLLibAPI extends Serializable {
       algo = algo,
       impurity = impurity,
       maxDepth = maxDepth,
-      numClassesForClassification = numClasses,
+      numClasses = numClasses,
       maxBins = maxBins,
       categoricalFeaturesInfo = categoricalFeaturesInfo.asScala.toMap)
     val cached = data.rdd.persist(StorageLevel.MEMORY_AND_DISK)
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
index 34ea0de706f08..54c301d3e9e14 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeans.scala
@@ -19,12 +19,11 @@ package org.apache.spark.mllib.clustering
 
 import scala.collection.mutable.ArrayBuffer
 
-import breeze.linalg.{DenseVector => BDV, Vector => BV, norm => breezeNorm}
-
 import org.apache.spark.annotation.Experimental
 import org.apache.spark.Logging
 import org.apache.spark.SparkContext._
 import org.apache.spark.mllib.linalg.{Vector, Vectors}
+import org.apache.spark.mllib.linalg.BLAS.{axpy, scal}
 import org.apache.spark.mllib.util.MLUtils
 import org.apache.spark.rdd.RDD
 import org.apache.spark.storage.StorageLevel
@@ -125,12 +124,12 @@ class KMeans private (
     }
 
     // Compute squared norms and cache them.
-    val norms = data.map(v => breezeNorm(v.toBreeze, 2.0))
+    val norms = data.map(Vectors.norm(_, 2.0))
     norms.persist()
-    val breezeData = data.map(_.toBreeze).zip(norms).map { case (v, norm) =>
-      new BreezeVectorWithNorm(v, norm)
+    val zippedData = data.zip(norms).map { case (v, norm) =>
+      new VectorWithNorm(v, norm)
     }
-    val model = runBreeze(breezeData)
+    val model = runAlgorithm(zippedData)
     norms.unpersist()
 
     // Warn at the end of the run as well, for increased visibility.
@@ -142,9 +141,9 @@ class KMeans private (
   }
 
   /**
-   * Implementation of K-Means using breeze.
+   * Implementation of K-Means algorithm.
    */
-  private def runBreeze(data: RDD[BreezeVectorWithNorm]): KMeansModel = {
+  private def runAlgorithm(data: RDD[VectorWithNorm]): KMeansModel = {
 
     val sc = data.sparkContext
 
@@ -170,9 +169,10 @@ class KMeans private (
 
     // Execute iterations of Lloyd's algorithm until all runs have converged
     while (iteration < maxIterations && !activeRuns.isEmpty) {
-      type WeightedPoint = (BV[Double], Long)
-      def mergeContribs(p1: WeightedPoint, p2: WeightedPoint): WeightedPoint = {
-        (p1._1 += p2._1, p1._2 + p2._2)
+      type WeightedPoint = (Vector, Long)
+      def mergeContribs(x: WeightedPoint, y: WeightedPoint): WeightedPoint = {
+        axpy(1.0, x._1, y._1)
+        (y._1, x._2 + y._2)
       }
 
       val activeCenters = activeRuns.map(r => centers(r)).toArray
@@ -185,16 +185,17 @@ class KMeans private (
         val thisActiveCenters = bcActiveCenters.value
         val runs = thisActiveCenters.length
         val k = thisActiveCenters(0).length
-        val dims = thisActiveCenters(0)(0).vector.length
+        val dims = thisActiveCenters(0)(0).vector.size
 
-        val sums = Array.fill(runs, k)(BDV.zeros[Double](dims).asInstanceOf[BV[Double]])
+        val sums = Array.fill(runs, k)(Vectors.zeros(dims))
         val counts = Array.fill(runs, k)(0L)
 
         points.foreach { point =>
           (0 until runs).foreach { i =>
             val (bestCenter, cost) = KMeans.findClosest(thisActiveCenters(i), point)
             costAccums(i) += cost
-            sums(i)(bestCenter) += point.vector
+            val sum = sums(i)(bestCenter)
+            axpy(1.0, point.vector, sum)
             counts(i)(bestCenter) += 1
           }
         }
@@ -212,8 +213,8 @@ class KMeans private (
         while (j < k) {
           val (sum, count) = totalContribs((i, j))
           if (count != 0) {
-            sum /= count.toDouble
-            val newCenter = new BreezeVectorWithNorm(sum)
+            scal(1.0 / count, sum)
+            val newCenter = new VectorWithNorm(sum)
             if (KMeans.fastSquaredDistance(newCenter, centers(run)(j)) > epsilon * epsilon) {
               changed = true
             }
@@ -245,18 +246,18 @@ class KMeans private (
 
     logInfo(s"The cost for the best run is $minCost.")
 
-    new KMeansModel(centers(bestRun).map(c => Vectors.fromBreeze(c.vector)))
+    new KMeansModel(centers(bestRun).map(_.vector))
   }
 
   /**
    * Initialize `runs` sets of cluster centers at random.
    */
-  private def initRandom(data: RDD[BreezeVectorWithNorm])
-  : Array[Array[BreezeVectorWithNorm]] = {
+  private def initRandom(data: RDD[VectorWithNorm])
+  : Array[Array[VectorWithNorm]] = {
     // Sample all the cluster centers in one pass to avoid repeated scans
     val sample = data.takeSample(true, runs * k, new XORShiftRandom().nextInt()).toSeq
     Array.tabulate(runs)(r => sample.slice(r * k, (r + 1) * k).map { v =>
-      new BreezeVectorWithNorm(v.vector.toDenseVector, v.norm)
+      new VectorWithNorm(Vectors.dense(v.vector.toArray), v.norm)
     }.toArray)
   }
 
@@ -269,8 +270,8 @@ class KMeans private (
    *
    * The original paper can be found at http://theory.stanford.edu/~sergei/papers/vldb12-kmpar.pdf.
    */
-  private def initKMeansParallel(data: RDD[BreezeVectorWithNorm])
-  : Array[Array[BreezeVectorWithNorm]] = {
+  private def initKMeansParallel(data: RDD[VectorWithNorm])
+  : Array[Array[VectorWithNorm]] = {
     // Initialize each run's center to a random point
     val seed = new XORShiftRandom().nextInt()
     val sample = data.takeSample(true, runs, seed).toSeq
@@ -376,8 +377,8 @@ object KMeans {
    * Returns the index of the closest center to the given point, as well as the squared distance.
    */
   private[mllib] def findClosest(
-      centers: TraversableOnce[BreezeVectorWithNorm],
-      point: BreezeVectorWithNorm): (Int, Double) = {
+      centers: TraversableOnce[VectorWithNorm],
+      point: VectorWithNorm): (Int, Double) = {
     var bestDistance = Double.PositiveInfinity
     var bestIndex = 0
     var i = 0
@@ -402,8 +403,8 @@ object KMeans {
    * Returns the K-means cost of a given point against the given cluster centers.
    */
   private[mllib] def pointCost(
-      centers: TraversableOnce[BreezeVectorWithNorm],
-      point: BreezeVectorWithNorm): Double =
+      centers: TraversableOnce[VectorWithNorm],
+      point: VectorWithNorm): Double =
     findClosest(centers, point)._2
 
   /**
@@ -411,26 +412,24 @@ object KMeans {
    * [[org.apache.spark.mllib.util.MLUtils#fastSquaredDistance]].
    */
   private[clustering] def fastSquaredDistance(
-      v1: BreezeVectorWithNorm,
-      v2: BreezeVectorWithNorm): Double = {
+      v1: VectorWithNorm,
+      v2: VectorWithNorm): Double = {
     MLUtils.fastSquaredDistance(v1.vector, v1.norm, v2.vector, v2.norm)
   }
 }
 
 /**
- * A breeze vector with its norm for fast distance computation.
+ * A vector with its norm for fast distance computation.
  *
  * @see [[org.apache.spark.mllib.clustering.KMeans#fastSquaredDistance]]
  */
 private[clustering]
-class BreezeVectorWithNorm(val vector: BV[Double], val norm: Double) extends Serializable {
-
-  def this(vector: BV[Double]) = this(vector, breezeNorm(vector, 2.0))
+class VectorWithNorm(val vector: Vector, val norm: Double) extends Serializable {
 
-  def this(array: Array[Double]) = this(new BDV[Double](array))
+  def this(vector: Vector) = this(vector, Vectors.norm(vector, 2.0))
 
-  def this(v: Vector) = this(v.toBreeze)
+  def this(array: Array[Double]) = this(Vectors.dense(array))
 
   /** Converts the vector to a dense vector. */
-  def toDense = new BreezeVectorWithNorm(vector.toDenseVector, norm)
+  def toDense = new VectorWithNorm(Vectors.dense(vector.toArray), norm)
 }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeansModel.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeansModel.scala
index 12a3d91cd31a6..3b95a9e6936e8 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeansModel.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/KMeansModel.scala
@@ -32,14 +32,14 @@ class KMeansModel (val clusterCenters: Array[Vector]) extends Serializable {
 
   /** Returns the cluster index that a given point belongs to. */
   def predict(point: Vector): Int = {
-    KMeans.findClosest(clusterCentersWithNorm, new BreezeVectorWithNorm(point))._1
+    KMeans.findClosest(clusterCentersWithNorm, new VectorWithNorm(point))._1
   }
 
   /** Maps given points to their cluster indices. */
   def predict(points: RDD[Vector]): RDD[Int] = {
     val centersWithNorm = clusterCentersWithNorm
     val bcCentersWithNorm = points.context.broadcast(centersWithNorm)
-    points.map(p => KMeans.findClosest(bcCentersWithNorm.value, new BreezeVectorWithNorm(p))._1)
+    points.map(p => KMeans.findClosest(bcCentersWithNorm.value, new VectorWithNorm(p))._1)
   }
 
   /** Maps given points to their cluster indices. */
@@ -53,9 +53,9 @@ class KMeansModel (val clusterCenters: Array[Vector]) extends Serializable {
   def computeCost(data: RDD[Vector]): Double = {
     val centersWithNorm = clusterCentersWithNorm
     val bcCentersWithNorm = data.context.broadcast(centersWithNorm)
-    data.map(p => KMeans.pointCost(bcCentersWithNorm.value, new BreezeVectorWithNorm(p))).sum()
+    data.map(p => KMeans.pointCost(bcCentersWithNorm.value, new VectorWithNorm(p))).sum()
   }
 
-  private def clusterCentersWithNorm: Iterable[BreezeVectorWithNorm] =
-    clusterCenters.map(new BreezeVectorWithNorm(_))
+  private def clusterCentersWithNorm: Iterable[VectorWithNorm] =
+    clusterCenters.map(new VectorWithNorm(_))
 }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LocalKMeans.scala b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LocalKMeans.scala
index f0722d7c14a46..b2f140e1b1352 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/clustering/LocalKMeans.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/clustering/LocalKMeans.scala
@@ -19,9 +19,9 @@ package org.apache.spark.mllib.clustering
 
 import scala.util.Random
 
-import breeze.linalg.{Vector => BV, DenseVector => BDV, norm => breezeNorm}
-
 import org.apache.spark.Logging
+import org.apache.spark.mllib.linalg.Vectors
+import org.apache.spark.mllib.linalg.BLAS.{axpy, scal}
 
 /**
  * An utility object to run K-means locally. This is private to the ML package because it's used
@@ -35,14 +35,14 @@ private[mllib] object LocalKMeans extends Logging {
    */
   def kMeansPlusPlus(
       seed: Int,
-      points: Array[BreezeVectorWithNorm],
+      points: Array[VectorWithNorm],
       weights: Array[Double],
       k: Int,
       maxIterations: Int
-  ): Array[BreezeVectorWithNorm] = {
+  ): Array[VectorWithNorm] = {
     val rand = new Random(seed)
-    val dimensions = points(0).vector.length
-    val centers = new Array[BreezeVectorWithNorm](k)
+    val dimensions = points(0).vector.size
+    val centers = new Array[VectorWithNorm](k)
 
     // Initialize centers by sampling using the k-means++ procedure.
     centers(0) = pickWeighted(rand, points, weights).toDense
@@ -75,14 +75,12 @@ private[mllib] object LocalKMeans extends Logging {
     while (moved && iteration < maxIterations) {
       moved = false
       val counts = Array.fill(k)(0.0)
-      val sums = Array.fill(k)(
-        BDV.zeros[Double](dimensions).asInstanceOf[BV[Double]]
-      )
+      val sums = Array.fill(k)(Vectors.zeros(dimensions))
       var i = 0
       while (i < points.length) {
         val p = points(i)
         val index = KMeans.findClosest(centers, p)._1
-        breeze.linalg.axpy(weights(i), p.vector, sums(index))
+        axpy(weights(i), p.vector, sums(index))
         counts(index) += weights(i)
         if (index != oldClosest(i)) {
           moved = true
@@ -97,8 +95,8 @@ private[mllib] object LocalKMeans extends Logging {
           // Assign center to a random point
           centers(j) = points(rand.nextInt(points.length)).toDense
         } else {
-          sums(j) /= counts(j)
-          centers(j) = new BreezeVectorWithNorm(sums(j))
+          scal(1.0 / counts(j), sums(j))
+          centers(j) = new VectorWithNorm(sums(j))
         }
         j += 1
       }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/feature/Normalizer.scala b/mllib/src/main/scala/org/apache/spark/mllib/feature/Normalizer.scala
index a9c2e23717896..1ced26a9b70a2 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/feature/Normalizer.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/feature/Normalizer.scala
@@ -17,8 +17,6 @@
 
 package org.apache.spark.mllib.feature
 
-import breeze.linalg.{norm => brzNorm}
-
 import org.apache.spark.annotation.Experimental
 import org.apache.spark.mllib.linalg.{DenseVector, SparseVector, Vector, Vectors}
 
