Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[JVM-Packages] Allow XGBoost jvm package run on GPU without spark-rapids #11184

Merged
merged 1 commit into from
Jan 26, 2025
Merged

[JVM-Packages] Allow XGBoost jvm package run on GPU without spark-rapids #11184

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion ...s/xgboost4j-spark-gpu/src/main/scala/ml/dmlc/xgboost4j/scala/spark/GpuXGBoostPlugin.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
Copyright (c) 2024 by Contributors
Copyright (c) 2024-2025 by Contributors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
Expand Down
33 changes: 17 additions & 16 deletions ...oost4j-spark-gpu/src/test/scala/ml/dmlc/xgboost4j/scala/spark/GpuXGBoostPluginSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,24 +90,23 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
val df = Seq((1.0f, 2.0f, 0.0f),
(2.0f, 3.0f, 1.0f)
).toDF("c1", "c2", "label")
val classifier = new XGBoostClassifier()
assert(classifier.getPlugin.isDefined)
assert(classifier.getPlugin.get.isEnabled(df) === expected)
assert(PluginUtils.getPlugin.isDefined)
assert(PluginUtils.getPlugin.get.isEnabled(df) === expected)
}

// spark.rapids.sql.enabled is not set explicitly, default to true
withSparkSession(new SparkConf(), spark => {
checkIsEnabled(spark, true)
checkIsEnabled(spark, expected = true)
})

// set spark.rapids.sql.enabled to false
withCpuSparkSession() { spark =>
checkIsEnabled(spark, false)
checkIsEnabled(spark, expected = false)
}

// set spark.rapids.sql.enabled to true
withGpuSparkSession() { spark =>
checkIsEnabled(spark, true)
checkIsEnabled(spark, expected = true)
}
}

Expand All @@ -122,7 +121,7 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
).toDF("c1", "c2", "weight", "margin", "label", "other")
val classifier = new XGBoostClassifier()

val plugin = classifier.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
val plugin = PluginUtils.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
intercept[IllegalArgumentException] {
plugin.validate(classifier, df)
}
Expand Down Expand Up @@ -156,9 +155,9 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
var classifier = new XGBoostClassifier()
.setNumWorkers(3)
.setFeaturesCol(features)
assert(classifier.getPlugin.isDefined)
assert(classifier.getPlugin.get.isInstanceOf[GpuXGBoostPlugin])
var out = classifier.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
assert(PluginUtils.getPlugin.isDefined)
assert(PluginUtils.getPlugin.get.isInstanceOf[GpuXGBoostPlugin])
var out = PluginUtils.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
.preprocess(classifier, df)

assert(out.schema.names.contains("c1") && out.schema.names.contains("c2"))
Expand All @@ -172,7 +171,7 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
.setWeightCol("weight")
.setBaseMarginCol("margin")
.setDevice("cuda")
out = classifier.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
out = PluginUtils.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
.preprocess(classifier, df)

assert(out.schema.names.contains("c1") && out.schema.names.contains("c2"))
Expand Down Expand Up @@ -207,7 +206,7 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
.setDevice("cuda")
.setMissing(missing)

val rdd = classifier.getPlugin.get.buildRddWatches(classifier, df)
val rdd = PluginUtils.getPlugin.get.buildRddWatches(classifier, df)
val result = rdd.mapPartitions { iter =>
val watches = iter.next()
val size = watches.size
Expand Down Expand Up @@ -271,7 +270,7 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
.setMissing(missing)
.setEvalDataset(eval)

val rdd = classifier.getPlugin.get.buildRddWatches(classifier, train)
val rdd = PluginUtils.getPlugin.get.buildRddWatches(classifier, train)
val result = rdd.mapPartitions { iter =>
val watches = iter.next()
val size = watches.size
Expand Down Expand Up @@ -324,7 +323,7 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
.setLabelCol("label")
.setDevice("cuda")

assert(estimator.getPlugin.isDefined && estimator.getPlugin.get.isEnabled(df))
assert(PluginUtils.getPlugin.isDefined && PluginUtils.getPlugin.get.isEnabled(df))

val out = estimator.fit(df).transform(df)
// Transform should not discard the other columns of the transforming dataframe
Expand Down Expand Up @@ -528,7 +527,8 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
.setGroupCol(group)
.setDevice("cuda")

