Skip to content

[SPARK-9312][ML] Add RawPrediction, numClasses, and numFeatures for OneVsRestModel #21044

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

Closed
lu-wang-dl wants to merge 5 commits into from
Closed

[SPARK-9312][ML] Add RawPrediction, numClasses, and numFeatures for OneVsRestModel #21044

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
0cfc20a
add rawPrediction as an output column;
lu-wang-dl Apr 11, 2018
2a47e2b
make rawPrediction optionall
lu-wang-dl Apr 13, 2018
0c32fca
change the tmp array name
lu-wang-dl Apr 13, 2018
ebf4a6c
fix typos and comments
lu-wang-dl Apr 13, 2018
b3c7fec
add require in OneVsRestModel
lu-wang-dl Apr 13, 2018
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
48 changes: 38 additions & 10 deletions mllib/src/main/scala/org/apache/spark/ml/classification/OneVsRest.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ import org.apache.spark.SparkContext
import org.apache.spark.annotation.Since
import org.apache.spark.ml._
import org.apache.spark.ml.attribute._
import org.apache.spark.ml.linalg.Vector
import org.apache.spark.ml.linalg.{Vector, Vectors}
import org.apache.spark.ml.param.{Param, ParamMap, ParamPair, Params}
import org.apache.spark.ml.param.shared.{HasParallelism, HasWeightCol}
import org.apache.spark.ml.util._
Expand All @@ -55,7 +55,7 @@ private[ml] trait ClassifierTypeTrait {
/**
* Params for [[OneVsRest]].
*/
private[ml] trait OneVsRestParams extends PredictorParams
private[ml] trait OneVsRestParams extends ClassifierParams
with ClassifierTypeTrait with HasWeightCol {

/**
Expand Down Expand Up @@ -138,6 +138,12 @@ final class OneVsRestModel private[ml] (
@Since("1.4.0") val models: Array[_ <: ClassificationModel[_, _]])
extends Model[OneVsRestModel] with OneVsRestParams with MLWritable {

Copy link
Member

@jkbradley jkbradley Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's add a require() statement here which checks that models.nonEmpty is true (to throw an exception upon construction, rather than when numFeatures calls models.head below). Just to be safe...

@Since("2.4.0")
val numClasses: Int = models.length

@Since("2.4.0")
val numFeatures: Int = models.head.numFeatures

/** @group setParam */
@Since("2.1.0")
def setFeaturesCol(value: String): this.type = set(featuresCol, value)
Expand All @@ -146,6 +152,10 @@ final class OneVsRestModel private[ml] (
@Since("2.1.0")
def setPredictionCol(value: String): this.type = set(predictionCol, value)

/** @group setParam */
@Since("2.4.0")
def setRawPredictionCol(value: String): this.type = set(rawPredictionCol, value)
Copy link
Member

@jkbradley jkbradley Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You'll need to add this to the Estimator too.


@Since("1.4.0")
override def transformSchema(schema: StructType): StructType = {
validateAndTransformSchema(schema, fitting = false, getClassifier.featuresDataType)
Expand Down Expand Up @@ -181,6 +191,7 @@ final class OneVsRestModel private[ml] (
val updateUDF = udf { (predictions: Map[Int, Double], prediction: Vector) =>
predictions + ((index, prediction(1)))
}

model.setFeaturesCol($(featuresCol))
val transformedDataset = model.transform(df).select(columns: _*)
val updatedDataset = transformedDataset
Expand All @@ -195,15 +206,32 @@ final class OneVsRestModel private[ml] (
newDataset.unpersist()
}

// output the index of the classifier with highest confidence as prediction
val labelUDF = udf { (predictions: Map[Int, Double]) =>
predictions.maxBy(_._2)._1.toDouble
}
// output the RawPrediction as vector
if (getRawPredictionCol != "") {
val rawPredictionUDF = udf { (predictions: Map[Int, Double]) =>
val predArray = Array.fill[Double](numClasses)(0.0)
Copy link
Member

@jkbradley jkbradley Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This causes a subtle ContextCleaner bug: numClasses refers to a field of the class OneVsRestModel, so when Spark's closure capture serializes this UDF to send to executors, it will end up sending the entire OneVsRestModel object, rather than just the value for numClasses. Make a local copy of the value numClasses within the transform() method to avoid this issue.

predictions.foreach { case (idx, value) => predArray(idx) = value }
Vectors.dense(predArray)
}

// output the index of the classifier with highest confidence as prediction
val labelUDF = udf { (predictions: Vector) => predictions.argmax.toDouble }
Copy link
Contributor

@WeichenXu123 WeichenXu123 Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

==> udf { (rawPredictions: Vector) => ... }


// output label and label metadata as prediction
aggregatedDataset
.withColumn($(predictionCol), labelUDF(col(accColName)), labelMetadata)
.drop(accColName)
aggregatedDataset
.withColumn(getRawPredictionCol, rawPredictionUDF(col(accColName)))
.withColumn(getPredictionCol, labelUDF(col(getRawPredictionCol)), labelMetadata)
.drop(accColName)
}
else {
Copy link
Member

@jkbradley jkbradley Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Scala style: This should go on the previous line: } else {

// output the index of the classifier with highest confidence as prediction
val labelUDF = udf { (predictions: Map[Int, Double]) =>
predictions.maxBy(_._2)._1.toDouble
}
// output confidence as rwa prediction, label and label metadata as prediction
Copy link
Contributor

@WeichenXu123 WeichenXu123 Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

rwa -> raw

Copy link
Member

@jkbradley jkbradley Apr 13, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This comment seems to be in the wrong part of the code. Also there's a typo

aggregatedDataset
.withColumn(getPredictionCol, labelUDF(col(accColName)), labelMetadata)
.drop(accColName)
}
}

@Since("1.4.1")
Expand Down
7 changes: 5 additions & 2 deletions mllib/src/test/scala/org/apache/spark/ml/classification/OneVsRestSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,11 +72,12 @@ class OneVsRestSuite extends MLTest with DefaultReadWriteTest {
.setClassifier(new LogisticRegression)
assert(ova.getLabelCol === "label")
assert(ova.getPredictionCol === "prediction")
assert(ova.getRawPredictionCol === "rawPrediction")
val ovaModel = ova.fit(dataset)

MLTestingUtils.checkCopyAndUids(ova, ovaModel)

assert(ovaModel.models.length === numClasses)
assert(ovaModel.numClasses === numClasses)
val transformedDataset = ovaModel.transform(dataset)

// check for label metadata in prediction col
Expand Down Expand Up @@ -179,6 +180,7 @@ class OneVsRestSuite extends MLTest with DefaultReadWriteTest {
val dataset2 = dataset.select(col("label").as("y"), col("features").as("fea"))
ovaModel.setFeaturesCol("fea")
ovaModel.setPredictionCol("pred")
ovaModel.setRawPredictionCol("")
val transformedDataset = ovaModel.transform(dataset2)
val outputFields = transformedDataset.schema.fieldNames.toSet
assert(outputFields === Set("y", "fea", "pred"))
Expand All @@ -190,7 +192,8 @@ class OneVsRestSuite extends MLTest with DefaultReadWriteTest {
val ovr = new OneVsRest()
.setClassifier(logReg)
val output = ovr.fit(dataset).transform(dataset)
assert(output.schema.fieldNames.toSet === Set("label", "features", "prediction"))
assert(output.schema.fieldNames.toSet
=== Set("label", "features", "prediction", "rawPrediction"))
}

test("SPARK-21306: OneVsRest should support setWeightCol") {
Expand Down