@@ -47,7 +45,7 @@ class Normalizer(p: Double) extends VectorTransformer {
    * @return normalized vector. If the norm of the input is zero, it will return the input vector.
    */
   override def transform(vector: Vector): Vector = {
-    val norm = brzNorm(vector.toBreeze, p)
+    val norm = Vectors.norm(vector, p)
 
     if (norm != 0.0) {
       // For dense vector, we've to allocate new memory for new output vector.
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/BLAS.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/BLAS.scala
index 89539e600f48c..9fed513becddc 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/BLAS.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/BLAS.scala
@@ -72,17 +72,21 @@ private[spark] object BLAS extends Serializable with Logging {
    * y += a * x
    */
   private def axpy(a: Double, x: SparseVector, y: DenseVector): Unit = {
-    val nnz = x.indices.size
+    val xValues = x.values
+    val xIndices = x.indices
+    val yValues = y.values
+    val nnz = xIndices.size
+
     if (a == 1.0) {
       var k = 0
       while (k < nnz) {
-        y.values(x.indices(k)) += x.values(k)
+        yValues(xIndices(k)) += xValues(k)
         k += 1
       }
     } else {
       var k = 0
       while (k < nnz) {
-        y.values(x.indices(k)) += a * x.values(k)
+        yValues(xIndices(k)) += a * xValues(k)
         k += 1
       }
     }
@@ -92,7 +96,9 @@ private[spark] object BLAS extends Serializable with Logging {
    * dot(x, y)
    */
   def dot(x: Vector, y: Vector): Double = {
-    require(x.size == y.size)
+    require(x.size == y.size,
+      "BLAS.dot(x: Vector, y:Vector) was given Vectors with non-matching sizes:" +
+      " x.size = " + x.size + ", y.size = " + y.size)
     (x, y) match {
       case (dx: DenseVector, dy: DenseVector) =>
         dot(dx, dy)
@@ -119,11 +125,15 @@ private[spark] object BLAS extends Serializable with Logging {
    * dot(x, y)
    */
   private def dot(x: SparseVector, y: DenseVector): Double = {
-    val nnz = x.indices.size
+    val xValues = x.values
+    val xIndices = x.indices
+    val yValues = y.values
+    val nnz = xIndices.size
+
     var sum = 0.0
     var k = 0
     while (k < nnz) {
-      sum += x.values(k) * y.values(x.indices(k))
+      sum += xValues(k) * yValues(xIndices(k))
       k += 1
     }
     sum
@@ -133,19 +143,24 @@ private[spark] object BLAS extends Serializable with Logging {
    * dot(x, y)
    */
   private def dot(x: SparseVector, y: SparseVector): Double = {
+    val xValues = x.values
+    val xIndices = x.indices
+    val yValues = y.values
+    val yIndices = y.indices
+    val nnzx = xIndices.size
+    val nnzy = yIndices.size
+
     var kx = 0
-    val nnzx = x.indices.size
     var ky = 0
-    val nnzy = y.indices.size
     var sum = 0.0
     // y catching x
     while (kx < nnzx && ky < nnzy) {
-      val ix = x.indices(kx)
-      while (ky < nnzy && y.indices(ky) < ix) {
+      val ix = xIndices(kx)
+      while (ky < nnzy && yIndices(ky) < ix) {
         ky += 1
       }
-      if (ky < nnzy && y.indices(ky) == ix) {
-        sum += x.values(kx) * y.values(ky)
+      if (ky < nnzy && yIndices(ky) == ix) {
+        sum += xValues(kx) * yValues(ky)
         ky += 1
       }
       kx += 1
@@ -163,21 +178,25 @@ private[spark] object BLAS extends Serializable with Logging {
       case dy: DenseVector =>
         x match {
           case sx: SparseVector =>
+            val sxIndices = sx.indices
+            val sxValues = sx.values
+            val dyValues = dy.values
+            val nnz = sxIndices.size
+
             var i = 0
             var k = 0
-            val nnz = sx.indices.size
             while (k < nnz) {
-              val j = sx.indices(k)
+              val j = sxIndices(k)
               while (i < j) {
-                dy.values(i) = 0.0
+                dyValues(i) = 0.0
                 i += 1
               }
-              dy.values(i) = sx.values(k)
+              dyValues(i) = sxValues(k)
               i += 1
               k += 1
             }
             while (i < n) {
-              dy.values(i) = 0.0
+              dyValues(i) = 0.0
               i += 1
             }
           case dx: DenseVector =>
@@ -311,6 +330,8 @@ private[spark] object BLAS extends Serializable with Logging {
       s"The columns of C don't match the columns of B. C: ${C.numCols}, A: $nB")
 
     val Avals = A.values
+    val Bvals = B.values
+    val Cvals = C.values
     val Arows = if (!transA) A.rowIndices else A.colPtrs
     val Acols = if (!transA) A.colPtrs else A.rowIndices
 
@@ -327,11 +348,11 @@ private[spark] object BLAS extends Serializable with Logging {
             val indEnd = Arows(rowCounterForA + 1)
             var sum = 0.0
             while (i < indEnd) {
-              sum += Avals(i) * B.values(Bstart + Acols(i))
+              sum += Avals(i) * Bvals(Bstart + Acols(i))
               i += 1
             }
             val Cindex = Cstart + rowCounterForA
-            C.values(Cindex) = beta * C.values(Cindex) + sum * alpha
+            Cvals(Cindex) = beta * Cvals(Cindex) + sum * alpha
             rowCounterForA += 1
           }
           colCounterForB += 1
@@ -349,7 +370,7 @@ private[spark] object BLAS extends Serializable with Logging {
               i += 1
             }
             val Cindex = Cstart + rowCounter
-            C.values(Cindex) = beta * C.values(Cindex) + sum * alpha
+            Cvals(Cindex) = beta * Cvals(Cindex) + sum * alpha
             rowCounter += 1
           }
           colCounterForB += 1
@@ -357,7 +378,7 @@ private[spark] object BLAS extends Serializable with Logging {
       }
     } else {
       // Scale matrix first if `beta` is not equal to 0.0
-      if (beta != 0.0){
+      if (beta != 0.0) {
         f2jBLAS.dscal(C.values.length, beta, C.values, 1)
       }
       // Perform matrix multiplication and add to C. The rows of A are multiplied by the columns of
@@ -371,9 +392,9 @@ private[spark] object BLAS extends Serializable with Logging {
           while (colCounterForA < kA) {
             var i = Acols(colCounterForA)
             val indEnd = Acols(colCounterForA + 1)
-            val Bval = B.values(Bstart + colCounterForA) * alpha
-            while (i < indEnd){
-              C.values(Cstart + Arows(i)) += Avals(i) * Bval
+            val Bval = Bvals(Bstart + colCounterForA) * alpha
+            while (i < indEnd) {
+              Cvals(Cstart + Arows(i)) += Avals(i) * Bval
               i += 1
             }
             colCounterForA += 1
@@ -384,12 +405,12 @@ private[spark] object BLAS extends Serializable with Logging {
         while (colCounterForB < nB) {
           var colCounterForA = 0 // The column of A to multiply with the row of B
           val Cstart = colCounterForB * mA
-          while (colCounterForA < kA){
+          while (colCounterForA < kA) {
             var i = Acols(colCounterForA)
             val indEnd = Acols(colCounterForA + 1)
             val Bval = B(colCounterForB, colCounterForA) * alpha
-            while (i < indEnd){
-              C.values(Cstart + Arows(i)) += Avals(i) * Bval
+            while (i < indEnd) {
+              Cvals(Cstart + Arows(i)) += Avals(i) * Bval
               i += 1
             }
             colCounterForA += 1
@@ -484,41 +505,43 @@ private[spark] object BLAS extends Serializable with Logging {
       beta: Double,
       y: DenseVector): Unit =  {
 
-    val mA: Int = if(!trans) A.numRows else A.numCols
-    val nA: Int = if(!trans) A.numCols else A.numRows
+    val xValues = x.values
+    val yValues = y.values
+
+    val mA: Int = if (!trans) A.numRows else A.numCols
+    val nA: Int = if (!trans) A.numCols else A.numRows
 
     val Avals = A.values
     val Arows = if (!trans) A.rowIndices else A.colPtrs
     val Acols = if (!trans) A.colPtrs else A.rowIndices
-
     // Slicing is easy in this case. This is the optimal multiplication setting for sparse matrices
-    if (trans){
+    if (trans) {
       var rowCounter = 0
-      while (rowCounter < mA){
+      while (rowCounter < mA) {
         var i = Arows(rowCounter)
         val indEnd = Arows(rowCounter + 1)
         var sum = 0.0
-        while(i < indEnd){
-          sum += Avals(i) * x.values(Acols(i))
+        while (i < indEnd) {
+          sum += Avals(i) * xValues(Acols(i))
           i += 1
         }
-        y.values(rowCounter) =  beta * y.values(rowCounter) + sum * alpha
+        yValues(rowCounter) = beta * yValues(rowCounter) + sum * alpha
         rowCounter += 1
       }
     } else {
       // Scale vector first if `beta` is not equal to 0.0
-      if (beta != 0.0){
+      if (beta != 0.0) {
         scal(beta, y)
       }
       // Perform matrix-vector multiplication and add to y
       var colCounterForA = 0
-      while (colCounterForA < nA){
+      while (colCounterForA < nA) {
         var i = Acols(colCounterForA)
         val indEnd = Acols(colCounterForA + 1)
-        val xVal = x.values(colCounterForA) * alpha
-        while (i < indEnd){
+        val xVal = xValues(colCounterForA) * alpha
+        while (i < indEnd) {
           val rowIndex = Arows(i)
-          y.values(rowIndex) += Avals(i) * xVal
+          yValues(rowIndex) += Avals(i) * xVal
           i += 1
         }
         colCounterForA += 1
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
index c6d5fe5bc678c..47d1a76fa361d 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
@@ -261,6 +261,57 @@ object Vectors {
         sys.error("Unsupported Breeze vector type: " + v.getClass.getName)
     }
   }
+
+  /**
+   * Returns the p-norm of this vector.
+   * @param vector input vector.
+   * @param p norm.
+   * @return norm in L^p^ space.
+   */
+  private[spark] def norm(vector: Vector, p: Double): Double = {
+    require(p >= 1.0)
+    val values = vector match {
+      case dv: DenseVector => dv.values
+      case sv: SparseVector => sv.values
+      case v => throw new IllegalArgumentException("Do not support vector type " + v.getClass)
+    }
+    val size = values.size
+
+    if (p == 1) {
+      var sum = 0.0
+      var i = 0
+      while (i < size) {
+        sum += math.abs(values(i))
+        i += 1
+      }
+      sum
+    } else if (p == 2) {
+      var sum = 0.0
+      var i = 0
+      while (i < size) {
+        sum += values(i) * values(i)
+        i += 1
+      }
+      math.sqrt(sum)
+    } else if (p == Double.PositiveInfinity) {
+      var max = 0.0
+      var i = 0
+      while (i < size) {
+        val value = math.abs(values(i))
+        if (value > max) max = value
+        i += 1
+      }
+      max
+    } else {
+      var sum = 0.0
+      var i = 0
+      while (i < size) {
+        sum += math.pow(math.abs(values(i)), p)
+        i += 1
+      }
+      math.pow(sum, 1.0 / p)
+    }
+  }
 }
 
 /**
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/DecisionTree.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/DecisionTree.scala
index 3d91867c896d9..73e7e32c6db31 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/DecisionTree.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/DecisionTree.scala
@@ -136,7 +136,7 @@ object DecisionTree extends Serializable with Logging {
    * @param impurity impurity criterion used for information gain calculation
    * @param maxDepth Maximum depth of the tree.
    *                 E.g., depth 0 means 1 leaf node; depth 1 means 1 internal node + 2 leaf nodes.
-   * @param numClassesForClassification number of classes for classification. Default value of 2.
+   * @param numClasses number of classes for classification. Default value of 2.
    * @return DecisionTreeModel that can be used for prediction
    */
   def train(
@@ -144,8 +144,8 @@ object DecisionTree extends Serializable with Logging {
       algo: Algo,
       impurity: Impurity,
       maxDepth: Int,
-      numClassesForClassification: Int): DecisionTreeModel = {
-    val strategy = new Strategy(algo, impurity, maxDepth, numClassesForClassification)
+      numClasses: Int): DecisionTreeModel = {
+    val strategy = new Strategy(algo, impurity, maxDepth, numClasses)
     new DecisionTree(strategy).run(input)
   }
 
@@ -164,7 +164,7 @@ object DecisionTree extends Serializable with Logging {
    * @param impurity criterion used for information gain calculation
    * @param maxDepth Maximum depth of the tree.
    *                 E.g., depth 0 means 1 leaf node; depth 1 means 1 internal node + 2 leaf nodes.
-   * @param numClassesForClassification number of classes for classification. Default value of 2.
+   * @param numClasses number of classes for classification. Default value of 2.
    * @param maxBins maximum number of bins used for splitting features
    * @param quantileCalculationStrategy  algorithm for calculating quantiles
    * @param categoricalFeaturesInfo Map storing arity of categorical features.
@@ -177,11 +177,11 @@ object DecisionTree extends Serializable with Logging {
       algo: Algo,
       impurity: Impurity,
       maxDepth: Int,
-      numClassesForClassification: Int,
+      numClasses: Int,
       maxBins: Int,
       quantileCalculationStrategy: QuantileStrategy,
       categoricalFeaturesInfo: Map[Int,Int]): DecisionTreeModel = {
-    val strategy = new Strategy(algo, impurity, maxDepth, numClassesForClassification, maxBins,
+    val strategy = new Strategy(algo, impurity, maxDepth, numClasses, maxBins,
       quantileCalculationStrategy, categoricalFeaturesInfo)
     new DecisionTree(strategy).run(input)
   }
@@ -191,7 +191,7 @@ object DecisionTree extends Serializable with Logging {
    *
    * @param input Training dataset: RDD of [[org.apache.spark.mllib.regression.LabeledPoint]].
    *              Labels should take values {0, 1, ..., numClasses-1}.
-   * @param numClassesForClassification number of classes for classification.
+   * @param numClasses number of classes for classification.
    * @param categoricalFeaturesInfo Map storing arity of categorical features.
    *                                E.g., an entry (n -> k) indicates that feature n is categorical
    *                                with k categories indexed from 0: {0, 1, ..., k-1}.
@@ -206,13 +206,13 @@ object DecisionTree extends Serializable with Logging {
    */
   def trainClassifier(
       input: RDD[LabeledPoint],
-      numClassesForClassification: Int,
+      numClasses: Int,
       categoricalFeaturesInfo: Map[Int, Int],
       impurity: String,
       maxDepth: Int,
       maxBins: Int): DecisionTreeModel = {
     val impurityType = Impurities.fromString(impurity)
-    train(input, Classification, impurityType, maxDepth, numClassesForClassification, maxBins, Sort,
+    train(input, Classification, impurityType, maxDepth, numClasses, maxBins, Sort,
       categoricalFeaturesInfo)
   }
 
