Question

我的数据框有两列，数据如下

+----+-----------------+
|acct|           device|
+----+-----------------+
|   B|       List(3, 4)|
|   C|       List(3, 5)|
|   A|       List(2, 6)|
|   B|List(3, 11, 4, 9)|
|   C|       List(5, 6)|
|   A|List(2, 10, 7, 6)|
+----+-----------------+

我需要结果如下

+----+-----------------+
|acct|           device|
+----+-----------------+
|   B|List(3, 4, 11, 9)|
|   C|    List(3, 5, 6)|
|   A|List(2, 6, 7, 10)|
+----+-----------------+

我尝试如下，但似乎无法正常工作

df.groupBy("acct").agg(concat("device"))

df.groupBy("acct").agg(collect_set("device"))

请告诉我如何使用Scala实现这一目标？

Answer 1

您可以从爆炸device列开始，然后继续操作 - 但请注意，它可能无法保留列表的顺序（无论如何都不保证在任何组中）：

val result = df.withColumn("device", explode($"device"))
  .groupBy("acct")
  .agg(collect_set("device"))

result.show(truncate = false)
// +----+-------------------+
// |acct|collect_set(device)|
// +----+-------------------+
// |B   |[9, 3, 4, 11]      |
// |C   |[5, 6, 3]          |
// |A   |[2, 6, 10, 7]      |
// +----+-------------------+

Answer 2

您可以尝试使用collect_set和Window。在你的情况下：

df.withColumn("device", collect_set("device").over(Window.partitionBy("acct")))

Answer 3

可能的另一个选项比explode选项更好：创建自己的 UserDefinedAggregationFunction ，将列表合并到不同的集合中。

您必须按如下方式延长UserDefinedAggregateFunction：

class MergeListsUDAF extends UserDefinedAggregateFunction {

  override def inputSchema: StructType = StructType(Seq(StructField("a", ArrayType(IntegerType))))

  override def bufferSchema: StructType = inputSchema

  override def dataType: DataType = ArrayType(IntegerType)

  override def deterministic: Boolean = true

  override def initialize(buffer: MutableAggregationBuffer): Unit = buffer.update(0, mutable.Seq[Int]())

  override def update(buffer: MutableAggregationBuffer, input: Row): Unit = {
    val existing = buffer.getAs[mutable.Seq[Int]](0)
    val newList = input.getAs[mutable.Seq[Int]](0)
    val result = (existing ++ newList).distinct
    buffer.update(0, result)
  }

  override def merge(buffer1: MutableAggregationBuffer, buffer2: Row): Unit = update(buffer1, buffer2)

  override def evaluate(buffer: Row): Any = buffer.getAs[mutable.Seq[Int]](0)
}

并像这样使用它：

val mergeUDAF = new MergeListsUDAF()

df.groupBy("acct").agg(mergeUDAF($"device"))

如何在Dataframe Spark Scala

3 个答案: