在使用Spark ml时， VectorAssembler使用null异常

SparkException: Values to assemble cannot be null

代码为：

val Array(trainingData, testData) = dataset.randomSplit(Array(0.7,0.3))
val assembler = new VectorAssembler()
      .setInputCols(len_df.select("Length","Breadth").columns)
      .setOutputCol("features")
val data = assembler
      .transform(len_df)

当调用assembler时，报异常：

[Stage 151:==>                                                    (9 + 2) / 200]16/12/28 20:13:57 WARN scheduler.TaskSetManager: Lost task 31.0 in stage 151.0 (TID 8922, slave1.hadoop.ml): org.apache.spark.SparkException: Values to assemble cannot be null.
    at org.apache.spark.ml.feature.VectorAssembler$$anonfun$assemble$1.apply(VectorAssembler.scala:159)
    at org.apache.spark.ml.feature.VectorAssembler$$anonfun$assemble$1.apply(VectorAssembler.scala:142)
    at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
    at scala.collection.mutable.WrappedArray.foreach(WrappedArray.scala:35)
    at org.apache.spark.ml.feature.VectorAssembler$.assemble(VectorAssembler.scala:142)
    at org.apache.spark.ml.feature.VectorAssembler$$anonfun$3.apply(VectorAssembler.scala:98)
    at org.apache.spark.ml.feature.VectorAssembler$$anonfun$3.apply(VectorAssembler.scala:97)
    at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source)
    at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
    at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$8$$anon$1.hasNext(WholeStageCodegenExec.scala:370)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
    at scala.collection.Iterator$class.foreach(Iterator.scala:893)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1336)
    at scala.collection.TraversableOnce$class.foldLeft(TraversableOnce.scala:157)
    at scala.collection.AbstractIterator.foldLeft(Iterator.scala:1336)
    at scala.collection.TraversableOnce$class.aggregate(TraversableOnce.scala:214)

经查找资料，找到解决方法：

Spark < 2.4

There is nothing wrong with VectorAssembler. Spark Vector just cannot contain null values.

VectorAssembler没有问题，但是Spark Vector不支持null值存在，如果有null存在，则会报 Values to assemble cannot be null 异常

import org.apache.spark.ml.feature.VectorAssembler

val df = Seq(
  (Some(1.0), None), (None, Some(2.0)), (Some(3.0), Some(4.0))
).toDF("x1", "x2")

val assembler = new VectorAssembler()
  .setInputCols(df.columns).setOutputCol("features")

assembler.transform(df).show(3)

org.apache.spark.SparkException: Failed to execute user defined function($anonfun$3: (struct<x1:double,x2:double>) => vector)
...
Caused by: org.apache.spark.SparkException: Values to assemble cannot be null.

Null对于ML算法没有意义，并且不能使用scala.Double表示。

方式一：可以放弃

assembler.transform(df.na.drop).show(2)

+---+---+---------+
| x1| x2| features|
+---+---+---------+
|3.0|4.0|[3.0,4.0]|
+---+---+---------+

方式二： 填充/用平均值替换缺失值等

// For example with averages
val replacements: Map[String,Any] = Map("x1" -> 2.0, "x2" -> 3.0)
assembler.transform(df.na.fill(replacements)).show(3)

+---+---+---------+
| x1| x2| features|
+---+---+---------+
|1.0|3.0|[1.0,3.0]|
|2.0|2.0|[2.0,2.0]|
|3.0|4.0|[3.0,4.0]|
+---+---+---------+

Spark >= 2.4

在Spark 2.4 之后， VectorAssembler 继承了 HasHandleInvalid， 可以选择skip跳过：

assembler.setHandleInvalid("skip").transform(df).show

+---+---+---------+
| x1| x2| features|
+---+---+---------+
|3.0|4.0|[3.0,4.0]|
+---+---+---------+

keep (note that ML algorithms are unlikely to handle this correctly):

二： 保留：

注： ML算法不太可能正确处理这个问题

assembler.setHandleInvalid("keep").transform(df).show

+----+----+---------+
|  x1|  x2| features|
+----+----+---------+
| 1.0|null|[1.0,NaN]|
|null| 2.0|[NaN,2.0]|
| 3.0| 4.0|[3.0,4.0]|
+----+----+---------+

或者 报异常，默认是异常