Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[jvm-packages] unify the set features API (#7692)

Browse Source 
xgboost4j-spark provides 2 sets of API for setting features, one for CPU, another for GPU, which may cause confusion.

This PR removes the GPU API and adds an override CPU function setFeaturesCol to accept Array[String] parameters.
This commit is contained in:
Bobby Wang 2022-02-23 03:37:25 +08:00 committed by GitHub
parent c859764d29
commit e3e6de5ed9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 32 additions and 49 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
jvm-packages/xgboost4j-spark-gpu/src/test/scala/ml/dmlc/xgboost4j/scala/rapids/spark/GpuXGBoostClassifierSuite.scala
View File

@ -1,5 +1,5 @@
/* /*
Copyright (c) 2021 by Contributors Copyright (c) 2021-2022 by Contributors
Licensed under the Apache License, Version 2.0 (the "License"); Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License. you may not use this file except in compliance with the License.
@ -39,7 +39,7 @@ class GpuXGBoostClassifierSuite extends GpuTestSuite {
StructField("f10", FloatType), StructField("f11", FloatType), StructField("f12", FloatType), StructField("f10", FloatType), StructField("f11", FloatType), StructField("f12", FloatType),
StructField(labelName, FloatType) StructField(labelName, FloatType)
)) ))
val featureNames = schema.fieldNames.filter(s => !s.equals(labelName)).toSeq val featureNames = schema.fieldNames.filter(s => !s.equals(labelName))
test("The transform result should be same for several runs on same model") { test("The transform result should be same for several runs on same model") {
withGpuSparkSession(enableCsvConf()) { spark => withGpuSparkSession(enableCsvConf()) { spark =>
@ -90,7 +90,7 @@ class GpuXGBoostClassifierSuite extends GpuTestSuite {
.csv(dataPath).randomSplit(Array(0.7, 0.3), seed = 1) .csv(dataPath).randomSplit(Array(0.7, 0.3), seed = 1)
val classifier = new XGBoostClassifier(xgbParam) val classifier = new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.setTreeMethod("gpu_hist") .setTreeMethod("gpu_hist")
(classifier.fit(rawInput), testDf) (classifier.fit(rawInput), testDf)
@ -155,12 +155,12 @@ class GpuXGBoostClassifierSuite extends GpuTestSuite {
"please refer to setFeaturesCols")) "please refer to setFeaturesCols"))
val left = cpuModel val left = cpuModel
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.transform(testDf) .transform(testDf)
.collect() .collect()
val right = cpuModelFromFile val right = cpuModelFromFile
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.transform(testDf) .transform(testDf)
.collect() .collect()
@ -177,7 +177,7 @@ class GpuXGBoostClassifierSuite extends GpuTestSuite {
.csv(dataPath).randomSplit(Array(0.7, 0.3), seed = 1) .csv(dataPath).randomSplit(Array(0.7, 0.3), seed = 1)
val classifier = new XGBoostClassifier(xgbParam) val classifier = new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.setTreeMethod("gpu_hist") .setTreeMethod("gpu_hist")
classifier.fit(rawInput) classifier.fit(rawInput)

22
jvm-packages/xgboost4j-spark-gpu/src/test/scala/ml/dmlc/xgboost4j/scala/rapids/spark/GpuXGBoostGeneralSuite.scala
View File

