srowen commented on a change in pull request #31588:
URL: https://github.com/apache/spark/pull/31588#discussion_r584715387
##########
File path: mllib/src/main/scala/org/apache/spark/ml/feature/VectorSlicer.scala
##########
@@ -110,22 +110,21 @@ final class VectorSlicer @Since("1.5.0") (@Since("1.5.0")
override val uid: Stri
}
// Prepare output attributes
- val inds = getSelectedFeatureIndices(dataset.schema)
- val selectedAttrs = inputAttr.attributes.map { attrs =>
- inds.map(index => attrs(index))
- }
+ val selectedIndices = getSelectedFeatureIndices(dataset.schema)
+ val selectedAttrs = inputAttr.attributes.map { attrs =>
selectedIndices.map(attrs.apply) }
val outputAttr = selectedAttrs match {
case Some(attrs) => new AttributeGroup($(outputCol), attrs)
- case None => new AttributeGroup($(outputCol), inds.length)
+ case None => new AttributeGroup($(outputCol), selectedIndices.length)
}
- // Select features
+ val sorted = selectedIndices.length > 1 &&
selectedIndices.sliding(2).forall(t => t(1) > t(0))
Review comment:
Not quite sure what you're proposing; I just have a moderately strong
preference for not putting this logic in N callers if possible. OK, factoring
it out to a utility class is a little better. Is the runtime check for
sortedness just too expensive and/or not reusable in other functions?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
