How to build a large distributed [sparse] matrix in Apache Spark 1.0?

问题

I have an RDD as such: byUserHour: org.apache.spark.rdd.RDD[(String, String, Int)] I would like to create a sparse matrix of the data for calculations like median, mean, etc. The RDD contains the row_id, column_id and value. I have two Arrays containing the row_id and column_id strings for lookup.

Here is my attempt:

import breeze.linalg._
val builder = new CSCMatrix.Builder[Int](rows=BCnUsers.value.toInt,cols=broadcastTimes.value.size)
byUserHour.foreach{x =>
  val row = userids.indexOf(x._1)
  val col = broadcastTimes.value.indexOf(x._2)
  builder.add(row,col,x._3)}
builder.result()

Here is my error:

14/06/10 16:39:34 INFO DAGScheduler: Failed to run foreach at <console>:38
org.apache.spark.SparkException: Job aborted due to stage failure: Task not serializable: java.io.NotSerializableException: breeze.linalg.CSCMatrix$Builder$mcI$sp
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1033)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1017)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1015)
    at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
    at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
    at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1015)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$submitMissingTasks(DAGScheduler.scala:770)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$submitStage(DAGScheduler.scala:713)
    at org.apache.spark.scheduler.DAGScheduler.handleJobSubmitted(DAGScheduler.scala:697)
    at org.apache.spark.scheduler.DAGSchedulerEventProcessActor$$anonfun$receive$2.applyOrElse(DAGScheduler.scala:1176)
    at akka.actor.ActorCell.receiveMessage(ActorCell.scala:498)
    at akka.actor.ActorCell.invoke(ActorCell.scala:456)
    at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:237)
    at akka.dispatch.Mailbox.run(Mailbox.scala:219)
    at akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:386)
    at scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)
    at scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)
    at scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)
    at scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)

My dataset is quite large so I would like to do this distributed if possible. Any help would be appreciated.

---

Progress update:

CSCMartix is not meant to work in Spark. However, there is RowMatrix which extends DistributedMatrix. RowMatrix does have a method, computeColumnSummaryStatistics(), that should compute some of the stats I am looking for. I know MLlib is growing everyday so I will watch for updates, but in the meantime I will try to make an RDD[Vector] to feed RowMatrix. Noting that RowMatrix is experimental and represents a row-oriented distributed Matrix with no meaningful row indices.

回答1:

Starting with mapping a little different byUserHour is now an RDD[(String, (String, Int))] . Because RowMatrix does not preserve order of rows I groupByKey on the row_id. Perhaps in the future I will figure out how to do this with a sparse matrix.

val byUser = byUserHour.groupByKey // RDD[(String, Iterable[(String, Int)])]
val times = countHour.map(x => x._1.split("\\+")(1)).distinct.collect.sortWith(_ < _)  // Array[String]
val broadcastTimes = sc.broadcast(times) // Broadcast[Array[String]]

val userMaps = byUser.mapValues { 
  x => x.map{
    case(time,cnt) => time -> cnt
  }.toMap
}  // RDD[(String, scala.collection.immutable.Map[String,Int])]


val rows = userMaps.map {
  case(u,ut) => (u.toDouble +: broadcastTimes.value.map(ut.getOrElse(_,0).toDouble))}       // RDD[Array[Double]]


import org.apache.spark.mllib.linalg.{Vector, Vectors}
val rowVec = rows.map(x => Vectors.dense(x)) // RDD[org.apache.spark.mllib.linalg.Vector]

import org.apache.spark.mllib.linalg.distributed._
val countMatrix = new RowMatrix(rowVec)
val stats = countMatrix.computeColumnSummaryStatistics()
val meanvec = stats.mean

回答2:

You can do that with a CoordinateMatrix:

import org.apache.spark.mllib.linalg.distributed.{CoordinateMatrix, MatrixEntry}

val sparseMatrix = new CoordinateMatrix(byUserHour.map {
  case (row, col, data) => MatrixEntry(row, col, data) 
})

来源：https://stackoverflow.com/questions/24147186/how-to-build-a-large-distributed-sparse-matrix-in-apache-spark-1-0