@ -1,5 +1,5 @@
/* /*
Copyright (c) 2021 by Contributors Copyright (c) 2021-2022 by Contributors
Licensed under the Apache License, Version 2.0 (the "License"); Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License. you may not use this file except in compliance with the License.
@ -28,7 +28,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
private val labelName = "label_col" private val labelName = "label_col"
private val weightName = "weight_col" private val weightName = "weight_col"
private val baseMarginName = "margin_col" private val baseMarginName = "margin_col"
private val featureNames = Seq("f1", "f2", "f3") private val featureNames = Array("f1", "f2", "f3")
private val allColumnNames = featureNames :+ weightName :+ baseMarginName :+ labelName private val allColumnNames = featureNames :+ weightName :+ baseMarginName :+ labelName
private val trainingData = Seq( private val trainingData = Seq(
// f1, f2, f3, weight, margin, label // f1, f2, f3, weight, margin, label
@ -68,7 +68,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
val xgbParam = Map("eta" -> 0.1f, "max_depth" -> 2, "objective" -> "multi:softprob", val xgbParam = Map("eta" -> 0.1f, "max_depth" -> 2, "objective" -> "multi:softprob",
"num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist") "num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist")
new XGBoostClassifier(xgbParam) new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.fit(trainingDf) .fit(trainingDf)
} }
@ -84,7 +84,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
"num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist") "num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist")
val thrown1 = intercept[IllegalArgumentException] { val thrown1 = intercept[IllegalArgumentException] {
new XGBoostClassifier(xgbParam) new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.fit(trainingDf) .fit(trainingDf)
} }
@ -93,7 +93,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
trainingDf = originalDf.withColumn(labelName, col(labelName).cast(StringType)) trainingDf = originalDf.withColumn(labelName, col(labelName).cast(StringType))
val thrown2 = intercept[IllegalArgumentException] { val thrown2 = intercept[IllegalArgumentException] {
new XGBoostClassifier(xgbParam) new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.fit(trainingDf) .fit(trainingDf)
} }
@ -117,7 +117,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
val thrown1 = intercept[IllegalArgumentException] { val thrown1 = intercept[IllegalArgumentException] {
new XGBoostClassifier(xgbParam) new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.fit(trainingDf) .fit(trainingDf)
} }
assert(thrown1.getMessage.contains("label does not exist.")) assert(thrown1.getMessage.contains("label does not exist."))
@ -132,7 +132,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
"num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "hist") "num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "hist")
val thrown = intercept[IllegalArgumentException] { val thrown = intercept[IllegalArgumentException] {
new XGBoostClassifier(xgbParam) new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.fit(trainingDf) .fit(trainingDf)
} }
@ -149,7 +149,7 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
val xgbParam = Map("eta" -> 0.1f, "max_depth" -> 2, "objective" -> "multi:softprob", val xgbParam = Map("eta" -> 0.1f, "max_depth" -> 2, "objective" -> "multi:softprob",
"num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist") "num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist")
val model1 = new XGBoostClassifier(xgbParam) val model1 = new XGBoostClassifier(xgbParam)
.setFeaturesCols(featureNames) .setFeaturesCol(featureNames)
.setLabelCol(labelName) .setLabelCol(labelName)
.setEvalSets(Map("eval1" -> eval1, "eval2" -> eval2)) .setEvalSets(Map("eval1" -> eval1, "eval2" -> eval2))
.fit(trainingDf) .fit(trainingDf)
@ -166,7 +166,6 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
test("test persistence of XGBoostClassifier and XGBoostClassificationModel") { test("test persistence of XGBoostClassifier and XGBoostClassificationModel") {
val xgbcPath = new File(tempDir.toFile, "xgbc").getPath val xgbcPath = new File(tempDir.toFile, "xgbc").getPath
withGpuSparkSession() { spark => withGpuSparkSession() { spark =>
import spark.implicits._
val xgbParam = Map("eta" -> 0.1f, "max_depth" -> 2, "objective" -> "multi:softprob", val xgbParam = Map("eta" -> 0.1f, "max_depth" -> 2, "objective" -> "multi:softprob",
"num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist", "num_class" -> 3, "num_round" -> 5, "num_workers" -> 1, "tree_method" -> "gpu_hist",
"features_cols" -> featureNames, "label_col" -> labelName) "features_cols" -> featureNames, "label_col" -> labelName)
@ -174,7 +173,10 @@ class GpuXGBoostGeneralSuite extends GpuTestSuite {
xgbc.write.overwrite().save(xgbcPath) xgbc.write.overwrite().save(xgbcPath)
val paramMap2 = XGBoostClassifier.load(xgbcPath).MLlib2XGBoostParams val paramMap2 = XGBoostClassifier.load(xgbcPath).MLlib2XGBoostParams
xgbParam.foreach { xgbParam.foreach {
case (k, v) => assert(v.toString == paramMap2(k).toString) case (k, v: Array[String]) =>
assert(v.sameElements(paramMap2(k).asInstanceOf[Array[String]]))
case (k, v) =>
assert(v.toString == paramMap2(k).toString)
} }
} }
} }

2
jvm-packages/xgboost4j-spark-gpu/src/test/scala/ml/dmlc/xgboost4j/scala/rapids/spark/GpuXGBoostRegressorSuite.scala
View File

@ -35,7 +35,7 @@ class GpuXGBoostRegressorSuite extends GpuTestSuite {
StructField("f3", FloatType), StructField("f3", FloatType),
StructField(groupName, IntegerType))) StructField(groupName, IntegerType)))
val featureNames = schema.fieldNames.filter(s => val featureNames = schema.fieldNames.filter(s =>
!(s.equals(labelName) || s.equals(groupName))).toSeq !(s.equals(labelName) || s.equals(groupName)))
test("The transform result should be same for several runs on same model") { test("The transform result should be same for several runs on same model") {
withGpuSparkSession(enableCsvConf()) { spark => withGpuSparkSession(enableCsvConf()) { spark =>