@@ -221,12 +221,12 @@ object DecisionTree extends Serializable with Logging {
    */
   def trainClassifier(
       input: JavaRDD[LabeledPoint],
-      numClassesForClassification: Int,
+      numClasses: Int,
       categoricalFeaturesInfo: java.util.Map[java.lang.Integer, java.lang.Integer],
       impurity: String,
       maxDepth: Int,
       maxBins: Int): DecisionTreeModel = {
-    trainClassifier(input.rdd, numClassesForClassification,
+    trainClassifier(input.rdd, numClasses,
       categoricalFeaturesInfo.asInstanceOf[java.util.Map[Int, Int]].asScala.toMap,
       impurity, maxDepth, maxBins)
   }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala
index 482d3395516e7..e9304b5e5c650 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/RandomForest.scala
@@ -56,7 +56,7 @@ import org.apache.spark.util.Utils
  *                 etc.
  * @param numTrees If 1, then no bootstrapping is used.  If > 1, then bootstrapping is done.
  * @param featureSubsetStrategy Number of features to consider for splits at each node.
- *                              Supported: "auto" (default), "all", "sqrt", "log2", "onethird".
+ *                              Supported: "auto", "all", "sqrt", "log2", "onethird".
  *                              If "auto" is set, this parameter is set based on numTrees:
  *                                if numTrees == 1, set to "all";
  *                                if numTrees > 1 (forest) set to "sqrt" for classification and
@@ -269,7 +269,7 @@ object RandomForest extends Serializable with Logging {
    * @param strategy Parameters for training each tree in the forest.
    * @param numTrees Number of trees in the random forest.
    * @param featureSubsetStrategy Number of features to consider for splits at each node.
-   *                              Supported: "auto" (default), "all", "sqrt", "log2", "onethird".
+   *                              Supported: "auto", "all", "sqrt", "log2", "onethird".
    *                              If "auto" is set, this parameter is set based on numTrees:
    *                                if numTrees == 1, set to "all";
    *                                if numTrees > 1 (forest) set to "sqrt".
@@ -293,13 +293,13 @@ object RandomForest extends Serializable with Logging {
    *
    * @param input Training dataset: RDD of [[org.apache.spark.mllib.regression.LabeledPoint]].
    *              Labels should take values {0, 1, ..., numClasses-1}.
-   * @param numClassesForClassification number of classes for classification.
+   * @param numClasses number of classes for classification.
    * @param categoricalFeaturesInfo Map storing arity of categorical features.
    *                                E.g., an entry (n -> k) indicates that feature n is categorical
    *                                with k categories indexed from 0: {0, 1, ..., k-1}.
    * @param numTrees Number of trees in the random forest.
    * @param featureSubsetStrategy Number of features to consider for splits at each node.
-   *                              Supported: "auto" (default), "all", "sqrt", "log2", "onethird".
+   *                              Supported: "auto", "all", "sqrt", "log2", "onethird".
    *                              If "auto" is set, this parameter is set based on numTrees:
    *                                if numTrees == 1, set to "all";
    *                                if numTrees > 1 (forest) set to "sqrt".
@@ -315,7 +315,7 @@ object RandomForest extends Serializable with Logging {
    */
   def trainClassifier(
       input: RDD[LabeledPoint],
-      numClassesForClassification: Int,
+      numClasses: Int,
       categoricalFeaturesInfo: Map[Int, Int],
       numTrees: Int,
       featureSubsetStrategy: String,
@@ -325,7 +325,7 @@ object RandomForest extends Serializable with Logging {
       seed: Int = Utils.random.nextInt()): RandomForestModel = {
     val impurityType = Impurities.fromString(impurity)
     val strategy = new Strategy(Classification, impurityType, maxDepth,
-      numClassesForClassification, maxBins, Sort, categoricalFeaturesInfo)
+      numClasses, maxBins, Sort, categoricalFeaturesInfo)
     trainClassifier(input, strategy, numTrees, featureSubsetStrategy, seed)
   }
 
@@ -334,7 +334,7 @@ object RandomForest extends Serializable with Logging {
    */
   def trainClassifier(
       input: JavaRDD[LabeledPoint],
-      numClassesForClassification: Int,
+      numClasses: Int,
       categoricalFeaturesInfo: java.util.Map[java.lang.Integer, java.lang.Integer],
       numTrees: Int,
       featureSubsetStrategy: String,
@@ -342,7 +342,7 @@ object RandomForest extends Serializable with Logging {
       maxDepth: Int,
       maxBins: Int,
       seed: Int): RandomForestModel = {
-    trainClassifier(input.rdd, numClassesForClassification,
+    trainClassifier(input.rdd, numClasses,
       categoricalFeaturesInfo.asInstanceOf[java.util.Map[Int, Int]].asScala.toMap,
       numTrees, featureSubsetStrategy, impurity, maxDepth, maxBins, seed)
   }
@@ -355,7 +355,7 @@ object RandomForest extends Serializable with Logging {
    * @param strategy Parameters for training each tree in the forest.
    * @param numTrees Number of trees in the random forest.
    * @param featureSubsetStrategy Number of features to consider for splits at each node.
-   *                              Supported: "auto" (default), "all", "sqrt", "log2", "onethird".
+   *                              Supported: "auto", "all", "sqrt", "log2", "onethird".
    *                              If "auto" is set, this parameter is set based on numTrees:
    *                                if numTrees == 1, set to "all";
    *                                if numTrees > 1 (forest) set to "onethird".
@@ -384,7 +384,7 @@ object RandomForest extends Serializable with Logging {
    *                                with k categories indexed from 0: {0, 1, ..., k-1}.
    * @param numTrees Number of trees in the random forest.
    * @param featureSubsetStrategy Number of features to consider for splits at each node.
-   *                              Supported: "auto" (default), "all", "sqrt", "log2", "onethird".
+   *                              Supported: "auto", "all", "sqrt", "log2", "onethird".
    *                              If "auto" is set, this parameter is set based on numTrees:
    *                                if numTrees == 1, set to "all";
    *                                if numTrees > 1 (forest) set to "onethird".
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/BoostingStrategy.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/BoostingStrategy.scala
index e703adbdbfbb3..cf51d041c65a9 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/BoostingStrategy.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/BoostingStrategy.scala
@@ -51,7 +51,7 @@ case class BoostingStrategy(
   private[tree] def assertValid(): Unit = {
     treeStrategy.algo match {
       case Classification =>
-        require(treeStrategy.numClassesForClassification == 2,
+        require(treeStrategy.numClasses == 2,
           "Only binary classification is supported for boosting.")
       case Regression =>
         // nothing
@@ -80,12 +80,12 @@ object BoostingStrategy {
     treeStrategy.maxDepth = 3
     algo match {
       case "Classification" =>
-        treeStrategy.numClassesForClassification = 2
+        treeStrategy.numClasses = 2
         new BoostingStrategy(treeStrategy, LogLoss)
       case "Regression" =>
         new BoostingStrategy(treeStrategy, SquaredError)
       case _ =>
-        throw new IllegalArgumentException(s"$algo is not supported by the boosting.")
+        throw new IllegalArgumentException(s"$algo is not supported by boosting.")
     }
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala
index d75f38433c081..d5cd89ab94e81 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/configuration/Strategy.scala
@@ -37,7 +37,7 @@ import org.apache.spark.mllib.tree.configuration.QuantileStrategy._
  *                 Supported for Regression: [[org.apache.spark.mllib.tree.impurity.Variance]].
  * @param maxDepth Maximum depth of the tree.
  *                 E.g., depth 0 means 1 leaf node; depth 1 means 1 internal node + 2 leaf nodes.
- * @param numClassesForClassification Number of classes for classification.
+ * @param numClasses Number of classes for classification.
  *                                    (Ignored for regression.)
  *                                    Default value is 2 (binary classification).
  * @param maxBins Maximum number of bins used for discretizing continuous features and
@@ -73,7 +73,7 @@ class Strategy (
     @BeanProperty var algo: Algo,
     @BeanProperty var impurity: Impurity,
     @BeanProperty var maxDepth: Int,
-    @BeanProperty var numClassesForClassification: Int = 2,
+    @BeanProperty var numClasses: Int = 2,
     @BeanProperty var maxBins: Int = 32,
     @BeanProperty var quantileCalculationStrategy: QuantileStrategy = Sort,
     @BeanProperty var categoricalFeaturesInfo: Map[Int, Int] = Map[Int, Int](),
@@ -86,7 +86,7 @@ class Strategy (
     @BeanProperty var checkpointInterval: Int = 10) extends Serializable {
 
   def isMulticlassClassification =
-    algo == Classification && numClassesForClassification > 2
+    algo == Classification && numClasses > 2
   def isMulticlassWithCategoricalFeatures
     = isMulticlassClassification && (categoricalFeaturesInfo.size > 0)
 
@@ -97,10 +97,10 @@ class Strategy (
       algo: Algo,
       impurity: Impurity,
       maxDepth: Int,
-      numClassesForClassification: Int,
+      numClasses: Int,
       maxBins: Int,
       categoricalFeaturesInfo: java.util.Map[java.lang.Integer, java.lang.Integer]) {
-    this(algo, impurity, maxDepth, numClassesForClassification, maxBins, Sort,
+    this(algo, impurity, maxDepth, numClasses, maxBins, Sort,
       categoricalFeaturesInfo.asInstanceOf[java.util.Map[Int, Int]].asScala.toMap)
   }
 
@@ -117,8 +117,8 @@ class Strategy (
    */
   def setCategoricalFeaturesInfo(
       categoricalFeaturesInfo: java.util.Map[java.lang.Integer, java.lang.Integer]): Unit = {
-    setCategoricalFeaturesInfo(
-      categoricalFeaturesInfo.asInstanceOf[java.util.Map[Int, Int]].asScala.toMap)
+    this.categoricalFeaturesInfo =
+      categoricalFeaturesInfo.asInstanceOf[java.util.Map[Int, Int]].asScala.toMap
   }
 
   /**
@@ -128,9 +128,9 @@ class Strategy (
   private[tree] def assertValid(): Unit = {
     algo match {
       case Classification =>
-        require(numClassesForClassification >= 2,
-          s"DecisionTree Strategy for Classification must have numClassesForClassification >= 2," +
-          s" but numClassesForClassification = $numClassesForClassification.")
+        require(numClasses >= 2,
+          s"DecisionTree Strategy for Classification must have numClasses >= 2," +
+          s" but numClasses = $numClasses.")
         require(Set(Gini, Entropy).contains(impurity),
           s"DecisionTree Strategy given invalid impurity for Classification: $impurity." +
           s"  Valid settings: Gini, Entropy")
@@ -160,7 +160,7 @@ class Strategy (
 
   /** Returns a shallow copy of this instance. */
   def copy: Strategy = {
-    new Strategy(algo, impurity, maxDepth, numClassesForClassification, maxBins,
+    new Strategy(algo, impurity, maxDepth, numClasses, maxBins,
       quantileCalculationStrategy, categoricalFeaturesInfo, minInstancesPerNode, minInfoGain,
       maxMemoryInMB, subsamplingRate, useNodeIdCache, checkpointDir, checkpointInterval)
   }
@@ -176,9 +176,9 @@ object Strategy {
   def defaultStrategy(algo: String): Strategy = algo match {
     case "Classification" =>
       new Strategy(algo = Classification, impurity = Gini, maxDepth = 10,
-        numClassesForClassification = 2)
+        numClasses = 2)
     case "Regression" =>
       new Strategy(algo = Regression, impurity = Variance, maxDepth = 10,
-        numClassesForClassification = 0)
+        numClasses = 0)
   }
 }
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala
index 5bc0f2635c6b1..951733fada6be 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/tree/impl/DecisionTreeMetadata.scala
@@ -110,7 +110,7 @@ private[tree] object DecisionTreeMetadata extends Logging {
     val numFeatures = input.take(1)(0).features.size
     val numExamples = input.count()
     val numClasses = strategy.algo match {
-      case Classification => strategy.numClassesForClassification
+      case Classification => strategy.numClasses
       case Regression => 0
     }
 
