Apache Spark UserDefinedAggregateFunction combining maps

CombineMaps.scala

import org.apache.spark.SparkContext
import org.apache.spark.sql.expressions.{MutableAggregationBuffer, UserDefinedAggregateFunction}
import org.apache.spark.sql.types._
import org.apache.spark.sql.{Column, Row, SQLContext}

/***
  * UDAF combining maps, overriding any duplicate key with "latest" value
  * @param keyType DataType of Map key
  * @param valueType DataType of Value key
  * @param merge function to merge values of identical keys
  * @tparam K key type
  * @tparam V value type
  */
class CombineMaps[K, V](keyType: DataType, valueType: DataType, merge: (V, V) => V) extends UserDefinedAggregateFunction {
  override def inputSchema: StructType = new StructType().add("map", dataType)
  override def bufferSchema: StructType = inputSchema
  override def dataType: DataType = MapType(keyType, valueType)
  override def deterministic: Boolean = true

  override def initialize(buffer: MutableAggregationBuffer): Unit = buffer.update(0, Map[K, V]())

  override def update(buffer: MutableAggregationBuffer, input: Row): Unit = {
    val map1 = buffer.getAs[Map[K, V]](0)
    val map2 = input.getAs[Map[K, V]](0)
    val result = map1 ++ map2.map { case (k,v) => k -> map1.get(k).map(merge(v, _)).getOrElse(v) }
    buffer.update(0, result)
  }

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

  override def evaluate(buffer: Row): Any = buffer.getAs[Map[K, V]](0)
}

object Example {
  def main(args: Array[String]): Unit = {
    import org.apache.spark.sql.functions._

    val sc: SparkContext = new SparkContext("local", "test")
    val sqlContext = new SQLContext(sc)

    import sqlContext.implicits._
    val input = sc.parallelize(Seq(
      (1, Map("John" -> 12.5, "Alice" -> 5.5)),
      (1, Map("Jim" -> 16.5, "Alice" -> 4.0)),
      (2, Map("John" -> 13.5, "Jim" -> 2.5))
    )).toDF("id", "scoreMap")

    // instantiate a CombineMaps with the relevant types:
    val combineMaps = new CombineMaps[String, Double](StringType, DoubleType, _ + _)

    // groupBy and aggregate
    val result = input.groupBy("id").agg(combineMaps(col("scoreMap")))

    result.show(truncate = false)
    // +---+--------------------------------------------+
    // |id |CombineMaps(scoreMap)                       |
    // +---+--------------------------------------------+
    // |1  |Map(John -> 12.5, Alice -> 9.5, Jim -> 16.5)|
    // |2  |Map(John -> 13.5, Jim -> 2.5)               |
    // +---+--------------------------------------------+

  }
}

Very nice example, thank you! I have a question, though. What purpose do the input and buffer schemas serve? I can't seem to get them to do anything. I had expected inputSchema to evaluate whether the correct columns and types were passed in, but that doesn't seem to be true.

Some cells can be null, so you probably need to check for that using if (!input.isNullAt(0))
This was very helpful 👍