6
jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostClassifier.scala
View File

@ -1,5 +1,5 @@
/* /*
Copyright (c) 2014,2021 by Contributors Copyright (c) 2014-2022 by Contributors
Licensed under the Apache License, Version 2.0 (the "License"); Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License. you may not use this file except in compliance with the License.
@ -148,7 +148,7 @@ class XGBoostClassifier (
* This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires * This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires
* all feature columns must be numeric types. * all feature columns must be numeric types.
*/ */
def setFeaturesCols(value: Seq[String]): this.type = def setFeaturesCol(value: Array[String]): this.type =
set(featuresCols, value) set(featuresCols, value)
// called at the start of fit/train when 'eval_metric' is not defined // called at the start of fit/train when 'eval_metric' is not defined
@ -264,7 +264,7 @@ class XGBoostClassificationModel private[ml](
* This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires * This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires
* all feature columns must be numeric types. * all feature columns must be numeric types.
*/ */
def setFeaturesCols(value: Seq[String]): this.type = def setFeaturesCol(value: Array[String]): this.type =
set(featuresCols, value) set(featuresCols, value)
/** /**

6
jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostRegressor.scala
View File

@ -1,5 +1,5 @@
/* /*
Copyright (c) 2014,2021 by Contributors Copyright (c) 2014-2022 by Contributors
Licensed under the Apache License, Version 2.0 (the "License"); Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License. you may not use this file except in compliance with the License.
@ -150,7 +150,7 @@ class XGBoostRegressor (
* This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires * This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires
* all feature columns must be numeric types. * all feature columns must be numeric types.
*/ */
def setFeaturesCols(value: Seq[String]): this.type = def setFeaturesCols(value: Array[String]): this.type =
set(featuresCols, value) set(featuresCols, value)
// called at the start of fit/train when 'eval_metric' is not defined // called at the start of fit/train when 'eval_metric' is not defined
@ -257,7 +257,7 @@ class XGBoostRegressionModel private[ml] (
* This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires * This API is only used in GPU train pipeline of xgboost4j-spark-gpu, which requires
* all feature columns must be numeric types. * all feature columns must be numeric types.
*/ */
def setFeaturesCols(value: Seq[String]): this.type = def setFeaturesCols(value: Array[String]): this.type =
set(featuresCols, value) set(featuresCols, value)
/** /**

33
jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/params/GpuParams.scala
View File

@ -1,5 +1,5 @@
/* /*
Copyright (c) 2021 by Contributors Copyright (c) 2021-2022 by Contributors
Licensed under the Apache License, Version 2.0 (the "License"); Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License. you may not use this file except in compliance with the License.
@ -16,38 +16,19 @@
package ml.dmlc.xgboost4j.scala.spark.params package ml.dmlc.xgboost4j.scala.spark.params
import org.json4s.DefaultFormats import org.apache.spark.ml.param.{Params, StringArrayParam}
import org.json4s.jackson.JsonMethods.{compact, parse, render}
import org.apache.spark.ml.param.{BooleanParam, Param, Params}
trait GpuParams extends Params { trait GpuParams extends Params {
/** /**
* Param for the names of feature columns. * Param for the names of feature columns for GPU pipeline.
* @group param * @group param
*/ */
final val featuresCols: StringSeqParam = new StringSeqParam(this, "featuresCols", final val featuresCols: StringArrayParam = new StringArrayParam(this, "featuresCols",
"a sequence of feature column names.") "an array of feature column names for GPU pipeline.")
setDefault(featuresCols, Seq.empty[String]) setDefault(featuresCols, Array.empty[String])
/** @group getParam */ /** @group getParam */
final def getFeaturesCols: Seq[String] = $(featuresCols) final def getFeaturesCols: Array[String] = $(featuresCols)
} }
class StringSeqParam(
parent: Params,
name: String,
doc: String) extends Param[Seq[String]](parent, name, doc) {
override def jsonEncode(value: Seq[String]): String = {
import org.json4s.JsonDSL._
compact(render(value))
}
override def jsonDecode(json: String): Seq[String] = {
implicit val formats = DefaultFormats
parse(json).extract[Seq[String]]
}
}