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala b/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
index 9353351af72a0..b0d05ae33e1b5 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala
@@ -19,7 +19,7 @@ package org.apache.spark.mllib.util
 
 import scala.reflect.ClassTag
 
-import breeze.linalg.{Vector => BV, DenseVector => BDV, SparseVector => BSV,
+import breeze.linalg.{DenseVector => BDV, SparseVector => BSV,
   squaredDistance => breezeSquaredDistance}
 
 import org.apache.spark.annotation.Experimental
@@ -28,7 +28,8 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.rdd.PartitionwiseSampledRDD
 import org.apache.spark.util.random.BernoulliCellSampler
 import org.apache.spark.mllib.regression.LabeledPoint
-import org.apache.spark.mllib.linalg.{Vector, Vectors}
+import org.apache.spark.mllib.linalg.{SparseVector, Vector, Vectors}
+import org.apache.spark.mllib.linalg.BLAS.dot
 import org.apache.spark.storage.StorageLevel
 import org.apache.spark.streaming.StreamingContext
 import org.apache.spark.streaming.dstream.DStream
@@ -281,9 +282,9 @@ object MLUtils {
    * @return squared distance between v1 and v2 within the specified precision
    */
   private[mllib] def fastSquaredDistance(
-      v1: BV[Double],
+      v1: Vector,
       norm1: Double,
-      v2: BV[Double],
+      v2: Vector,
       norm2: Double,
       precision: Double = 1e-6): Double = {
     val n = v1.size
@@ -306,16 +307,19 @@ object MLUtils {
      */
     val precisionBound1 = 2.0 * EPSILON * sumSquaredNorm / (normDiff * normDiff + EPSILON)
     if (precisionBound1 < precision) {
-      sqDist = sumSquaredNorm - 2.0 * v1.dot(v2)
-    } else if (v1.isInstanceOf[BSV[Double]] || v2.isInstanceOf[BSV[Double]]) {
-      val dot = v1.dot(v2)
-      sqDist = math.max(sumSquaredNorm - 2.0 * dot, 0.0)
-      val precisionBound2 = EPSILON * (sumSquaredNorm + 2.0 * math.abs(dot)) / (sqDist + EPSILON)
+      sqDist = sumSquaredNorm - 2.0 * dot(v1, v2)
+    } else if (v1.isInstanceOf[SparseVector] || v2.isInstanceOf[SparseVector]) {
+      val dotValue = dot(v1, v2)
+      sqDist = math.max(sumSquaredNorm - 2.0 * dotValue, 0.0)
+      val precisionBound2 = EPSILON * (sumSquaredNorm + 2.0 * math.abs(dotValue)) /
+        (sqDist + EPSILON)
       if (precisionBound2 > precision) {
-        sqDist = breezeSquaredDistance(v1, v2)
+        // TODO: breezeSquaredDistance is slow,
+        // so we should replace it with our own implementation.
+        sqDist = breezeSquaredDistance(v1.toBreeze, v2.toBreeze)
       }
     } else {
-      sqDist = breezeSquaredDistance(v1, v2)
+      sqDist = breezeSquaredDistance(v1.toBreeze, v2.toBreeze)
     }
     sqDist
   }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/linalg/VectorsSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/linalg/VectorsSuite.scala
index 9492f604af4d5..f99f01450992a 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/linalg/VectorsSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/linalg/VectorsSuite.scala
@@ -21,6 +21,7 @@ import breeze.linalg.{DenseMatrix => BDM}
 import org.scalatest.FunSuite
 
 import org.apache.spark.SparkException
+import org.apache.spark.mllib.util.TestingUtils._
 
 class VectorsSuite extends FunSuite {
 
@@ -197,4 +198,27 @@ class VectorsSuite extends FunSuite {
     assert(svMap.get(2) === Some(3.1))
     assert(svMap.get(3) === Some(0.0))
   }
+
+  test("vector p-norm") {
+    val dv = Vectors.dense(0.0, -1.2, 3.1, 0.0, -4.5, 1.9)
+    val sv = Vectors.sparse(6, Seq((1, -1.2), (2, 3.1), (3, 0.0), (4, -4.5), (5, 1.9)))
+
+    assert(Vectors.norm(dv, 1.0) ~== dv.toArray.foldLeft(0.0)((a, v) =>
+      a + math.abs(v)) relTol 1E-8)
+    assert(Vectors.norm(sv, 1.0) ~== sv.toArray.foldLeft(0.0)((a, v) =>
+      a + math.abs(v)) relTol 1E-8)
+
+    assert(Vectors.norm(dv, 2.0) ~== math.sqrt(dv.toArray.foldLeft(0.0)((a, v) =>
+      a + v * v)) relTol 1E-8)
+    assert(Vectors.norm(sv, 2.0) ~== math.sqrt(sv.toArray.foldLeft(0.0)((a, v) =>
+      a + v * v)) relTol 1E-8)
+
+    assert(Vectors.norm(dv, Double.PositiveInfinity) ~== dv.toArray.map(math.abs).max relTol 1E-8)
+    assert(Vectors.norm(sv, Double.PositiveInfinity) ~== sv.toArray.map(math.abs).max relTol 1E-8)
+
+    assert(Vectors.norm(dv, 3.7) ~== math.pow(dv.toArray.foldLeft(0.0)((a, v) =>
+      a + math.pow(math.abs(v), 3.7)), 1.0 / 3.7) relTol 1E-8)
+    assert(Vectors.norm(sv, 3.7) ~== math.pow(sv.toArray.foldLeft(0.0)((a, v) =>
+      a + math.pow(math.abs(v), 3.7)), 1.0 / 3.7) relTol 1E-8)
+  }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala
index 972c905ec9ffa..9347eaf9221a8 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/DecisionTreeSuite.scala
@@ -57,7 +57,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
       Classification,
       Gini,
       maxDepth = 2,
-      numClassesForClassification = 2,
+      numClasses = 2,
       maxBins = 100,
       categoricalFeaturesInfo = Map(0 -> 2, 1-> 2))
 
@@ -81,7 +81,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
       Classification,
       Gini,
       maxDepth = 2,
-      numClassesForClassification = 2,
+      numClasses = 2,
       maxBins = 100,
       categoricalFeaturesInfo = Map(0 -> 3, 1 -> 3))
 
@@ -177,7 +177,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
       Classification,
       Gini,
       maxDepth = 2,
-      numClassesForClassification = 100,
+      numClasses = 100,
       maxBins = 100,
       categoricalFeaturesInfo = Map(0 -> 3, 1-> 3))
 
@@ -271,7 +271,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
       Classification,
       Gini,
       maxDepth = 2,
-      numClassesForClassification = 100,
+      numClasses = 100,
       maxBins = 100,
       categoricalFeaturesInfo = Map(0 -> 10, 1-> 10))
     // 2^(10-1) - 1 > 100, so categorical features will be ordered
@@ -295,7 +295,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val strategy = new Strategy(
       Classification,
       Gini,
-      numClassesForClassification = 2,
+      numClasses = 2,
       maxDepth = 2,
       maxBins = 100,
       categoricalFeaturesInfo = Map(0 -> 3, 1-> 3))
@@ -377,7 +377,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     assert(arr.length === 1000)
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(Classification, Gini, maxDepth = 3,
-      numClassesForClassification = 2, maxBins = 100)
+      numClasses = 2, maxBins = 100)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
     assert(!metadata.isUnordered(featureIndex = 0))
     assert(!metadata.isUnordered(featureIndex = 1))
@@ -401,7 +401,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     assert(arr.length === 1000)
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(Classification, Gini, maxDepth = 3,
-      numClassesForClassification = 2, maxBins = 100)
+      numClasses = 2, maxBins = 100)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
     assert(!metadata.isUnordered(featureIndex = 0))
     assert(!metadata.isUnordered(featureIndex = 1))
@@ -426,7 +426,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     assert(arr.length === 1000)
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(Classification, Entropy, maxDepth = 3,
-      numClassesForClassification = 2, maxBins = 100)
+      numClasses = 2, maxBins = 100)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
     assert(!metadata.isUnordered(featureIndex = 0))
     assert(!metadata.isUnordered(featureIndex = 1))
@@ -451,7 +451,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     assert(arr.length === 1000)
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(Classification, Entropy, maxDepth = 3,
-      numClassesForClassification = 2, maxBins = 100)
+      numClasses = 2, maxBins = 100)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
     assert(!metadata.isUnordered(featureIndex = 0))
     assert(!metadata.isUnordered(featureIndex = 1))
@@ -485,7 +485,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
 
     // Train a 1-node model
     val strategyOneNode = new Strategy(Classification, Entropy, maxDepth = 1,
-      numClassesForClassification = 2, maxBins = 100)
+      numClasses = 2, maxBins = 100)
     val modelOneNode = DecisionTree.train(rdd, strategyOneNode)
     val rootNode1 = modelOneNode.topNode.deepCopy()
     val rootNode2 = modelOneNode.topNode.deepCopy()
@@ -545,7 +545,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val arr = DecisionTreeSuite.generateCategoricalDataPointsForMulticlass()
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 3, categoricalFeaturesInfo = Map(0 -> 3, 1 -> 3))
+      numClasses = 3, categoricalFeaturesInfo = Map(0 -> 3, 1 -> 3))
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
     assert(strategy.isMulticlassClassification)
     assert(metadata.isUnordered(featureIndex = 0))
@@ -568,7 +568,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     arr(3) = new LabeledPoint(1.0, Vectors.dense(3.0))
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 2)
+      numClasses = 2)
 
     val model = DecisionTree.train(rdd, strategy)
     DecisionTreeSuite.validateClassifier(model, arr, 1.0)
@@ -585,7 +585,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
 
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 2)
+      numClasses = 2)
 
     val model = DecisionTree.train(rdd, strategy)
     DecisionTreeSuite.validateClassifier(model, arr, 1.0)
@@ -600,7 +600,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val arr = DecisionTreeSuite.generateCategoricalDataPointsForMulticlass()
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 3, maxBins = maxBins,
+      numClasses = 3, maxBins = maxBins,
       categoricalFeaturesInfo = Map(0 -> 3, 1 -> 3))
     assert(strategy.isMulticlassClassification)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
@@ -629,7 +629,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val arr = DecisionTreeSuite.generateContinuousDataPointsForMulticlass()
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 3, maxBins = 100)
+      numClasses = 3, maxBins = 100)
     assert(strategy.isMulticlassClassification)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
 
@@ -650,7 +650,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val arr = DecisionTreeSuite.generateContinuousDataPointsForMulticlass()
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 3, maxBins = 100, categoricalFeaturesInfo = Map(0 -> 3))
+      numClasses = 3, maxBins = 100, categoricalFeaturesInfo = Map(0 -> 3))
     assert(strategy.isMulticlassClassification)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
     assert(metadata.isUnordered(featureIndex = 0))
@@ -671,7 +671,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val arr = DecisionTreeSuite.generateCategoricalDataPointsForMulticlassForOrderedFeatures()
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 3, maxBins = 100,
+      numClasses = 3, maxBins = 100,
       categoricalFeaturesInfo = Map(0 -> 10, 1 -> 10))
     assert(strategy.isMulticlassClassification)
     val metadata = DecisionTreeMetadata.buildMetadata(rdd, strategy)
@@ -692,7 +692,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val arr = DecisionTreeSuite.generateCategoricalDataPointsForMulticlassForOrderedFeatures()
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 4,
-      numClassesForClassification = 3, maxBins = 10,
+      numClasses = 3, maxBins = 10,
       categoricalFeaturesInfo = Map(0 -> 10, 1 -> 10))
     assert(strategy.isMulticlassClassification)
 
@@ -708,7 +708,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
 
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini,
-      maxDepth = 2, numClassesForClassification = 2, minInstancesPerNode = 2)
+      maxDepth = 2, numClasses = 2, minInstancesPerNode = 2)
 
     val model = DecisionTree.train(rdd, strategy)
     assert(model.topNode.isLeaf)
@@ -737,7 +737,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val rdd = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini,
       maxBins = 2, maxDepth = 2, categoricalFeaturesInfo = Map(0 -> 2, 1-> 2),
-      numClassesForClassification = 2, minInstancesPerNode = 2)
+      numClasses = 2, minInstancesPerNode = 2)
 
     val rootNode = DecisionTree.train(rdd, strategy).topNode
 
@@ -755,7 +755,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
 
     val input = sc.parallelize(arr)
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 2,
-      numClassesForClassification = 2, minInfoGain = 1.0)
+      numClasses = 2, minInfoGain = 1.0)
 
     val model = DecisionTree.train(input, strategy)
     assert(model.topNode.isLeaf)
@@ -781,7 +781,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val input = sc.parallelize(arr)
 
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 1,
-      numClassesForClassification = 2, categoricalFeaturesInfo = Map(0 -> 3))
+      numClasses = 2, categoricalFeaturesInfo = Map(0 -> 3))
     val metadata = DecisionTreeMetadata.buildMetadata(input, strategy)
     val (splits, bins) = DecisionTree.findSplitsBins(input, metadata)
 
@@ -824,7 +824,7 @@ class DecisionTreeSuite extends FunSuite with MLlibTestSparkContext {
     val input = sc.parallelize(arr)
 
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 5,
-      numClassesForClassification = 2, categoricalFeaturesInfo = Map(0 -> 3))
+      numClasses = 2, categoricalFeaturesInfo = Map(0 -> 3))
     val metadata = DecisionTreeMetadata.buildMetadata(input, strategy)
     val (splits, bins) = DecisionTree.findSplitsBins(input, metadata)
 
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala
index d4d54cf4c9e2a..3aa97e544680b 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/GradientBoostedTreesSuite.scala
@@ -100,7 +100,7 @@ class GradientBoostedTreesSuite extends FunSuite with MLlibTestSparkContext {
         val rdd = sc.parallelize(GradientBoostedTreesSuite.data, 2)
 
         val treeStrategy = new Strategy(algo = Classification, impurity = Variance, maxDepth = 2,
-          numClassesForClassification = 2, categoricalFeaturesInfo = Map.empty,
+          numClasses = 2, categoricalFeaturesInfo = Map.empty,
           subsamplingRate = subsamplingRate)
         val boostingStrategy =
           new BoostingStrategy(treeStrategy, LogLoss, numIterations, learningRate)
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala
index 90a8c2dfdab80..f7f0f20c6c125 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/tree/RandomForestSuite.scala
@@ -57,7 +57,7 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext {
     " comparing DecisionTree vs. RandomForest(numTrees = 1)") {
     val categoricalFeaturesInfo = Map.empty[Int, Int]
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 2,
-      numClassesForClassification = 2, categoricalFeaturesInfo = categoricalFeaturesInfo)
+      numClasses = 2, categoricalFeaturesInfo = categoricalFeaturesInfo)
     binaryClassificationTestWithContinuousFeatures(strategy)
   }
 
