Skip to content

[SPARK-19746][ML] Faster indexing for logistic aggregator #17078

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
sethah wants to merge 4 commits into from
Closed

[SPARK-19746][ML] Faster indexing for logistic aggregator #17078

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3bea389
better indexing for logistic agg
sethah Feb 27, 2017
b194a27
transient coefficients and test
sethah Feb 27, 2017
61d22e7
check error msg
sethah Feb 27, 2017
44ee113
add type
sethah Feb 27, 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
11 changes: 8 additions & 3 deletions mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1431,7 +1431,12 @@ private class LogisticAggregator(
private var weightSum = 0.0
private var lossSum = 0.0

private val gradientSumArray = Array.fill[Double](coefficientSize)(0.0D)
@transient private lazy val coefficientsArray: Array[Double] = bcCoefficients.value match {
case DenseVector(values) => values
case _ => throw new IllegalArgumentException(s"coefficients only supports dense vector but " +
s"got type ${bcCoefficients.value.getClass}.)")
}
private val gradientSumArray = new Array[Double](coefficientSize)

if (multinomial && numClasses <= 2) {
logInfo(s"Multinomial logistic regression for binary classification yields separate " +
Expand All @@ -1447,7 +1452,7 @@ private class LogisticAggregator(
label: Double): Unit = {

val localFeaturesStd = bcFeaturesStd.value
val localCoefficients = bcCoefficients.value
val localCoefficients = coefficientsArray
val localGradientArray = gradientSumArray
val margin = - {
var sum = 0.0
Expand Down Expand Up @@ -1491,7 +1496,7 @@ private class LogisticAggregator(
logistic regression without pivoting.
*/
val localFeaturesStd = bcFeaturesStd.value
val localCoefficients = bcCoefficients.value
val localCoefficients = coefficientsArray
val localGradientArray = gradientSumArray

// marginOfLabel is margins(label) in the formula
Expand Down
26 changes: 26 additions & 0 deletions mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -456,6 +456,32 @@ class LogisticRegressionSuite
assert(blrModel.intercept !== 0.0)
}

test("sparse coefficients in LogisticAggregator") {
val bcCoefficientsBinary = spark.sparkContext.broadcast(Vectors.sparse(2, Array(0), Array(1.0)))
val bcFeaturesStd = spark.sparkContext.broadcast(Array(1.0))
val binaryAgg = new LogisticAggregator(bcCoefficientsBinary, bcFeaturesStd, 2,
fitIntercept = true, multinomial = false)
val thrownBinary = withClue("binary logistic aggregator cannot handle sparse coefficients") {
Copy link
Member

@dbtsai dbtsai Feb 27, 2017

Choose a reason for hiding this comment

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

I think we should handle sparse coefficients for further performance improvement. But not in this PR.

intercept[IllegalArgumentException] {
binaryAgg.add(Instance(1.0, 1.0, Vectors.dense(1.0)))
}
}
assert(thrownBinary.getMessage.contains("coefficients only supports dense"))

val bcCoefficientsMulti = spark.sparkContext.broadcast(Vectors.sparse(6, Array(0), Array(1.0)))
val multinomialAgg = new LogisticAggregator(bcCoefficientsMulti, bcFeaturesStd, 3,
fitIntercept = true, multinomial = true)
val thrown = withClue("multinomial logistic aggregator cannot handle sparse coefficients") {
intercept[IllegalArgumentException] {
multinomialAgg.add(Instance(1.0, 1.0, Vectors.dense(1.0)))
}
}
assert(thrown.getMessage.contains("coefficients only supports dense"))
bcCoefficientsBinary.destroy(blocking = false)
bcFeaturesStd.destroy(blocking = false)
bcCoefficientsMulti.destroy(blocking = false)
}

test("overflow prediction for multiclass") {
val model = new LogisticRegressionModel("mLogReg",
Matrices.dense(3, 2, Array(0.0, 0.0, 0.0, 1.0, 2.0, 3.0)),
Expand Down