Skip to content

[SPARK-21108] [ML] convert LinearSVC to aggregator framework #18315

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
hhbyyh wants to merge 9 commits into from
Closed

[SPARK-21108] [ML] convert LinearSVC to aggregator framework #18315

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
071ce88
add hinge agg
YY-OnCall Jun 15, 2017
87c7279
resolve conflict
YY-OnCall Jul 26, 2017
c8228fb
add unit test
YY-OnCall Jul 26, 2017
94e0250
add 1 un
YY-OnCall Jul 28, 2017
bc33cf9
Merge remote-tracking branch 'upstream/master' into svcAggregator
YY-OnCall Aug 19, 2017
3d7bfce
unit test update
YY-OnCall Aug 20, 2017
b19ca41
Merge remote-tracking branch 'upstream/master' into svcAggregator
YY-OnCall Aug 23, 2017
30e8646
Merge remote-tracking branch 'upstream/master' into svcAggregator
YY-OnCall Aug 23, 2017
4173084
update std case
YY-OnCall Aug 23, 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
Prev Previous commit
Next Next commit
unit test update
  • Loading branch information
@YY-OnCall
YY-OnCall committed Aug 20, 2017
commit 3d7bfce704efb252c93f599b8af0829555c5fe7f
3 changes: 2 additions & 1 deletion mllib/src/main/scala/org/apache/spark/ml/classification/LinearSVC.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,12 +214,13 @@ class LinearSVC @Since("2.2.0") (
}

val featuresStd = summarizer.variance.toArray.map(math.sqrt)
val getFeaturesStd = (j: Int) => featuresStd(j)
val regParamL2 = $(regParam)
val bcFeaturesStd = instances.context.broadcast(featuresStd)
val regularization = if (regParamL2 != 0.0) {
val shouldApply = (idx: Int) => idx >= 0 && idx < numFeatures
Some(new L2Regularization(regParamL2, shouldApply,
if ($(standardization)) None else Some(featuresStd)))
if ($(standardization)) None else Some(getFeaturesStd)))
} else {
None
}
Expand Down
2 changes: 1 addition & 1 deletion mllib/src/test/scala/org/apache/spark/ml/classification/LinearSVCSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,7 +171,7 @@ class LinearSVCSuite extends SparkFunSuite with MLlibTestSparkContext with Defau
assert(model2.intercept !== 0.0)
}

test("sparse coefficients in SVCAggregator") {
test("sparse coefficients in HingeAggregator") {
val bcCoefficients = spark.sparkContext.broadcast(Vectors.sparse(2, Array(0), Array(1.0)))
val bcFeaturesStd = spark.sparkContext.broadcast(Array(1.0))
val agg = new HingeAggregator(bcFeaturesStd, true)(bcCoefficients)
Expand Down
14 changes: 7 additions & 7 deletions mllib/src/test/scala/org/apache/spark/ml/optim/aggregator/HingeAggregatorSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@ class HingeAggregatorSuite extends SparkFunSuite with MLlibTestSparkContext {
import DifferentiableLossAggregatorSuite.getClassificationSummarizers

@transient var instances: Array[Instance] = _
@transient var instancesConstantFeature: Array[Instance] = _

override def beforeAll(): Unit = {
super.beforeAll()
Expand All @@ -35,6 +36,10 @@ class HingeAggregatorSuite extends SparkFunSuite with MLlibTestSparkContext {
Instance(1.0, 0.5, Vectors.dense(1.5, 1.0)),
Instance(0.0, 0.3, Vectors.dense(4.0, 0.5))
)
instancesConstantFeature = Array(
Instance(0.0, 0.1, Vectors.dense(1.0, 2.0)),
Instance(1.0, 0.5, Vectors.dense(1.0, 1.0)),
Instance(1.0, 0.3, Vectors.dense(1.0, 0.5)))
}

/** Get summary statistics for some data and create a new HingeAggregator. */
Expand Down Expand Up @@ -74,7 +79,7 @@ class HingeAggregatorSuite extends SparkFunSuite with MLlibTestSparkContext {
}
}

test("check sizes binomial") {
test("check sizes") {
val rng = new scala.util.Random
val numFeatures = instances.head.features.size
val coefWithIntercept = Vectors.dense(Array.fill(numFeatures + 1)(rng.nextDouble))
Expand All @@ -89,8 +94,7 @@ class HingeAggregatorSuite extends SparkFunSuite with MLlibTestSparkContext {
assert(aggNoIntercept.gradient.size === numFeatures)
}


test("check correctness binomial") {
test("check correctness") {
val coefArray = Array(1.0, 2.0)
val intercept = 1.0
val numFeatures = instances.head.features.size
Expand Down Expand Up @@ -134,10 +138,6 @@ class HingeAggregatorSuite extends SparkFunSuite with MLlibTestSparkContext {
}

Copy link
Contributor

@MLnick MLnick Jul 27, 2017

Choose a reason for hiding this comment

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

The other aggregator tests have one for "zero standard deviation". We should add one here too.

Copy link
Contributor Author

@hhbyyh hhbyyh Aug 20, 2017

Choose a reason for hiding this comment

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

Sure. Added.

test("check with zero standard deviation") {
Copy link
Contributor

@sethah sethah Aug 21, 2017

Choose a reason for hiding this comment

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

As we found out in #18896, this test is not thorough enough. We should check all elements of the gradient for correctness.

val instancesConstantFeature = Array(
Instance(0.0, 0.1, Vectors.dense(1.0, 2.0)),
Instance(1.0, 0.5, Vectors.dense(1.0, 1.0)),
Instance(1.0, 0.3, Vectors.dense(1.0, 0.5)))
val binaryCoefArray = Array(1.0, 2.0)
val intercept = 1.0
val aggConstantFeatureBinary = getNewAggregator(instancesConstantFeature,
Expand Down