@@ -65,7 +65,7 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext {
     " comparing DecisionTree vs. RandomForest(numTrees = 1)") {
     val categoricalFeaturesInfo = Map.empty[Int, Int]
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 2,
-      numClassesForClassification = 2, categoricalFeaturesInfo = categoricalFeaturesInfo,
+      numClasses = 2, categoricalFeaturesInfo = categoricalFeaturesInfo,
       useNodeIdCache = true)
     binaryClassificationTestWithContinuousFeatures(strategy)
   }
@@ -93,7 +93,7 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext {
     " comparing DecisionTree vs. RandomForest(numTrees = 1)") {
     val categoricalFeaturesInfo = Map.empty[Int, Int]
     val strategy = new Strategy(algo = Regression, impurity = Variance,
-      maxDepth = 2, maxBins = 10, numClassesForClassification = 2,
+      maxDepth = 2, maxBins = 10, numClasses = 2,
       categoricalFeaturesInfo = categoricalFeaturesInfo)
     regressionTestWithContinuousFeatures(strategy)
   }
@@ -102,7 +102,7 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext {
     " comparing DecisionTree vs. RandomForest(numTrees = 1)") {
     val categoricalFeaturesInfo = Map.empty[Int, Int]
     val strategy = new Strategy(algo = Regression, impurity = Variance,
-      maxDepth = 2, maxBins = 10, numClassesForClassification = 2,
+      maxDepth = 2, maxBins = 10, numClasses = 2,
       categoricalFeaturesInfo = categoricalFeaturesInfo, useNodeIdCache = true)
     regressionTestWithContinuousFeatures(strategy)
   }
@@ -169,14 +169,14 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext {
   test("Binary classification with continuous features: subsampling features") {
     val categoricalFeaturesInfo = Map.empty[Int, Int]
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 2,
-      numClassesForClassification = 2, categoricalFeaturesInfo = categoricalFeaturesInfo)
+      numClasses = 2, categoricalFeaturesInfo = categoricalFeaturesInfo)
     binaryClassificationTestWithContinuousFeaturesAndSubsampledFeatures(strategy)
   }
 
   test("Binary classification with continuous features and node Id cache: subsampling features") {
     val categoricalFeaturesInfo = Map.empty[Int, Int]
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 2,
-      numClassesForClassification = 2, categoricalFeaturesInfo = categoricalFeaturesInfo,
+      numClasses = 2, categoricalFeaturesInfo = categoricalFeaturesInfo,
       useNodeIdCache = true)
     binaryClassificationTestWithContinuousFeaturesAndSubsampledFeatures(strategy)
   }
@@ -191,7 +191,7 @@ class RandomForestSuite extends FunSuite with MLlibTestSparkContext {
     val input = sc.parallelize(arr)
 
     val strategy = new Strategy(algo = Classification, impurity = Gini, maxDepth = 5,
-      numClassesForClassification = 3, categoricalFeaturesInfo = categoricalFeaturesInfo)
+      numClasses = 3, categoricalFeaturesInfo = categoricalFeaturesInfo)
     val model = RandomForest.trainClassifier(input, strategy, numTrees = 2,
       featureSubsetStrategy = "sqrt", seed = 12345)
     EnsembleTestHelper.validateClassifier(model, arr, 1.0)
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/util/MLUtilsSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/util/MLUtilsSuite.scala
index 88bc49cc61f94..df07987093fbf 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/util/MLUtilsSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/util/MLUtilsSuite.scala
@@ -44,18 +44,19 @@ class MLUtilsSuite extends FunSuite with MLlibTestSparkContext {
   test("fast squared distance") {
     val a = (30 to 0 by -1).map(math.pow(2.0, _)).toArray
     val n = a.length
-    val v1 = new BDV[Double](a)
-    val norm1 = breezeNorm(v1, 2.0)
+    val v1 = Vectors.dense(a)
+    val norm1 = Vectors.norm(v1, 2.0)
     val precision = 1e-6
     for (m <- 0 until n) {
       val indices = (0 to m).toArray
       val values = indices.map(i => a(i))
-      val v2 = new BSV[Double](indices, values, n)
-      val norm2 = breezeNorm(v2, 2.0)
-      val squaredDist = breezeSquaredDistance(v1, v2)
+      val v2 = Vectors.sparse(n, indices, values)
+      val norm2 = Vectors.norm(v2, 2.0)
+      val squaredDist = breezeSquaredDistance(v1.toBreeze, v2.toBreeze)
       val fastSquaredDist1 = fastSquaredDistance(v1, norm1, v2, norm2, precision)
       assert((fastSquaredDist1 - squaredDist) <= precision * squaredDist, s"failed with m = $m")
-      val fastSquaredDist2 = fastSquaredDistance(v1, norm1, v2.toDenseVector, norm2, precision)
+      val fastSquaredDist2 =
+        fastSquaredDistance(v1, norm1, Vectors.dense(v2.toArray), norm2, precision)
       assert((fastSquaredDist2 - squaredDist) <= precision * squaredDist, s"failed with m = $m")
     }
   }
diff --git a/project/SparkBuild.scala b/project/SparkBuild.scala
index b16ed66aeb3c3..6ff08723772aa 100644
--- a/project/SparkBuild.scala
+++ b/project/SparkBuild.scala
@@ -365,7 +365,10 @@ object Unidoc {
         "mllib.classification", "mllib.clustering", "mllib.evaluation.binary", "mllib.linalg",
         "mllib.linalg.distributed", "mllib.optimization", "mllib.rdd", "mllib.recommendation",
         "mllib.regression", "mllib.stat", "mllib.tree", "mllib.tree.configuration",
-        "mllib.tree.impurity", "mllib.tree.model", "mllib.util"
+        "mllib.tree.impurity", "mllib.tree.model", "mllib.util",
+        "mllib.evaluation", "mllib.feature", "mllib.random", "mllib.stat.correlation",
+        "mllib.stat.test", "mllib.tree.impl", "mllib.tree.loss",
+        "ml", "ml.classification", "ml.evaluation", "ml.feature", "ml.param", "ml.tuning"
       ),
       "-group", "Spark SQL", packageList("sql.api.java", "sql.api.java.types", "sql.hive.api.java"),
       "-noqualifier", "java.lang"
diff --git a/python/pyspark/mllib/tree.py b/python/pyspark/mllib/tree.py
index 46e253991aa56..66702478474dc 100644
--- a/python/pyspark/mllib/tree.py
+++ b/python/pyspark/mllib/tree.py
@@ -250,7 +250,7 @@ def _train(cls, data, algo, numClasses, categoricalFeaturesInfo, numTrees,
         return RandomForestModel(model)
 
     @classmethod
-    def trainClassifier(cls, data, numClassesForClassification, categoricalFeaturesInfo, numTrees,
+    def trainClassifier(cls, data, numClasses, categoricalFeaturesInfo, numTrees,
                         featureSubsetStrategy="auto", impurity="gini", maxDepth=4, maxBins=32,
                         seed=None):
         """
@@ -259,7 +259,7 @@ def trainClassifier(cls, data, numClassesForClassification, categoricalFeaturesI
 
         :param data: Training dataset: RDD of LabeledPoint. Labels should take
                values {0, 1, ..., numClasses-1}.
-        :param numClassesForClassification: number of classes for classification.
+        :param numClasses: number of classes for classification.
         :param categoricalFeaturesInfo: Map storing arity of categorical features.
                E.g., an entry (n -> k) indicates that feature n is categorical
                with k categories indexed from 0: {0, 1, ..., k-1}.
@@ -320,7 +320,7 @@ def trainClassifier(cls, data, numClassesForClassification, categoricalFeaturesI
         >>> model.predict(rdd).collect()
         [1.0, 0.0]
         """
-        return cls._train(data, "classification", numClassesForClassification,
+        return cls._train(data, "classification", numClasses,
                           categoricalFeaturesInfo, numTrees, featureSubsetStrategy, impurity,
                           maxDepth, maxBins, seed)
 
diff --git a/sbt/sbt b/sbt/sbt
index c172fa74bc771..0a251d97db95c 100755
--- a/sbt/sbt
+++ b/sbt/sbt
@@ -71,8 +71,8 @@ Usage: $script_name [options]
   -Dkey=val          pass -Dkey=val directly to the java runtime
   -J-X               pass option -X directly to the java runtime
                      (-J is stripped)
-  -S-X               add -X to sbt's scalacOptions (-J is stripped)
-  -PmavenProfiles     Enable a maven profile for the build.
+  -S-X               add -X to sbt's scalacOptions (-S is stripped)
+  -PmavenProfiles    Enable a maven profile for the build.
 
 In the case of duplicated or conflicting options, the order above
 shows precedence: JAVA_OPTS lowest, command line options highest.
diff --git a/sbt/sbt-launch-lib.bash b/sbt/sbt-launch-lib.bash
index 84a6f7a207186..fa7222d55a6db 100755
--- a/sbt/sbt-launch-lib.bash
+++ b/sbt/sbt-launch-lib.bash
@@ -104,7 +104,7 @@ addResidual () {
   residual_args=( "${residual_args[@]}" "$1" )
 }
 addDebugger () {
-  addJava "-Xdebug -Xrunjdwp:transport=dt_socket,server=y,suspend=n,address=$1"
+  addJava "-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=$1"
 }
 
 # a ham-fisted attempt to move some memory settings in concert
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SparkSQLParser.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SparkSQLParser.scala
index b198ed9936d95..f1a1ca6616a21 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SparkSQLParser.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SparkSQLParser.scala
@@ -97,10 +97,10 @@ class SqlLexical(val keywords: Seq[String]) extends StdLexical {
 
   /** Generate all variations of upper and lower case of a given string */
   def allCaseVersions(s: String, prefix: String = ""): Stream[String] = {
-    if (s == "") {
+    if (s.isEmpty) {
       Stream(prefix)
     } else {
-      allCaseVersions(s.tail, prefix + s.head.toLower) ++
+      allCaseVersions(s.tail, prefix + s.head.toLower) #:::
         allCaseVersions(s.tail, prefix + s.head.toUpper)
     }
   }
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala
index dc1d349f10f1b..a2bcd73b6074f 100755
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/SqlParser.scala
@@ -277,7 +277,8 @@ class SqlParser extends AbstractSparkSQLParser {
     | SUM   ~> "(" ~> DISTINCT ~> expression <~ ")" ^^ { case exp => SumDistinct(exp) }
     | COUNT ~  "(" ~> "*"                    <~ ")" ^^ { case _ => Count(Literal(1)) }
     | COUNT ~  "(" ~> expression             <~ ")" ^^ { case exp => Count(exp) }
-    | COUNT ~> "(" ~> DISTINCT ~> expression <~ ")" ^^ { case exp => CountDistinct(exp :: Nil) }
+    | COUNT ~> "(" ~> DISTINCT ~> repsep(expression, ",") <~ ")" ^^
+      { case exps => CountDistinct(exps) }
     | APPROXIMATE ~ COUNT ~ "(" ~ DISTINCT ~> expression <~ ")" ^^
       { case exp => ApproxCountDistinct(exp) }
     | APPROXIMATE ~> "(" ~> floatLit ~ ")" ~ COUNT ~ "(" ~ DISTINCT ~ expression <~ ")" ^^
@@ -340,18 +341,13 @@ class SqlParser extends AbstractSparkSQLParser {
     | floatLit ^^ { f => Literal(f.toDouble) }
     )
 
-  private val longMax = BigDecimal(s"${Long.MaxValue}")
-  private val longMin = BigDecimal(s"${Long.MinValue}")
-  private val intMax = BigDecimal(s"${Int.MaxValue}")
-  private val intMin = BigDecimal(s"${Int.MinValue}")
-
   private def toNarrowestIntegerType(value: String) = {
     val bigIntValue = BigDecimal(value)
 
     bigIntValue match {
-      case v if v < longMin || v > longMax => v
-      case v if v < intMin || v > intMax => v.toLong
-      case v => v.toInt
+      case v if bigIntValue.isValidInt => v.toIntExact
+      case v if bigIntValue.isValidLong => v.toLongExact
+      case v => v
     }
   }
 
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala
index 31dc5a58e68e5..70dabc4e6c2e9 100755
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/dsl/package.scala
@@ -147,6 +147,8 @@ package object dsl {
     def max(e: Expression) = Max(e)
     def upper(e: Expression) = Upper(e)
     def lower(e: Expression) = Lower(e)
+    def sqrt(e: Expression) = Sqrt(e)
+    def abs(e: Expression) = Abs(e)
 
     implicit class DslSymbol(sym: Symbol) extends ImplicitAttribute { def s = sym.name }
     // TODO more implicit class for literal?
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
index 39b120e8de485..bc45881e42748 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Expression.scala
@@ -153,6 +153,25 @@ abstract class Expression extends TreeNode[Expression] {
     }
   }
 
