如何编写一个简单的行集合的Spark UDAF？

Question

根据我的具体要求,我想编写一个UDAF,它只收集所有输入行.

输入是两列行,Double Type;

中间模式,"我想",是ArrayList(如果我错了,请纠正我)

返回的数据类型是ArrayList

我写了一篇关于我的UDAF的"想法",但我希望有人帮助我完成它.

class CollectorUDAF() extends UserDefinedAggregateFunction {

  // Input Data Type Schema
  def inputSchema: StructType = StructType(Array(StructField("value", DoubleType), StructField("y", DoubleType)))

  // Intermediate Schema
  def bufferSchema = util.ArrayList[Array(StructField("value", DoubleType), StructField("y", DoubleType)]

  // Returned Data Type .
  def dataType: DataType = util.ArrayList[Array(StructField("value", DoubleType), StructField("y", DoubleType)]

  // Self-explaining
  def deterministic = true

  // This function is called whenever key changes
  def initialize(buffer: MutableAggregationBuffer) = {

  }

  // Iterate over each entry of a group
  def update(buffer: MutableAggregationBuffer, input: Row) = {


  }

  // Called after all the entries are exhausted.
  def evaluate(buffer: Row) = {

  }

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

  }

}

Answer 1

如果我理解你的问题是正确的,那么以下是你的解决方案:

class CollectorUDAF() extends UserDefinedAggregateFunction {

  // Input Data Type Schema
  def inputSchema: StructType = new StructType().add("value", DataTypes.DoubleType).add("y", DataTypes.DoubleType)

  // Intermediate Schema
  val bufferFields : util.ArrayList[StructField] = new util.ArrayList[StructField]
  val bufferStructField : StructField = DataTypes.createStructField("array", DataTypes.createArrayType(DataTypes.StringType, true), true)
  bufferFields.add(bufferStructField)
  def bufferSchema: StructType = DataTypes.createStructType(bufferFields)

  // Returned Data Type .
  def dataType: DataType = DataTypes.createArrayType(DataTypes.DoubleType)

  // Self-explaining
  def deterministic = true

  // This function is called whenever key changes
  def initialize(buffer: MutableAggregationBuffer) = {
    buffer(0, new java.util.ArrayList[Double])
  }

  // Iterate over each entry of a group
  def update(buffer: MutableAggregationBuffer, input: Row) = {
    val DoubleList: util.ArrayList[Double]  = new util.ArrayList[Double](buffer.getList(0))
    DoubleList.add(input.getDouble(0))
    DoubleList.add(input.getDouble(1))
    buffer.update(0, DoubleList)
  }

  def merge(buffer1: MutableAggregationBuffer, buffer2: Row): Unit = {
    buffer1.update(0, buffer1.getList(0).toArray() ++ buffer2.getList(0).toArray())
  }
  // Called after all the entries are exhausted.
  def evaluate(buffer: Row) = {
    buffer.getList(0).toArray()
  }
}