将 Spark UDF 与结构序列一起使用 [英] Using Spark UDFs with struct sequences

问题描述

给定一个数据帧，其中一列是由以下序列生成的结构序列

Given a dataframe in which one column is a sequence of structs generated by the following sequence

val df = spark
  .range(10)
  .map((i) => (i % 2, util.Random.nextInt(10), util.Random.nextInt(10)))
  .toDF("a","b","c")
  .groupBy("a")
  .agg(collect_list(struct($"b",$"c")).as("my_list"))
df.printSchema
df.show(false)

输出

root
 |-- a: long (nullable = false)
 |-- my_list: array (nullable = true)
 |    |-- element: struct (containsNull = true)
 |    |    |-- b: integer (nullable = false)
 |    |    |-- c: integer (nullable = false)

+---+-----------------------------------+
|a  |my_list                            |
+---+-----------------------------------+
|0  |[[0,3], [9,5], [3,1], [4,2], [3,3]]|
|1  |[[1,7], [4,6], [5,9], [6,4], [3,9]]|
+---+-----------------------------------+

我需要在每个结构列表上运行一个函数.函数原型类似于下面的函数

I need to run a function over each struct list. The function prototype is similar to the function below

case class DataPoint(b: Int, c: Int)
def do_something_with_data(data: Seq[DataPoint]): Double = {
  // This is an example. I don't actually want the sum
  data.map(data_point => data_point.b + data_point.c).sum
}

我想将此函数的结果存储到另一个 DataFrame 列中.

I want to store the result of this function to another DataFrame column.

我试着跑

val my_udf = udf(do_something_with_data(_))
val df_with_result = df.withColumn("result", my_udf($"my_list"))
df_with_result.show(false)

得到

17/07/13 12:33:42 WARN TaskSetManager: Lost task 0.0 in stage 15.0 (TID 225, REDACTED, executor 0): org.apache.spark.SparkException: Failed to execute user defined function($anonfun$1: (array<struct<b:int,c:int>>) => double)
    at org.apache.spark.sql.catalyst.expressions.GeneratedClass$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:395)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$2.apply(SparkPlan.scala:234)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$2.apply(SparkPlan.scala:228)
    at org.apache.spark.rdd.RDD$$anonfun$mapPartitionsInternal$1$$anonfun$apply$25.apply(RDD.scala:827)
    at org.apache.spark.rdd.RDD$$anonfun$mapPartitionsInternal$1$$anonfun$apply$25.apply(RDD.scala:827)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87)
    at org.apache.spark.scheduler.Task.run(Task.scala:108)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:335)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
    at java.lang.Thread.run(Thread.java:745)
Caused by: java.lang.ClassCastException: org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema cannot be cast to $line27.$read$$iw$$iw$DataPoint
    at $line28.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$anonfun$do_something_with_data$1.apply(<console>:29)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234)
    at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
    at scala.collection.mutable.WrappedArray.foreach(WrappedArray.scala:35)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:234)
    at scala.collection.AbstractTraversable.map(Traversable.scala:104)
    at $line28.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw.do_something_with_data(<console>:29)
    at $line32.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$anonfun$1.apply(<console>:29)
    at $line32.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$anonfun$1.apply(<console>:29)

是否可以在不首先使用 DataFrame API 将我的行转换为容器结构的情况下使用这样的 UDF?

Is it possible to use a UDF like this without first casting my rows to a container struct with the DataFrame API?

做类似的事情:

case class MyRow(a: Long, my_list: Seq[DataPoint])
df.as[MyRow].map(_ => (a, my_list, my_udf(my_list)))

使用 DataSet api 有效，但如果可能，我更愿意坚持使用 DataFrame API.

using the DataSet api works, but I'd prefer to stick with the DataFrame API if possible.

推荐答案

您不能使用 case-class 作为 UDF 的输入参数(但您可以从 UDF 返回 case classes).要映射结构数组，您可以将 Seq[Row] 传递给您的 UDF:

You cannot use a case-class as the input-argument of your UDF (but you can return case classes from the UDF). To map an array of structs, you can pass in a Seq[Row] to your UDF:

val  my_uDF = udf((data: Seq[Row]) => {
  // This is an example. I don't actually want the sum
  data.map{case Row(x:Int,y:Int) => x+y}.sum
})

df.withColumn("result", my_udf($"my_list")).show

+---+--------------------+------+
|  a|             my_list|result|
+---+--------------------+------+
|  0|[[0,3], [5,5], [3...|    41|
|  1|[[0,9], [4,9], [6...|    54|
+---+--------------------+------+

这篇关于将 Spark UDF 与结构序列一起使用的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！