+  /**
+   * Evaluation helper function for 1 Fractional children expression.
+   * if the expression result is null, the evaluation result should be null.
+   */
+  @inline
+  protected final def f1(i: Row, e1: Expression, f: ((Fractional[Any], Any) => Any)): Any  = {
+    val evalE1 = e1.eval(i: Row)
+    if(evalE1 == null) {
+      null
+    } else {
+      e1.dataType match {
+        case ft: FractionalType =>
+          f.asInstanceOf[(Fractional[ft.JvmType], ft.JvmType) => ft.JvmType](
+            ft.fractional, evalE1.asInstanceOf[ft.JvmType])
+        case other => sys.error(s"Type $other does not support fractional operations")
+      }
+    }
+  }
+
   /**
    * Evaluation helper function for 2 Integral children expressions. Those expressions are
    * supposed to be in the same data type, and also the return type.
@@ -189,6 +208,28 @@ abstract class Expression extends TreeNode[Expression] {
     }
   }
 
+  /**
+   * Evaluation helper function for 1 Integral children expression.
+   * if the expression result is null, the evaluation result should be null.
+   */
+  @inline
+  protected final def i1(i: Row, e1: Expression, f: ((Integral[Any], Any) => Any)): Any  = {
+    val evalE1 = e1.eval(i)
+    if(evalE1 == null) {
+      null
+    } else {
+      e1.dataType match {
+        case i: IntegralType =>
+          f.asInstanceOf[(Integral[i.JvmType], i.JvmType) => i.JvmType](
+            i.integral, evalE1.asInstanceOf[i.JvmType])
+        case i: FractionalType =>
+          f.asInstanceOf[(Integral[i.JvmType], i.JvmType) => i.JvmType](
+            i.asIntegral, evalE1.asInstanceOf[i.JvmType])
+        case other => sys.error(s"Type $other does not support numeric operations")
+      }
+    }
+  }
+
   /**
    * Evaluation helper function for 2 Comparable children expressions. Those expressions are
    * supposed to be in the same data type, and the return type should be Integer:
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/arithmetic.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/arithmetic.scala
index d17c9553ac24e..61c26c50a6662 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/arithmetic.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/arithmetic.scala
@@ -19,7 +19,6 @@ package org.apache.spark.sql.catalyst.expressions
 
 import org.apache.spark.sql.catalyst.analysis.UnresolvedException
 import org.apache.spark.sql.catalyst.types._
-import scala.math.pow
 
 case class UnaryMinus(child: Expression) extends UnaryExpression {
   type EvaluatedType = Any
@@ -43,7 +42,7 @@ case class Sqrt(child: Expression) extends UnaryExpression {
   override def toString = s"SQRT($child)"
 
   override def eval(input: Row): Any = {
-    n1(child, input, ((na,a) => math.sqrt(na.toDouble(a))))
+    n1(child, input, (na,a) => math.sqrt(na.toDouble(a)))
   }
 }
 
@@ -106,11 +105,16 @@ case class Multiply(left: Expression, right: Expression) extends BinaryArithmeti
 case class Divide(left: Expression, right: Expression) extends BinaryArithmetic {
   def symbol = "/"
 
-  override def nullable = left.nullable || right.nullable || dataType.isInstanceOf[DecimalType]
+  override def nullable = true
 
-  override def eval(input: Row): Any = dataType match {
-    case _: FractionalType => f2(input, left, right, _.div(_, _))
-    case _: IntegralType => i2(input, left , right, _.quot(_, _))
+  override def eval(input: Row): Any = {
+    val evalE2 = right.eval(input)
+    dataType match {
+      case _ if evalE2 == null => null
+      case _ if evalE2 == 0 => null
+      case ft: FractionalType => f1(input, left, _.div(_, evalE2.asInstanceOf[ft.JvmType]))
+      case it: IntegralType => i1(input, left, _.quot(_, evalE2.asInstanceOf[it.JvmType]))
+    }
   }
 
 }
@@ -134,7 +138,7 @@ case class BitwiseAnd(left: Expression, right: Expression) extends BinaryArithme
     case ShortType => (evalE1.asInstanceOf[Short] & evalE2.asInstanceOf[Short]).toShort
     case IntegerType => evalE1.asInstanceOf[Int] & evalE2.asInstanceOf[Int]
     case LongType => evalE1.asInstanceOf[Long] & evalE2.asInstanceOf[Long]
-    case other => sys.error(s"Unsupported bitwise & operation on ${other}")
+    case other => sys.error(s"Unsupported bitwise & operation on $other")
   }
 }
 
@@ -149,7 +153,7 @@ case class BitwiseOr(left: Expression, right: Expression) extends BinaryArithmet
     case ShortType => (evalE1.asInstanceOf[Short] | evalE2.asInstanceOf[Short]).toShort
     case IntegerType => evalE1.asInstanceOf[Int] | evalE2.asInstanceOf[Int]
     case LongType => evalE1.asInstanceOf[Long] | evalE2.asInstanceOf[Long]
-    case other => sys.error(s"Unsupported bitwise | operation on ${other}")
+    case other => sys.error(s"Unsupported bitwise | operation on $other")
   }
 }
 
@@ -164,7 +168,7 @@ case class BitwiseXor(left: Expression, right: Expression) extends BinaryArithme
     case ShortType => (evalE1.asInstanceOf[Short] ^ evalE2.asInstanceOf[Short]).toShort
     case IntegerType => evalE1.asInstanceOf[Int] ^ evalE2.asInstanceOf[Int]
     case LongType => evalE1.asInstanceOf[Long] ^ evalE2.asInstanceOf[Long]
-    case other => sys.error(s"Unsupported bitwise ^ operation on ${other}")
+    case other => sys.error(s"Unsupported bitwise ^ operation on $other")
   }
 }
 
@@ -177,7 +181,7 @@ case class BitwiseNot(child: Expression) extends UnaryExpression {
   def dataType = child.dataType
   override def foldable = child.foldable
   def nullable = child.nullable
-  override def toString = s"-$child"
+  override def toString = s"~$child"
 
   override def eval(input: Row): Any = {
     val evalE = child.eval(input)
@@ -185,11 +189,11 @@ case class BitwiseNot(child: Expression) extends UnaryExpression {
       null
     } else {
       dataType match {
-        case ByteType => (~(evalE.asInstanceOf[Byte])).toByte
-        case ShortType => (~(evalE.asInstanceOf[Short])).toShort
-        case IntegerType => ~(evalE.asInstanceOf[Int])
-        case LongType => ~(evalE.asInstanceOf[Long])
-        case other => sys.error(s"Unsupported bitwise ~ operation on ${other}")
+        case ByteType => (~evalE.asInstanceOf[Byte]).toByte
+        case ShortType => (~evalE.asInstanceOf[Short]).toShort
+        case IntegerType => ~evalE.asInstanceOf[Int]
+        case LongType => ~evalE.asInstanceOf[Long]
+        case other => sys.error(s"Unsupported bitwise ~ operation on $other")
       }
     }
   }
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
index 67f8d411b6bb4..ab71e15e1f573 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
@@ -359,7 +359,24 @@ abstract class CodeGenerator[InType <: AnyRef, OutType <: AnyRef] extends Loggin
       case Add(e1, e2) =>      (e1, e2) evaluate { case (eval1, eval2) => q"$eval1 + $eval2" }
       case Subtract(e1, e2) => (e1, e2) evaluate { case (eval1, eval2) => q"$eval1 - $eval2" }
       case Multiply(e1, e2) => (e1, e2) evaluate { case (eval1, eval2) => q"$eval1 * $eval2" }
-      case Divide(e1, e2) =>   (e1, e2) evaluate { case (eval1, eval2) => q"$eval1 / $eval2" }
+      case Divide(e1, e2) =>
+        val eval1 = expressionEvaluator(e1)
+        val eval2 = expressionEvaluator(e2)
+
+        eval1.code ++ eval2.code ++
+        q"""
+          var $nullTerm = false
+          var $primitiveTerm: ${termForType(e1.dataType)} = 0
+
+          if (${eval1.nullTerm} || ${eval2.nullTerm} ) {
+            $nullTerm = true
+          } else if (${eval2.primitiveTerm} == 0)
+            $nullTerm = true
+          else {
+            $nullTerm = false
+            $primitiveTerm = ${eval1.primitiveTerm} / ${eval2.primitiveTerm}
+          }
+         """.children
 
       case IsNotNull(e) =>
         val eval = expressionEvaluator(e)
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/nullFunctions.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/nullFunctions.scala
index 84a3567895175..08b982bc671e7 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/nullFunctions.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/nullFunctions.scala
@@ -45,9 +45,9 @@ case class Coalesce(children: Seq[Expression]) extends Expression {
   override def eval(input: Row): Any = {
     var i = 0
     var result: Any = null
-    while(i < children.size && result == null) {
-      result = children(i).eval(input)
-      i += 1
+    val childIterator = children.iterator
+    while (childIterator.hasNext && result == null) {
+      result = childIterator.next().eval(input)
     }
     result
   }
diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala
index 3f5b9f698f827..cd2f67f448b0b 100644
--- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala
+++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/ExpressionEvaluationSuite.scala
@@ -42,6 +42,21 @@ class ExpressionEvaluationSuite extends FunSuite {
     checkEvaluation(Literal(1) + Literal(1), 2)
   }
 
+  test("unary BitwiseNOT") {
+    checkEvaluation(BitwiseNot(1), -2)
+    assert(BitwiseNot(1).dataType === IntegerType)
+    assert(BitwiseNot(1).eval(EmptyRow).isInstanceOf[Int])
+    checkEvaluation(BitwiseNot(1.toLong), -2.toLong)
+    assert(BitwiseNot(1.toLong).dataType === LongType)
+    assert(BitwiseNot(1.toLong).eval(EmptyRow).isInstanceOf[Long])
+    checkEvaluation(BitwiseNot(1.toShort), -2.toShort)
+    assert(BitwiseNot(1.toShort).dataType === ShortType)
+    assert(BitwiseNot(1.toShort).eval(EmptyRow).isInstanceOf[Short])
+    checkEvaluation(BitwiseNot(1.toByte), -2.toByte)
+    assert(BitwiseNot(1.toByte).dataType === ByteType)
+    assert(BitwiseNot(1.toByte).eval(EmptyRow).isInstanceOf[Byte])
+  }
+
   /**
    * Checks for three-valued-logic.  Based on:
    * http://en.wikipedia.org/wiki/Null_(SQL)#Comparisons_with_NULL_and_the_three-valued_logic_.283VL.29
@@ -149,6 +164,21 @@ class ExpressionEvaluationSuite extends FunSuite {
     checkEvaluation(In(Literal(1), Seq(Literal(1), Literal(2))) && In(Literal(2), Seq(Literal(1), Literal(2))), true)
   }
 
+  test("Divide") {
+    checkEvaluation(Divide(Literal(2), Literal(1)), 2)
+    checkEvaluation(Divide(Literal(1.0), Literal(2.0)), 0.5)
+    checkEvaluation(Divide(Literal(1), Literal(2)), 0)
+    checkEvaluation(Divide(Literal(1), Literal(0)), null)
+    checkEvaluation(Divide(Literal(1.0), Literal(0.0)), null)
+    checkEvaluation(Divide(Literal(0.0), Literal(0.0)), null)
+    checkEvaluation(Divide(Literal(0), Literal(null, IntegerType)), null)
+    checkEvaluation(Divide(Literal(1), Literal(null, IntegerType)), null)
+    checkEvaluation(Divide(Literal(null, IntegerType), Literal(0)), null)
+    checkEvaluation(Divide(Literal(null, DoubleType), Literal(0.0)), null)
+    checkEvaluation(Divide(Literal(null, IntegerType), Literal(1)), null)
+    checkEvaluation(Divide(Literal(null, IntegerType), Literal(null, IntegerType)), null)
+  }
+
   test("INSET") {
     val hS = HashSet[Any]() + 1 + 2
     val nS = HashSet[Any]() + 1 + 2 + null
diff --git a/sql/core/src/main/java/org/apache/spark/sql/api/java/DataType.java b/sql/core/src/main/java/org/apache/spark/sql/api/java/DataType.java
index c38354039d686..c69bbd5736a5b 100644
--- a/sql/core/src/main/java/org/apache/spark/sql/api/java/DataType.java
+++ b/sql/core/src/main/java/org/apache/spark/sql/api/java/DataType.java
@@ -82,6 +82,11 @@ public abstract class DataType {
    */
   public static final ShortType ShortType = new ShortType();
 
+  /**
+   * Gets the NullType object.
+   */
+  public static final NullType NullType = new NullType();
+
   /**
    * Creates an ArrayType by specifying the data type of elements ({@code elementType}).
    * The field of {@code containsNull} is set to {@code true}.
diff --git a/core/src/main/scala/org/apache/spark/rdd/MappedRDD.scala b/sql/core/src/main/java/org/apache/spark/sql/api/java/NullType.java
similarity index 66%
rename from core/src/main/scala/org/apache/spark/rdd/MappedRDD.scala
rename to sql/core/src/main/java/org/apache/spark/sql/api/java/NullType.java
index 8d7c288593665..6d5ecdf46e551 100644
--- a/core/src/main/scala/org/apache/spark/rdd/MappedRDD.scala
+++ b/sql/core/src/main/java/org/apache/spark/sql/api/java/NullType.java
@@ -15,18 +15,13 @@
  * limitations under the License.
  */
 
-package org.apache.spark.rdd
+package org.apache.spark.sql.api.java;
 
