将Spark Scala数据集转换为特定的RDD格式

Question

我有一个看起来像这样的数据框：

+--------------------+-----------------+
|     recommendations|relevant_products|
+--------------------+-----------------+
|[12949, 12949, 71...|           [4343]|
|[12949, 12949, 71...|           [1589]|
|[12949, 12949, 71...|          [11497]|

evaluation_ds:org.apache.spark.sql.Dataset[docCompare] = [recommendations: array, relevant_products: array]

这是数据集中使用的类：case class docCompare (recommendations: Array[Int], relevant_products: Array[Int])

如何将其转换为以下格式的JavaRDD：

 org.apache.spark.rdd.RDD[(Array[?], Array[?])]

Answer 1

您只需将rdd应用于数据集，如下所示：

val evaluation_ds = Seq(
  (Seq(3446, 3843, 1809), Seq(1249)),
  (Seq(4557, 4954, 2920), Seq(2360))
).toDF("recommendations", "relevant_products").as[(Array[Int], Array[Int])]

import org.apache.spark.mllib.evaluation.RankingMetrics

val metrics = new RankingMetrics(evaluation_ds.rdd)
// metrics: org.apache.spark.mllib.evaluation.RankingMetrics[Int] = ...