Skip to content

[SPARK-21306][ML] OneVsRest should support setWeightCol #18554

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
facaiy wants to merge 16 commits into from
Closed

[SPARK-21306][ML] OneVsRest should support setWeightCol #18554

Show file tree
Hide file tree
Changes from 7 commits
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
f178595
BUG: cache weightCol if necessary
facaiy Jul 6, 2017
a3e6e96
TST: add unit test
facaiy Jul 6, 2017
931d02d
TST: concise comment
facaiy Jul 6, 2017
25d681f
ENH: python, cache weightCol
facaiy Jul 7, 2017
c380ba7
BUG: check weightCol defined
facaiy Jul 7, 2017
e511b90
TST: model is not None
facaiy Jul 7, 2017
1c215f3
CLN: use if-indent style
facaiy Jul 7, 2017
2d3ce90
ENH: add weightCol
facaiy Jul 13, 2017
ea92228
TST: modify test case
facaiy Jul 13, 2017
57fc4b3
ENH: add weightCol for python part
facaiy Jul 13, 2017
00a7ed8
CLN: revise comment
facaiy Jul 13, 2017
a57f096
TST: python test for setWeightCol
facaiy Jul 13, 2017
54e0fca
DOC: add description
facaiy Jul 13, 2017
9ba0e2b
TST: classifier doesn't have weightCol
facaiy Jul 19, 2017
db303a0
CLN: mv weightCol to the end
facaiy Jul 26, 2017
8c0beba
TST: remove ovr2 nad ovr4
facaiy Jul 26, 2017
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
8 changes: 7 additions & 1 deletion 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 @@ -34,6 +34,7 @@ import org.apache.spark.ml._
import org.apache.spark.ml.attribute._
import org.apache.spark.ml.linalg.Vector
import org.apache.spark.ml.param.{Param, ParamMap, ParamPair, Params}
import org.apache.spark.ml.param.shared.HasWeightCol
import org.apache.spark.ml.util._
import org.apache.spark.sql.{DataFrame, Dataset, Row}
import org.apache.spark.sql.functions._
Expand Down Expand Up @@ -317,7 +318,12 @@ final class OneVsRest @Since("1.4.0") (
val numClasses = MetadataUtils.getNumClasses(labelSchema).fold(computeNumClasses())(identity)
instr.logNumClasses(numClasses)

val multiclassLabeled = dataset.select($(labelCol), $(featuresCol))
val multiclassLabeled = getClassifier match {
// SPARK-21306: cache weightCol if necessary
case c: HasWeightCol if c.isDefined(c.weightCol) && c.getWeightCol.nonEmpty =>
dataset.select($(labelCol), $(featuresCol), c.getWeightCol)
case _ => dataset.select($(labelCol), $(featuresCol))
}
Copy link
Contributor

@yanboliang yanboliang Jul 11, 2017

Choose a reason for hiding this comment

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

OneVsRest is a classification estimator, I think we should make weightCol a member param of it like featuresCol. For example:

val dataset = dataset       // This dataset has column: a, b, c.
val ova = new OneVsRest().setFeaturesCol("a").setClassifier(new LogisticRegression().setFeaturesCol("b"))

The features column used by OneVsRest is a. The features column set for OneVsRest will override corresponding set in OneVsRest.classifier. So we should follow this way for weightCol as well. Thanks.

Copy link
Contributor Author

@facaiy facaiy Jul 12, 2017 •
edited
Loading

Choose a reason for hiding this comment

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

Hi, @yanboliang . As @MLnick said, not all classifiers inherits HasWeightCol, so it might cause confusion.

In my opinion, setWeightCol is an attribute owned by one specific classifier itself, like setProbabilityCol and setRawPredictionCol for Logistic Regreesion. So I'd suggest that user should configure the classifier itself, rather than OneVsRest.

Copy link
Contributor

@yanboliang yanboliang Jul 12, 2017 •
edited
Loading

Choose a reason for hiding this comment

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

@facaiy It doesn't matter. If the classifier doesn't inherit from HasWeightCol, we don't run setWeightCol for that classifier but to print out warning log to say weightCol doesn't take effect. You can refer these lines of code to learn how featuresCol be handled. We can do it in similar way. Thanks.


// persist if underlying dataset is not persistent.
val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE
Expand Down
8 changes: 8 additions & 0 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 @@ -156,6 +156,14 @@ class OneVsRestSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
assert(output.schema.fieldNames.toSet === Set("label", "features", "prediction"))
}

test("SPARK-21306: OneVsRest should cache weightCol if necessary") {
val dataset2 = dataset.withColumn("weight", lit(1))
val ova = new OneVsRest().setClassifier(new LogisticRegression().setWeightCol("weight"))
// failed if weightCol is not cached.
val ovaModel = ova.fit(dataset2)
assert(ovaModel !== null)
}

test("OneVsRest.copy and OneVsRestModel.copy") {
val lr = new LogisticRegression()
.setMaxIter(1)
Expand Down
7 changes: 6 additions & 1 deletion python/pyspark/ml/classification.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1546,7 +1546,12 @@ def _fit(self, dataset):

numClasses = int(dataset.agg({labelCol: "max"}).head()["max("+labelCol+")"]) + 1

multiclassLabeled = dataset.select(labelCol, featuresCol)
if (isinstance(classifier, HasWeightCol)
and classifier.isDefined(classifier.weightCol)
and classifier.getWeightCol()):
multiclassLabeled = dataset.select(labelCol, featuresCol, classifier.getWeightCol())
else:
multiclassLabeled = dataset.select(labelCol, featuresCol)

# persist if underlying dataset is not persistent.
handlePersistence = \
Expand Down
10 changes: 10 additions & 0 deletions python/pyspark/ml/tests.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1255,6 +1255,16 @@ def test_output_columns(self):
output = model.transform(df)
self.assertEqual(output.columns, ["label", "features", "prediction"])

def test_cache_weightCol_if_necessary(self):
df = self.spark.createDataFrame([(0.0, Vectors.dense(1.0, 0.8), 1.0),
(1.0, Vectors.sparse(2, [], []), 1.0),
(2.0, Vectors.dense(0.5, 0.5), 1.0)],
["label", "features", "weight"])
lr = LogisticRegression(maxIter=5, regParam=0.01, weightCol="weight")
ovr = OneVsRest(classifier=lr)
model = ovr.fit(df)
self.assertIsNotNone(model)


class HashingTFTest(SparkSessionTestCase):

Expand Down