-import scala.reflect.ClassTag
-
-import org.apache.spark.{Partition, TaskContext}
-
-private[spark]
-class MappedRDD[U: ClassTag, T: ClassTag](prev: RDD[T], f: T => U)
-  extends RDD[U](prev) {
-
-  override def getPartitions: Array[Partition] = firstParent[T].partitions
-
-  override def compute(split: Partition, context: TaskContext) =
-    firstParent[T].iterator(split, context).map(f)
+/**
+ * The data type representing null and NULL values.
+ *
+ * {@code NullType} is represented by the singleton object {@link DataType#NullType}.
+ */
+public class NullType extends DataType {
+  protected NullType() {}
 }
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/SchemaRDD.scala b/sql/core/src/main/scala/org/apache/spark/sql/SchemaRDD.scala
index 452baab8eb889..c6d4dabf83bc4 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/SchemaRDD.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/SchemaRDD.scala
@@ -225,6 +225,8 @@ class SchemaRDD(
    * {{{
    *   schemaRDD.limit(10)
    * }}}
+   * 
+   * @group Query
    */
   def limit(limitNum: Int): SchemaRDD =
     new SchemaRDD(sqlContext, Limit(Literal(limitNum), logicalPlan))
@@ -355,6 +357,8 @@ class SchemaRDD(
    * Return the number of elements in the RDD. Unlike the base RDD implementation of count, this
    * implementation leverages the query optimizer to compute the count on the SchemaRDD, which
    * supports features such as filter pushdown.
+   * 
+   * @group Query
    */
   @Experimental
   override def count(): Long = aggregate(Count(Literal(1))).collect().head.getLong(0)
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/package.scala b/sql/core/src/main/scala/org/apache/spark/sql/package.scala
index 51dad54f1a3f3..1fd8e6220f83b 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/package.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/package.scala
@@ -263,6 +263,16 @@ package object sql {
   @DeveloperApi
   val ShortType = catalyst.types.ShortType
 
+  /**
+   * :: DeveloperApi ::
+   *
+   * The data type representing `NULL` values.
+   *
+   * @group dataType
+   */
+  @DeveloperApi
+  val NullType = catalyst.types.NullType
+  
   /**
    * :: DeveloperApi ::
    *
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/parquet/ParquetTableOperations.scala b/sql/core/src/main/scala/org/apache/spark/sql/parquet/ParquetTableOperations.scala
index 0e36852ddd9b0..232ef90b017a7 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/parquet/ParquetTableOperations.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/parquet/ParquetTableOperations.scala
@@ -302,11 +302,14 @@ case class InsertIntoParquetTable(
       val committer = format.getOutputCommitter(hadoopContext)
       committer.setupTask(hadoopContext)
       val writer = format.getRecordWriter(hadoopContext)
-      while (iter.hasNext) {
-        val row = iter.next()
-        writer.write(null, row)
+      try {
+        while (iter.hasNext) {
+          val row = iter.next()
+          writer.write(null, row)
+        }
+      } finally {
+        writer.close(hadoopContext)
       }
-      writer.close(hadoopContext)
       committer.commitTask(hadoopContext)
       1
     }
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala b/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala
index bea12e6dd674e..2e0c6c51c00e5 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala
@@ -22,6 +22,7 @@ import org.apache.hadoop.fs.{FileStatus, FileSystem, Path}
 import org.apache.hadoop.conf.{Configurable, Configuration}
 import org.apache.hadoop.io.Writable
 import org.apache.hadoop.mapreduce.{JobContext, InputSplit, Job}
+import org.apache.spark.sql.catalyst.expressions.codegen.GeneratePredicate
 
 import parquet.hadoop.ParquetInputFormat
 import parquet.hadoop.util.ContextUtil
@@ -31,16 +32,16 @@ import org.apache.spark.{Partition => SparkPartition, Logging}
 import org.apache.spark.rdd.{NewHadoopPartition, RDD}
 
 import org.apache.spark.sql.{SQLConf, Row, SQLContext}
-import org.apache.spark.sql.catalyst.expressions.{SpecificMutableRow, And, Expression, Attribute}
-import org.apache.spark.sql.catalyst.types.{IntegerType, StructField, StructType}
+import org.apache.spark.sql.catalyst.expressions._
+import org.apache.spark.sql.catalyst.types.{StringType, IntegerType, StructField, StructType}
 import org.apache.spark.sql.sources._
 
 import scala.collection.JavaConversions._
 
 /**
  * Allows creation of parquet based tables using the syntax
- * `CREATE TABLE ... USING org.apache.spark.sql.parquet`.  Currently the only option required
- * is `path`, which should be the location of a collection of, optionally partitioned,
+ * `CREATE TEMPORARY TABLE ... USING org.apache.spark.sql.parquet`.  Currently the only option 
+ * required is `path`, which should be the location of a collection of, optionally partitioned,
  * parquet files.
  */
 class DefaultSource extends RelationProvider {
@@ -49,7 +50,7 @@ class DefaultSource extends RelationProvider {
       sqlContext: SQLContext,
       parameters: Map[String, String]): BaseRelation = {
     val path =
-      parameters.getOrElse("path", sys.error("'path' must be specifed for parquet tables."))
+      parameters.getOrElse("path", sys.error("'path' must be specified for parquet tables."))
 
     ParquetRelation2(path)(sqlContext)
   }
@@ -151,8 +152,6 @@ case class ParquetRelation2(path: String)(@transient val sqlContext: SQLContext)
   override def buildScan(output: Seq[Attribute], predicates: Seq[Expression]): RDD[Row] = {
     // This is mostly a hack so that we can use the existing parquet filter code.
     val requiredColumns = output.map(_.name)
-    // TODO: Parquet filters should be based on data sources API, not catalyst expressions.
-    val filters = DataSourceStrategy.selectFilters(predicates)
 
     val job = new Job(sparkContext.hadoopConfiguration)
     ParquetInputFormat.setReadSupportClass(job, classOf[RowReadSupport])
@@ -160,38 +159,40 @@ case class ParquetRelation2(path: String)(@transient val sqlContext: SQLContext)
 
     val requestedSchema = StructType(requiredColumns.map(schema(_)))
 
-    // TODO: Make folder based partitioning a first class citizen of the Data Sources API.
-    val partitionFilters = filters.collect {
-      case e @ EqualTo(attr, value) if partitionKeys.contains(attr) =>
-        logInfo(s"Parquet scan partition filter: $attr=$value")
-        (p: Partition) => p.partitionValues(attr) == value
-
-      case e @ In(attr, values) if partitionKeys.contains(attr) =>
-        logInfo(s"Parquet scan partition filter: $attr IN ${values.mkString("{", ",", "}")}")
-        val set = values.toSet
-        (p: Partition) => set.contains(p.partitionValues(attr))
-
-      case e @ GreaterThan(attr, value) if partitionKeys.contains(attr) =>
-        logInfo(s"Parquet scan partition filter: $attr > $value")
-        (p: Partition) => p.partitionValues(attr).asInstanceOf[Int] > value.asInstanceOf[Int]
-
-      case e @ GreaterThanOrEqual(attr, value) if partitionKeys.contains(attr) =>
-        logInfo(s"Parquet scan partition filter: $attr >= $value")
-        (p: Partition) => p.partitionValues(attr).asInstanceOf[Int] >= value.asInstanceOf[Int]
+    val partitionKeySet = partitionKeys.toSet
+    val rawPredicate =
+      predicates
+        .filter(_.references.map(_.name).toSet.subsetOf(partitionKeySet))
+        .reduceOption(And)
+        .getOrElse(Literal(true))
+
+    // Translate the predicate so that it reads from the information derived from the
+    // folder structure
+    val castedPredicate = rawPredicate transform {
+      case a: AttributeReference =>
+        val idx = partitionKeys.indexWhere(a.name == _)
+        BoundReference(idx, IntegerType, nullable = true)
+    }
 
-      case e @ LessThan(attr, value) if partitionKeys.contains(attr) =>
-        logInfo(s"Parquet scan partition filter: $attr < $value")
-        (p: Partition) => p.partitionValues(attr).asInstanceOf[Int] < value.asInstanceOf[Int]
+    val inputData = new GenericMutableRow(partitionKeys.size)
+    val pruningCondition = InterpretedPredicate(castedPredicate)
 
-      case e @ LessThanOrEqual(attr, value) if partitionKeys.contains(attr) =>
-        logInfo(s"Parquet scan partition filter: $attr <= $value")
-        (p: Partition) => p.partitionValues(attr).asInstanceOf[Int] <= value.asInstanceOf[Int]
-    }
+    val selectedPartitions =
+      if (partitionKeys.nonEmpty && predicates.nonEmpty) {
+        partitions.filter { part =>
+          inputData(0) = part.partitionValues.values.head
+          pruningCondition(inputData)
+        }
+      } else {
+        partitions
+      }
 
-    val selectedPartitions = partitions.filter(p => partitionFilters.forall(_(p)))
     val fs = FileSystem.get(new java.net.URI(path), sparkContext.hadoopConfiguration)
     val selectedFiles = selectedPartitions.flatMap(_.files).map(f => fs.makeQualified(f.getPath))
-    org.apache.hadoop.mapreduce.lib.input.FileInputFormat.setInputPaths(job, selectedFiles:_*)
+    // FileInputFormat cannot handle empty lists.
+    if (selectedFiles.nonEmpty) {
+      org.apache.hadoop.mapreduce.lib.input.FileInputFormat.setInputPaths(job, selectedFiles: _*)
+    }
 
     // Push down filters when possible
     predicates
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala
index 9168ca2fc6fec..ca510cb0b07e3 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/ddl.scala
@@ -67,7 +67,7 @@ private[sql] class DDLParser extends StandardTokenParsers with PackratParsers wi
   protected lazy val ddl: Parser[LogicalPlan] = createTable
 
   /**
-   * CREATE FOREIGN TEMPORARY TABLE avroTable
+   * CREATE TEMPORARY TABLE avroTable
    * USING org.apache.spark.sql.avro
    * OPTIONS (path "../hive/src/test/resources/data/files/episodes.avro")
    */
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
index 2b8fc05fc0102..939b4e15163a6 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/sources/interfaces.scala
@@ -1,19 +1,19 @@
 /*
-* Licensed to the Apache Software Foundation (ASF) under one or more
-* contributor license agreements.  See the NOTICE file distributed with
-* this work for additional information regarding copyright ownership.
-* The ASF licenses this file to You under the Apache License, Version 2.0
-* (the "License"); you may not use this file except in compliance with
-* the License.  You may obtain a copy of the License at
-*
-*    http://www.apache.org/licenses/LICENSE-2.0
-*
-* Unless required by applicable law or agreed to in writing, software
-* distributed under the License is distributed on an "AS IS" BASIS,
-* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-* See the License for the specific language governing permissions and
-* limitations under the License.
-*/
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
 package org.apache.spark.sql.sources
 
 import org.apache.spark.annotation.{Experimental, DeveloperApi}
@@ -111,5 +111,3 @@ abstract class PrunedFilteredScan extends BaseRelation {
 abstract class CatalystScan extends BaseRelation {
   def buildScan(requiredColumns: Seq[Attribute], filters: Seq[Expression]): RDD[Row]
 }
-
-
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/types/util/DataTypeConversions.scala b/sql/core/src/main/scala/org/apache/spark/sql/types/util/DataTypeConversions.scala
index d4258156f18f6..4160a80621c77 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/types/util/DataTypeConversions.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/types/util/DataTypeConversions.scala
@@ -62,6 +62,7 @@ protected[sql] object DataTypeConversions {
     case IntegerType => JDataType.IntegerType
     case LongType => JDataType.LongType
     case ShortType => JDataType.ShortType
+    case NullType => JDataType.NullType
 
     case arrayType: ArrayType => JDataType.createArrayType(
         asJavaDataType(arrayType.elementType), arrayType.containsNull)
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/DslQuerySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/DslQuerySuite.scala
index 94bd97758fe94..1a330a2bb6d46 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/DslQuerySuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/DslQuerySuite.scala
@@ -282,4 +282,72 @@ class DslQuerySuite extends QueryTest {
       (1, "1", "11") :: (2, "2", "22") :: (3, "3", "33") :: Nil
     )
   }
+
+  test("sqrt") {
+    checkAnswer(
+      testData.select(sqrt('key)).orderBy('key asc),
+      (1 to 100).map(n => Seq(math.sqrt(n)))
+    )
+
+    checkAnswer(
+      testData.select(sqrt('value), 'key).orderBy('key asc, 'value asc),
+      (1 to 100).map(n => Seq(math.sqrt(n), n))
+    )
+
+    checkAnswer(
+      testData.select(sqrt(Literal(null))),
+      (1 to 100).map(_ => Seq(null))
+    )
+  }
+
+  test("abs") {
+    checkAnswer(
+      testData.select(abs('key)).orderBy('key asc),
+      (1 to 100).map(n => Seq(n))
+    )
+
+    checkAnswer(
+      negativeData.select(abs('key)).orderBy('key desc),
+      (1 to 100).map(n => Seq(n))
+    )
+
+    checkAnswer(
+      testData.select(abs(Literal(null))),
+      (1 to 100).map(_ => Seq(null))
+    )
+  }
+
+  test("upper") {
+    checkAnswer(
+      lowerCaseData.select(upper('l)),
+      ('a' to 'd').map(c => Seq(c.toString.toUpperCase()))
+    )
+
+    checkAnswer(
+      testData.select(upper('value), 'key),
+      (1 to 100).map(n => Seq(n.toString, n))
+    )
+
+    checkAnswer(
+      testData.select(upper(Literal(null))),
+      (1 to 100).map(n => Seq(null))
+    )
+  }
+
+  test("lower") {
+    checkAnswer(
+      upperCaseData.select(lower('L)),
+      ('A' to 'F').map(c => Seq(c.toString.toLowerCase()))
+    )
+
+    checkAnswer(
+      testData.select(lower('value), 'key),
+      (1 to 100).map(n => Seq(n.toString, n))
+    )
+
+    checkAnswer(
+      testData.select(lower(Literal(null))),
+      (1 to 100).map(n => Seq(null))
+    )
+  }
 }
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala
index 84ee3051eb682..bcebce7603f13 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala
@@ -22,7 +22,6 @@ import java.util.TimeZone
 import org.scalatest.BeforeAndAfterAll
 
 import org.apache.spark.sql.catalyst.errors.TreeNodeException
-import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
 
