[GitHub] spark pull request: [SPARK-3366][MLLIB]Compute best splits distrib...

[chouqin <gi...@git.apache.org>]

Github user chouqin commented on a diff in the pull request:

    https://github.com/apache/spark/pull/2595#discussion_r18258483
  
    --- Diff: mllib/src/main/scala/org/apache/spark/mllib/tree/DecisionTree.scala ---
    @@ -518,30 +516,58 @@ object DecisionTree extends Serializable with Logging {
           agg
         }
     
    -    // Calculate bin aggregates.
    -    timer.start("aggregation")
    -    val binAggregates: DTStatsAggregator = {
    -      val initAgg = if (metadata.subsamplingFeatures) {
    -        new DTStatsAggregatorSubsampledFeatures(metadata, treeToNodeToIndexInfo)
    -      } else {
    -        new DTStatsAggregatorFixedFeatures(metadata, numNodes)
    -      }
    -      input.treeAggregate(initAgg)(binSeqOp, DTStatsAggregator.binCombOp)
    -    }
    -    timer.stop("aggregation")
    -
         // Calculate best splits for all nodes in the group
         timer.start("chooseSplits")
     
    +    // In each parition, iterate all instances and compute aggregate stats for each node,
    +    // yield an (nodeIndex, nodeAggregateStats) pair for each node.
    +    // After a `reduceByKey` operation,
    +    // stats of a node will be shuffled to a particular partition and be combined together,
    +    // then best splits for nodes are found there.
    +    // Finally, only best Splits for nodes are collected to driver to construct decision tree.
    +    val nodeToBestSplits: Map[Int, (Split, InformationGainStats, Predict)] =
    +      input.mapPartitions(points => {
    +        // Construct a nodeStatsAggregators array to hold node aggregate stats,
    +        // each node will have a nodeStatsAggregator
    +        val numNodes = nodeToFeatures.keys.size
    +        val nodeStatsAggregators = new Array[NodeStatsAggregator](numNodes)
    +        var nodeIndex = 0
    +        while (nodeIndex < numNodes) {
    +          nodeStatsAggregators(nodeIndex) =
    +            new NodeStatsAggregator(metadata, nodeToFeatures(nodeIndex))
    +          nodeIndex += 1
    +        }
    +
    +        // iterator all instances in current partition and update aggregate stats
    +        points.foreach(binSeqOp(nodeStatsAggregators, _))
    +
    +        // transform nodeStatsAggregators array to (nodeIndex, nodeAggregateStats) pairs,
    +        // which can be combined with other partition using `reduceByKey`
    +        nodeStatsAggregators.zipWithIndex.map(t => {
    +          (t._2, t._1)
    +        }).iterator
    +      }).reduceByKey((a, b) => a.merge(b))
    +        .map(t => {
    +          val nodeIndex = t._1
    +          val aggStats = t._2
    +          val featuresForNode = nodeToFeatures(nodeIndex)
    +
    +          // find best split for each node
    +          val (split: Split, stats: InformationGainStats, predict: Predict) =
    +            binsToBestSplit(aggStats, splits, featuresForNode, metadata)
    +          (nodeIndex, (split, stats, predict))
    +        }).collectAsMap().toMap
    --- End diff --
    
    this is necessary because `nodeToBestSplits` is defined as Map[...], which is of type 'scala.collection.Map' while `collectAsMap()` returns `scala.collection.immutable.Map`.
    
    I can remove type definition for `nodeToBestSplits` to avoid this. I made this type definition 
    originally because my IDE(idea intellij) can't infer type for it.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org