val processedDf = ranker.getPlugin.get.asInstanceOf[GpuXGBoostPlugin].preprocess(ranker, df)
val processedDf = PluginUtils.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
.preprocess(ranker, df)
processedDf.rdd.foreachPartition { iter => {
var prevGroup = Int.MinValue
while (iter.hasNext) {
Expand Down Expand Up @@ -575,7 +575,8 @@ class GpuXGBoostPluginSuite extends GpuTestSuite {
// The fix has replaced repartition with repartitionByRange which will put the
// instances with same group into the same partition
val ranker = new XGBoostRanker().setGroupCol("group").setNumWorkers(num_workers)
val processedDf = ranker.getPlugin.get.asInstanceOf[GpuXGBoostPlugin].preprocess(ranker, df)
val processedDf = PluginUtils.getPlugin.get.asInstanceOf[GpuXGBoostPlugin]
.preprocess(ranker, df)
val rows = processedDf
.select("group")
.mapPartitions { case iter =>
Expand Down
21 changes: 10 additions & 11 deletions ...kages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostEstimator.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
Copyright (c) 2024 by Contributors
Copyright (c) 2024-2025 by Contributors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -66,7 +66,7 @@ private[spark] trait NonParamVariables[T <: XGBoostEstimator[T, M], M <: XGBoost
}
}

private[spark] trait PluginMixin {
private[spark] object PluginUtils {
// Find the XGBoostPlugin by ServiceLoader
private val plugin: Option[XGBoostPlugin] = {
val classLoader = Option(Thread.currentThread().getContextClassLoader)
Expand All @@ -85,18 +85,17 @@ private[spark] trait PluginMixin {
}

/** Visible for testing */
protected[spark] def getPlugin: Option[XGBoostPlugin] = plugin
def getPlugin: Option[XGBoostPlugin] = plugin

protected def isPluginEnabled(dataset: Dataset[_]): Boolean = {
def isPluginEnabled(dataset: Dataset[_]): Boolean = {
plugin.map(_.isEnabled(dataset)).getOrElse(false)
}
}

private[spark] trait XGBoostEstimator[
Learner <: XGBoostEstimator[Learner, M], M <: XGBoostModel[M]] extends Estimator[M]
with XGBoostParams[Learner] with SparkParams[Learner] with ParamUtils[Learner]
with NonParamVariables[Learner, M] with ParamMapConversion with DefaultParamsWritable
with PluginMixin {
with NonParamVariables[Learner, M] with ParamMapConversion with DefaultParamsWritable {

protected val logger = LogFactory.getLog("XGBoostSpark")

Expand Down Expand Up @@ -428,8 +427,8 @@ private[spark] trait XGBoostEstimator[
protected def train(dataset: Dataset[_]): M = {
validate(dataset)

val rdd = if (isPluginEnabled(dataset)) {
getPlugin.get.buildRddWatches(this, dataset)
val rdd = if (PluginUtils.isPluginEnabled(dataset)) {
PluginUtils.getPlugin.get.buildRddWatches(this, dataset)
} else {
val (input, columnIndexes) = preprocess(dataset)
toRdd(input, columnIndexes)
Expand Down Expand Up @@ -466,7 +465,7 @@ private[spark] case class PredictedColumns(
* XGBoost base model
*/
private[spark] trait XGBoostModel[M <: XGBoostModel[M]] extends Model[M] with MLWritable
with XGBoostParams[M] with SparkParams[M] with ParamUtils[M] with PluginMixin {
with XGBoostParams[M] with SparkParams[M] with ParamUtils[M] {

protected val TMP_TRANSFORMED_COL = "_tmp_xgb_transformed_col"

Expand Down Expand Up @@ -597,8 +596,8 @@ private[spark] trait XGBoostModel[M <: XGBoostModel[M]] extends Model[M] with ML
}

override def transform(dataset: Dataset[_]): DataFrame = {
if (getPlugin.isDefined) {
return getPlugin.get.transform(this, dataset)
if (PluginUtils.isPluginEnabled(dataset)) {
return PluginUtils.getPlugin.get.transform(this, dataset)
}
validateFeatureType(dataset.schema)
val (schema, pred) = preprocess(dataset)
Expand Down
Loading