zhengruifeng commented on a change in pull request #26803: [SPARK-30178][ML]
RobustScaler support large numFeatures
URL: https://github.com/apache/spark/pull/26803#discussion_r357014334
##########
File path: mllib/src/main/scala/org/apache/spark/ml/feature/RobustScaler.scala
##########
@@ -147,49 +146,44 @@ class RobustScaler (override val uid: String)
override def fit(dataset: Dataset[_]): RobustScalerModel = {
transformSchema(dataset.schema, logging = true)
- val localRelativeError = $(relativeError)
- val summaries = dataset.select($(inputCol)).rdd.map {
- case Row(vec: Vector) => vec
- }.mapPartitions { iter =>
- var agg: Array[QuantileSummaries] = null
- while (iter.hasNext) {
- val vec = iter.next()
- if (agg == null) {
- agg = Array.fill(vec.size)(
- new QuantileSummaries(QuantileSummaries.defaultCompressThreshold,
localRelativeError))
- }
- require(vec.size == agg.length,
- s"Number of dimensions must be ${agg.length} but got ${vec.size}")
- var i = 0
- while (i < vec.size) {
- agg(i) = agg(i).insert(vec(i))
- i += 1
- }
- }
+ val vectors = dataset.select($(inputCol)).rdd.map { case Row(vec: Vector)
=> vec }
+ val numFeatures = MetadataUtils.getNumFeatures(dataset.schema($(inputCol)))
+ .getOrElse(vectors.first().size)
- if (agg == null) {
- Iterator.empty
- } else {
- Iterator.single(agg.map(_.compress))
- }
- }.treeReduce { (agg1, agg2) =>
- require(agg1.length == agg2.length)
- var i = 0
- while (i < agg1.length) {
- agg1(i) = agg1(i).merge(agg2(i))
- i += 1
+ val localRelativeError = $(relativeError)
+ val localUpper = $(upper)
+ val localLower = $(lower)
+
+ // compute scale by the logic in treeAggregate with depth=2
+ // TODO: design a common treeAggregateByKey in PairRDDFunctions?
+ val scale = math.max(math.ceil(math.sqrt(vectors.getNumPartitions)).toInt,
2)
Review comment:
> say there are M partitions of data, and N dimensions / features. This
produces N*sqrt(M) new partitions.
Sorry I am try to figure out what _partition_ here means. Does it means a
intermediate aggregator `QuantileSummaries`?
Or if you mean a _partition_ in internal rdds, I think it is still M
partitions, since all the ops
(`mapPartitionsWithIndex`->`aggregateByKey`->`map`->`reduceByKey`) before
`collect` do not change numPartitions.
The base logic is quite similar to `treeAggregate`:
Say a key (i=1), the orginal rdd has 16 partitions. There will be 16
aggregator `QuantileSummaries` for i=1 at first, one per partition.
Then after `aggregateByKey` there will partitally merged and there are 4
`QuantileSummaries`.
Finally `reduceByKey` merge them to generate only one `QuantileSummaries`
on some executor.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
