Spark Scala中的Pivot非数字表

Question

是否有可能在Spark Scala中透视非数值表?我复习了以下两个Stack问题。

如何透视数据框架?

Spark SQL的Case-When语句列表

按照"List in the Case-When"问题中的步骤，我可以转换数据，使每个数据类型都是一列，但每个实体-数据类型组合都有一行。

id    tag    value
1     US     foo
1     UK     bar
1     CA     baz
2     US     hoo
2     UK     hah
2     CA     wah
id    US    UK    CA
1     foo
1           bar
1                 baz
2     hoo
2           hah
3                 wah

是否有一个"第一个非空"函数，可以将每个实体的多行折叠成一个?

id    US    UK    CA
1     foo   bar   baz
2     hoo   hah   wah

Answer 1

这是一个完整的Scala类，它创建了一个示例数据框架，然后对它进行透视。它是针对这个问题的，所以我不知道它有多普遍有用。也没有经过广泛的测试，所以买家要小心。

import org.apache.spark.{SparkConf, SparkContext}
import org.apache.spark.sql.{Column, DataFrame, Row, SQLContext}
import org.apache.spark.sql.functions.{lit, when}
import org.apache.spark.sql.types.{IntegerType, StringType, StructField, StructType}
object DemoPivot {
  def main(args: Array[String]) = {
    def pivotColumn(df: DataFrame)(t: String): Column = {
      val col = when(df("tag") <=> lit(t), df("value"))
      col.alias(t)
    }
    def pivotFrame(sqlContext: SQLContext, df: DataFrame): DataFrame = {
      val tags = df.select("tag").distinct.map(r => r.getString(0)).collect.toList
      df.select(df("id") :: tags.map(pivotColumn(df)): _*)
    }
    def aggregateRows(value: Seq[Option[Any]], agg: Seq[Option[Any]]): Seq[Option[Any]] = {
      for (i <- 0 until Math.max(value.size, agg.size)) yield i match {
        case x if x > value.size => agg(x)
        case y if y > agg.size => value(y)
        case z if value(z).isEmpty => agg(z)
        case a => value(a)
      }
    }
    def collapseRows(sqlContext: SQLContext, df: DataFrame): DataFrame = {
      // RDDs cannot have null elements, so pack into Options and unpack before returning
      val rdd = df.map(row => (Some(row(0)), row.toSeq.tail.map(element => if (element == null) None else Some(element))))
      val agg = rdd.reduceByKey(aggregateRows)
      val aggRdd = agg.map{ case (key, list) => Row.fromSeq((key.get) :: (list.map(element => if (element.isDefined) element.get else null)).toList) }
      sqlContext.createDataFrame(aggRdd, df.schema)
    }
    val conf = new SparkConf().setAppName("Simple Pivot Demo")
    val sc = new SparkContext(conf)
    val sqlContext = new SQLContext(sc)
    val data = List((1, "US", "foo"), (1, "UK", "bar"), (1, "CA", "baz"),
                    (2, "US", "hoo"), (2, "UK", "hah"), (2, "CA", "wah"))
    val rows = data.map(d => Row.fromSeq(d.productIterator.toList))
    val fields = Array(StructField("id", IntegerType, nullable = false),
                       StructField("tag", StringType, nullable = false),
                       StructField("value", StringType, nullable = false))
    val df = sqlContext.createDataFrame(sc.parallelize(rows), StructType(fields))
    df.show()
    val pivoted = pivotFrame(sqlContext, df)
    pivoted.show()
    val collapsed = collapseRows(sqlContext, pivoted)
    collapsed.show()
  }
}

Answer 2

您可以考虑aggregate方法(或aggregateByKey)。您只需要编写适当的函数来在每个位置获取非空元素。