CODEWITHSUNDEEP
scalaapache-sparkdataframeapache-spark-sql
Babu
Babu

Reputation: 891

How do I groupby and concat a list in a Dataframe Spark Scala

I have a dataframe with two columns with data as below 

+----+-----------------+
|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)|
+----+-----------------+

And I need the result as below

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

I tried as below but ,it seems to be not working

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

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

Please let me know how can I achieve this using Scala?

Upvotes: 7

Views: 14451

Answers (3)

Tzach Zohar
Tzach Zohar

Reputation: 37842

Another option that might perform better than the explode option: creating your own UserDefinedAggregationFunction that merges lists into distinct sets.

You'll have to extend UserDefinedAggregateFunction as follows:

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)
}

And use it like so:

val mergeUDAF = new MergeListsUDAF()

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

Upvotes: 3

dportman
dportman

Reputation: 1109

You can try to use collect_set and Window. In your case:

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

Upvotes: 0

Tzach Zohar
Tzach Zohar

Reputation: 37842

You can start by exploding the device column and continue as you did - but note that it might not preserve the order of the lists (which anyway isn't guaranteed in any group by):

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]      |
// +----+-------------------+

Upvotes: 6

Related Questions