Github user karlhigley commented on a diff in the pull request:
https://github.com/apache/spark/pull/9843#discussion_r46000917
--- Diff: mllib/src/main/scala/org/apache/spark/mllib/feature/IDF.scala ---
@@ -218,7 +218,7 @@ private object IDFModel {
newValues(k) = values(k) * idf(indices(k))
k += 1
}
- Vectors.sparse(n, indices, newValues)
+ Vectors.sparse(n, indices, newValues).toSparse
--- End diff --
I can see an argument for consistent behavior with respect to explicit
zeros across the three versions of `Vectors.sparse`. Sorting only happens in
one of the three (as a way of detecting duplicate identifiers), and while
filtering doesn't seem that expensive in comparison, it would be significant
extra work in the other two versions.
The suggestion of optimizing performance with `idf.compressed` and extra
cases in `transform` makes sense to me, but seems like more than is required to
address this particular issue. The thought about `ArrayBuffer` suggests a
reasonably small edit to the existing `SparseVector` case, though. I'll push a
new commit with that change.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