 /* Implicits */
@@ -992,4 +991,11 @@ class SQLQuerySuite extends QueryTest with BeforeAndAfterAll {
       "nulldata2 on nulldata1.value <=> nulldata2.value"),
         (1 to 2).map(i => Seq(i)))
   }
+
+  test("Multi-column COUNT(DISTINCT ...)") {
+    val data = TestData(1,"val_1") :: TestData(2,"val_2") :: Nil
+    val rdd = sparkContext.parallelize((0 to 1).map(i => data(i)))
+    rdd.registerTempTable("distinctData")
+    checkAnswer(sql("SELECT COUNT(DISTINCT key,value) FROM distinctData"), 2)
+  }
 }
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/TestData.scala b/sql/core/src/test/scala/org/apache/spark/sql/TestData.scala
index 933e027436e75..bb553a0a1e50c 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/TestData.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/TestData.scala
@@ -32,6 +32,10 @@ object TestData {
     (1 to 100).map(i => TestData(i, i.toString))).toSchemaRDD
   testData.registerTempTable("testData")
 
+  val negativeData = TestSQLContext.sparkContext.parallelize(
+    (1 to 100).map(i => TestData(-i, (-i).toString))).toSchemaRDD
+  negativeData.registerTempTable("negativeData")
+
   case class LargeAndSmallInts(a: Int, b: Int)
   val largeAndSmallInts =
     TestSQLContext.sparkContext.parallelize(
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/api/java/JavaSQLSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/api/java/JavaSQLSuite.scala
index c9012c9e47cff..8afc3a9fb2187 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/api/java/JavaSQLSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/api/java/JavaSQLSuite.scala
@@ -68,6 +68,22 @@ class JavaSQLSuite extends FunSuite {
     javaSqlCtx.sql("SELECT * FROM people").collect()
   }
 
+  test("schema with null from JavaBeans") {
+    val person = new PersonBean
+    person.setName("Michael")
+    person.setAge(29)
+
+    val rdd = javaCtx.parallelize(person :: Nil)
+    val schemaRDD = javaSqlCtx.applySchema(rdd, classOf[PersonBean])
+
+    schemaRDD.registerTempTable("people")
+    val nullRDD = javaSqlCtx.sql("SELECT null FROM people")
+    val structFields = nullRDD.schema.getFields()
+    assert(structFields.size == 1)
+    assert(structFields(0).getDataType().isInstanceOf[NullType])
+    assert(nullRDD.collect.head.row === Seq(null))
+  }
+
   test("all types in JavaBeans") {
     val bean = new AllTypesBean
     bean.setStringField("")
diff --git a/sql/hive-thriftserver/src/main/scala/org/apache/spark/sql/hive/thriftserver/SparkSQLEnv.scala b/sql/hive-thriftserver/src/main/scala/org/apache/spark/sql/hive/thriftserver/SparkSQLEnv.scala
index 89732c939b0ec..158c225159720 100644
--- a/sql/hive-thriftserver/src/main/scala/org/apache/spark/sql/hive/thriftserver/SparkSQLEnv.scala
+++ b/sql/hive-thriftserver/src/main/scala/org/apache/spark/sql/hive/thriftserver/SparkSQLEnv.scala
@@ -32,11 +32,21 @@ private[hive] object SparkSQLEnv extends Logging {
 
   def init() {
     if (hiveContext == null) {
-      val sparkConf = new SparkConf()
+      val sparkConf = new SparkConf(loadDefaults = true)
+      val maybeSerializer = sparkConf.getOption("spark.serializer")
+      val maybeKryoReferenceTracking = sparkConf.getOption("spark.kryo.referenceTracking")
+
+      sparkConf
         .setAppName(s"SparkSQL::${java.net.InetAddress.getLocalHost.getHostName}")
         .set("spark.sql.hive.version", HiveShim.version)
-      sparkContext = new SparkContext(sparkConf)
+        .set(
+          "spark.serializer",
+          maybeSerializer.getOrElse("org.apache.spark.serializer.KryoSerializer"))
+        .set(
+          "spark.kryo.referenceTracking",
+          maybeKryoReferenceTracking.getOrElse("false"))
 
+      sparkContext = new SparkContext(sparkConf)
       sparkContext.addSparkListener(new StatsReportListener())
       hiveContext = new HiveContext(sparkContext)
 
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala
index 304b9a73ee91d..34fc21e61f60f 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveContext.scala
@@ -377,7 +377,7 @@ class HiveContext(sc: SparkContext) extends SQLContext(sc) {
         command.executeCollect().map(_.head.toString)
 
       case other =>
-        val result: Seq[Seq[Any]] = toRdd.map(_.copy()).collect().toSeq
+        val result: Seq[Seq[Any]] = other.executeCollect().toSeq
         // We need the types so we can output struct field names
         val types = analyzed.output.map(_.dataType)
         // Reformat to match hive tab delimited output.
@@ -416,6 +416,8 @@ object HiveContext {
     case (bin: Array[Byte], BinaryType) => new String(bin, "UTF-8")
     case (decimal: Decimal, DecimalType()) =>  // Hive strips trailing zeros so use its toString
       HiveShim.createDecimal(decimal.toBigDecimal.underlying()).toString
+    case (decimal: BigDecimal, DecimalType()) =>
+      HiveShim.createDecimal(decimal.underlying()).toString
     case (other, tpe) if primitiveTypes contains tpe => other.toString
   }
 
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala
index b9283f668a9b5..f4c42bbc5b03d 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveQl.scala
@@ -379,7 +379,7 @@ private[hive] object HiveQl {
   protected def nameExpressions(exprs: Seq[Expression]): Seq[NamedExpression] = {
     exprs.zipWithIndex.map {
       case (ne: NamedExpression, _) => ne
-      case (e, i) => Alias(e, s"c_$i")()
+      case (e, i) => Alias(e, s"_c$i")()
     }
   }
 
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveStrategies.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveStrategies.scala
index 56fc85239e1c0..edf291f917f07 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveStrategies.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveStrategies.scala
@@ -27,7 +27,7 @@ import org.apache.spark.sql.catalyst.planning._
 import org.apache.spark.sql.catalyst.plans._
 import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
 import org.apache.spark.sql.catalyst.types.StringType
-import org.apache.spark.sql.execution.{DescribeCommand, OutputFaker, SparkPlan}
+import org.apache.spark.sql.execution.{DescribeCommand, OutputFaker, SparkPlan, PhysicalRDD}
 import org.apache.spark.sql.hive
 import org.apache.spark.sql.hive.execution._
 import org.apache.spark.sql.parquet.ParquetRelation
@@ -104,53 +104,61 @@ private[hive] trait HiveStrategies {
           case a: AttributeReference => UnresolvedAttribute(a.name)
         })
 
-        if (relation.hiveQlTable.isPartitioned) {
-          val rawPredicate = pruningPredicates.reduceOption(And).getOrElse(Literal(true))
-          // Translate the predicate so that it automatically casts the input values to the correct
-          // data types during evaluation
-          val castedPredicate = rawPredicate transform {
-            case a: AttributeReference =>
-              val idx = relation.partitionKeys.indexWhere(a.exprId == _.exprId)
-              val key = relation.partitionKeys(idx)
-              Cast(BoundReference(idx, StringType, nullable = true), key.dataType)
-          }
-
-          val inputData = new GenericMutableRow(relation.partitionKeys.size)
-          val pruningCondition =
-            if(codegenEnabled) {
-              GeneratePredicate(castedPredicate)
-            } else {
-              InterpretedPredicate(castedPredicate)
+        try {
+          if (relation.hiveQlTable.isPartitioned) {
+            val rawPredicate = pruningPredicates.reduceOption(And).getOrElse(Literal(true))
+            // Translate the predicate so that it automatically casts the input values to the
+            // correct data types during evaluation.
+            val castedPredicate = rawPredicate transform {
+              case a: AttributeReference =>
+                val idx = relation.partitionKeys.indexWhere(a.exprId == _.exprId)
+                val key = relation.partitionKeys(idx)
+                Cast(BoundReference(idx, StringType, nullable = true), key.dataType)
             }
 
-          val partitions = relation.hiveQlPartitions.filter { part =>
-            val partitionValues = part.getValues
-            var i = 0
-            while (i < partitionValues.size()) {
-              inputData(i) = partitionValues(i)
-              i += 1
+            val inputData = new GenericMutableRow(relation.partitionKeys.size)
+            val pruningCondition =
+              if (codegenEnabled) {
+                GeneratePredicate(castedPredicate)
+              } else {
+                InterpretedPredicate(castedPredicate)
+              }
+
+            val partitions = relation.hiveQlPartitions.filter { part =>
+              val partitionValues = part.getValues
+              var i = 0
+              while (i < partitionValues.size()) {
+                inputData(i) = partitionValues(i)
+                i += 1
+              }
+              pruningCondition(inputData)
             }
-            pruningCondition(inputData)
-          }
 
-          hiveContext
-            .parquetFile(partitions.map(_.getLocation).mkString(","))
-            .addPartitioningAttributes(relation.partitionKeys)
-            .lowerCase
-            .where(unresolvedOtherPredicates)
-            .select(unresolvedProjection:_*)
-            .queryExecution
-            .executedPlan
-            .fakeOutput(projectList.map(_.toAttribute)):: Nil
-        } else {
-          hiveContext
-            .parquetFile(relation.hiveQlTable.getDataLocation.toString)
-            .lowerCase
-            .where(unresolvedOtherPredicates)
-            .select(unresolvedProjection:_*)
-            .queryExecution
-            .executedPlan
-            .fakeOutput(projectList.map(_.toAttribute)) :: Nil
+            hiveContext
+              .parquetFile(partitions.map(_.getLocation).mkString(","))
+              .addPartitioningAttributes(relation.partitionKeys)
+              .lowerCase
+              .where(unresolvedOtherPredicates)
+              .select(unresolvedProjection: _*)
+              .queryExecution
+              .executedPlan
+              .fakeOutput(projectList.map(_.toAttribute)) :: Nil
+          } else {
+            hiveContext
+              .parquetFile(relation.hiveQlTable.getDataLocation.toString)
+              .lowerCase
+              .where(unresolvedOtherPredicates)
+              .select(unresolvedProjection: _*)
+              .queryExecution
+              .executedPlan
+              .fakeOutput(projectList.map(_.toAttribute)) :: Nil
+          }
+        } catch {
+          // parquetFile will throw an exception when there is no data.
+          // TODO: Remove this hack for Spark 1.3.
+          case iae: java.lang.IllegalArgumentException
+              if iae.getMessage.contains("Can not create a Path from an empty string") =>
+            PhysicalRDD(plan.output, sparkContext.emptyRDD[Row]) :: Nil
         }
       case _ => Nil
     }
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/execution/HiveTableScan.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/execution/HiveTableScan.scala
index d39413a44a6cb..8bbcd6fec1f3b 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/execution/HiveTableScan.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/execution/HiveTableScan.scala
@@ -38,7 +38,7 @@ import org.apache.spark.sql.hive._
  * :: DeveloperApi ::
  * The Hive table scan operator.  Column and partition pruning are both handled.
  *
- * @param attributes Attributes to be fetched from the Hive table.
+ * @param requestedAttributes Attributes to be fetched from the Hive table.
  * @param relation The Hive table be be scanned.
  * @param partitionPruningPred An optional partition pruning predicate for partitioned table.
  */
diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveUdfs.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveUdfs.scala
index b255a2ebb9778..fecf8faaf4cda 100644
--- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveUdfs.scala
+++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/hiveUdfs.scala
@@ -279,7 +279,7 @@ private[hive] case class HiveGenericUdtf(
   }
 
   override protected def makeOutput() = {
-    // Use column names when given, otherwise c_1, c_2, ... c_n.
+    // Use column names when given, otherwise _c1, _c2, ... _cn.
     if (aliasNames.size == outputDataTypes.size) {
       aliasNames.zip(outputDataTypes).map {
         case (attrName, attrDataType) =>
@@ -288,7 +288,7 @@ private[hive] case class HiveGenericUdtf(
     } else {
       outputDataTypes.zipWithIndex.map {
         case (attrDataType, i) =>
-          AttributeReference(s"c_$i", attrDataType, nullable = true)()
+          AttributeReference(s"_c$i", attrDataType, nullable = true)()
       }
     }
   }
diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/parquet/parquetSuites.scala b/sql/hive/src/test/scala/org/apache/spark/sql/parquet/parquetSuites.scala
index 7159ebd0353ad..488ebba043794 100644
--- a/sql/hive/src/test/scala/org/apache/spark/sql/parquet/parquetSuites.scala
+++ b/sql/hive/src/test/scala/org/apache/spark/sql/parquet/parquetSuites.scala
@@ -218,6 +218,12 @@ abstract class ParquetTest extends QueryTest with BeforeAndAfterAll {
         10)
     }
 
+    test(s"non-existant partition $table") {
+      checkAnswer(
+        sql(s"SELECT COUNT(*) FROM $table WHERE p = 1000"),
+        0)
+    }
+
     test(s"multi-partition pruned count $table") {
       checkAnswer(
         sql(s"SELECT COUNT(*) FROM $table WHERE p IN (1,2,3)"),
diff --git a/yarn/common/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMasterArguments.scala b/yarn/common/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMasterArguments.scala
index 8b32c76d14037..d76a63276d752 100644
--- a/yarn/common/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMasterArguments.scala
+++ b/yarn/common/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMasterArguments.scala
@@ -36,7 +36,7 @@ class ApplicationMasterArguments(val args: Array[String]) {
 
     var args = inputArgs
 
-    while (! args.isEmpty) {
+    while (!args.isEmpty) {
       // --num-workers, --worker-memory, and --worker-cores are deprecated since 1.0,
       // the properties with executor in their names are preferred.
       args match {