Spark:列表的交集不起作用

时间:2015-03-12 01:19:11

标签: list scala apache-spark rdd

我有一个表格的RDD:

t1-> (Long, List[Long])

和表格列表

 t2-> List[Long]

我需要执行列表的并集和交集。我正在尝试以下代码:

val t1 = a.map(x => (x._1, (List(x._2)))).reduceByKey(_ ++ _) 
val t2 = b.map(x => (x._1, (List(x._2)))).reduceByKey(_ ++ _).map(x => x._2).collect()
val t3intersect = t1.map(x => (x._1, (x._2.intersect(t2))))
val t3union = t1.map(x => (x._1, (x._2.union(t2))))

当union返回正确的结果时,交集总是一个空列表。我无法确定问题所在。请帮忙!谢谢!

以下是一个例子:

(1, List(1596, 1617, 1929, 2399, 2674))
(2, List(1702, 1785, 1933, 2054, 2583, 2913))
(3, List(1982, 2002, 2048, 2341, 2666))

List(2002, 2399)

这应该返回交叉点:

(1, List(2399))
(2, List())
(3, List(2002))

和union:

(1, List(1596, 1617, 1929, 2399, 2674, 2002))
(2, List(1702, 1785, 1933, 2054, 2583, 2913, 2002, 2399))
(3, List(1982, 2002, 2048, 2341, 2666, 2399))

2 个答案:

答案 0 :(得分:1)

你的交叉口代码看起来不错。它应该工作。也可以尝试这样做以获得更清晰和可能的表现:

val t3intersect = t1.mapValues( _ intersect t2 )

编辑:我不知道什么是ab,以及从中获取t1t2背后的逻辑是什么,但如果你在Spark REPL中初始化t1t2如下,为了测试:

scala> val t1 = sc.parallelize( List(
     | (1, List(1596, 1617, 1929, 2399, 2674)),
     | (2, List(1702, 1785, 1933, 2054, 2583, 2913)),
     | (3, List(1982, 2002, 2048, 2341, 2666)) ), 2)
t1: org.apache.spark.rdd.RDD[(Int, List[Int])] = ParallelCollectionRDD[10] at parallelize at <console>:12

scala> val t2 = List(2002, 2399)
t2: List[Int] = List(2002, 2399)

然后您获得预期的结果:

scala> val tr = t1.mapValues( _ intersect t2 )
tr: org.apache.spark.rdd.RDD[(Int, List[Int])] = MappedValuesRDD[12] at mapValues at <console>:16

scala> tr.collect()
res13: Array[(Int, List[Int])] = Array((1,List(2399)), (2,List()), (3,List(2002)))

所以,请注意其他地方的错误。

答案 1 :(得分:1)

我已将您的问题案例转载如下:

object ItersectionList {

  def main(args: Array[String]) {
    val spConf = new SparkConf().setMaster("local[2]").setAppName("ItersectionList")
    val sc = new SparkContext(spConf)

    val a = Array(
      (1, List(1596, 1617, 1929, 2399, 2674)),
      (2, List(1702, 1785, 1933, 2054, 2583, 2913)),
      (3, List(1982, 2002, 2048, 2341, 2666))
    )

    val t2 = List(2002, 2399)

    val t1 = sc.makeRDD(a).map(x => (x._1, (List(x._2)))).reduceByKey(_ ++ _)
    val t3intersect = t1.map(x => (x._1, (x._2.intersect(t2))))
    val t3union = t1.map(x => (x._1, (x._2.union(t2))))

    t3intersect.foreach(println)
    t3union.foreach(println)

  }
}

结果如下:

Intersection:
(2,List())
(1,List())
(3,List())

Union: 
(2,List(List(1702, 1785, 1933, 2054, 2583, 2913), 2002, 2399))
(1,List(List(1596, 1617, 1929, 2399, 2674), 2002, 2399))
(3,List(List(1982, 2002, 2048, 2341, 2666), 2002, 2399))

我发现List(x._2)中存在map(x => (x._1, (List(x._2)))).reduceByKey(_ ++ _)问题,将List(a, b, c)更改为List(List(a, b, c))。由于List(List(a, b, c))List(a, b, c)不匹配,因此交集将为空。您可以按照以下方式删除List(),结果将是正确的。

 val t1 = sc.makeRDD(a).map(x => (x._1, x._2)).reduceByKey(_ ++ _)

 val t1 = sc.makeRDD(a).reduceByKey(_ ++ _)