Github user codedeft commented on a diff in the pull request:
https://github.com/apache/spark/pull/2868#discussion_r19249614
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/tree/impl/NodeIdCache.scala ---
@@ -0,0 +1,171 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.mllib.tree.impl
+
+import scala.collection.mutable
+
+import org.apache.spark.rdd.RDD
+import org.apache.spark.annotation.DeveloperApi
+import org.apache.spark.mllib.tree.configuration.FeatureType._
+import spire.implicits._
+import org.apache.spark.storage.StorageLevel
+import org.apache.spark.mllib.tree.model.{Bin, Split}
+
+/**
+ * :: DeveloperApi ::
+ * This is used by the node id cache to find the child id that a data
point would belong to.
+ * @param split Split information.
+ * @param leftChildIndex Left child index.
+ * @param rightChildIndex Right child index.
+ */
+@DeveloperApi
+private[tree] case class NodeIndexUpdater(
+ split: Split,
+ leftChildIndex: Int,
+ rightChildIndex: Int) {
+ /**
+ * Determine a child node index based on the feature value and the split.
+ * @param binnedFeatures Binned feature values.
+ * @param bins Bin information to convert the bin indices to approximate
feature values.
+ * @return Child node index to update to.
+ */
+ def updateNodeIndex(binnedFeatures: Array[Int], bins:
Array[Array[Bin]]): Int = {
+ if (split.featureType == Continuous) {
+ val featureIndex = split.feature
+ val binIndex = binnedFeatures(featureIndex)
+ val featureValueUpperBound =
bins(featureIndex)(binIndex).highSplit.threshold
+ if (featureValueUpperBound <= split.threshold) {
+ leftChildIndex
+ } else {
+ rightChildIndex
+ }
+ } else {
+ if
(split.categories.contains(binnedFeatures(split.feature).toDouble)) {
+ leftChildIndex
+ } else {
+ rightChildIndex
+ }
+ }
+ }
+}
+
+/**
+ * A given TreePoint would belong to a particular node per tree.
+ * This is used to keep track of which node for a particular tree that a
TreePoint belongs to.
+ * A separate RDD of Array[Int] needs to be maintained and updated at each
iteration.
+ * @param data The RDD of training rows.
+ * @param cur The initial values in the cache
+ * (should be an Array of all 1's (meaning the root nodes)).
+ * @param checkpointDir The checkpoint directory where
+ * the checkpointed files will be stored.
+ * @param checkpointInterval The checkpointing interval
+ * (how often should the cache be checkpointed.).
+ */
+@DeveloperApi
+private[tree] class NodeIdCache(
+ val data: RDD[BaggedPoint[TreePoint]],
+ var cur: RDD[Array[Int]],
+ val checkpointDir: Option[String],
+ val checkpointInterval: Int) {
+
+ // To keep track of the past checkpointed RDDs.
+ val checkpointQueue = mutable.Queue[RDD[Array[Int]]]()
+ var rddUpdateCount = 0
+ if (checkpointDir != None) {
+ cur.sparkContext.setCheckpointDir(checkpointDir.get)
+ }
+
+ /**
+ * Update the node index values in the cache.
+ * This updates the RDD and its lineage.
+ * TODO: Passing bin information to executors seems unnecessary and
costly.
--- End diff --
I think that this is fine as long as you get relatively balanced trees,
since the tree would be extremely big by the time you reach level 30.
However, based on my experience, the problem in practice is that you often
get unbalanced trees. E.g., when I train on 60,000 sample mnist without
pruning, I often got a tree with close to ~100 level deep, even though the
number of nodes was only around 5000 or so.
I will do it the way you suggested by simply calling Node.leftChildId,
Node.rightChildId.
For future, though, I think that a relatively easy way to do this might be
by assigning Ids in a FIFO fashion (keepLastAssignedId and +1 for every child
node that comes in). It will likely lead to a bit more complicated code, but I
think that it'll be easier than removing node Ids entirely.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
