Github user sethah commented on a diff in the pull request:
https://github.com/apache/spark/pull/15621#discussion_r85009838
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/optim/WeightedLeastSquares.scala ---
@@ -137,61 +133,68 @@ private[ml] class WeightedLeastSquares(
}
}
- // scale aBar to standardized space in-place
- val aBarValues = aBar.values
- var j = 0
- while (j < numFeatures) {
- if (aStd(j) == 0.0) {
- aBarValues(j) = 0.0
- } else {
- aBarValues(j) /= aStd(j)
- }
- j += 1
- }
+ val bBar = summary.bBar / bStd
+ val bbBar = summary.bbBar / (bStd * bStd)
- // scale abBar to standardized space in-place
- val abBarValues = abBar.values
- val aStdValues = aStd.values
- j = 0
- while (j < numFeatures) {
- if (aStdValues(j) == 0.0) {
- abBarValues(j) = 0.0
- } else {
- abBarValues(j) /= (aStdValues(j) * bStd)
+ val aStd = summary.aStd
+ val aBar = {
+ val _aBar = summary.aBar
+ var i = 0
+ // scale aBar to standardized space in-place
+ while (i < numFeatures) {
+ if (aStd(i) == 0.0) {
+ _aBar.values(i) = 0.0
--- End diff --
General suggestion, the following two are not equivalent:
````scala
while (j < numFeatures) {
_abar.values(i) /= aStd(i)
}
````
````scala
val _abarValues = _abar.values
while (j < numFeatures) {
_abarValues(i) /= aStd(i)
}
````
This is because of how scala compiles class member vals, you are actually
accessing a `getter` method when you call `_abar.values`. For speed, I think we
should revert this back to making explicit pointers to the values. In this
case, this probably isn't a big deal and likely wouldn't matter for anything
but `aaBar` since it's `k by k` (roughly) in dimension. Still, I don't think
it's a bad idea. Thoughts?
---
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]
