Question

是否可以在Spark Scala中使用非数字值旋转表格？我已经回顾了以下两个Stack问题。

List in the Case-When Statement in Spark SQL

按照案例中的＆＃34;列表中的步骤 - 何时＆＃34;问题，我可以转换我的数据，以便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＆＃34;首先是非空的＆＃34;可以将每个实体的多行折叠成一个？

的函数

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

Answer 1

这是一个完整的Scala类，它创建一个示例数据框，然后将其转动。它是针对这个问题的，所以我不知道它有多普遍有用。还没有经过广泛测试，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

您可以考虑使用aggregate方法（或aggregateByKey）。你只需要编写适当的函数来获得每个位置的非null元素。

在Spark Scala中旋转非数字表

2 个答案: