Github user srowen commented on a diff in the pull request:
https://github.com/apache/spark/pull/22986#discussion_r233232112
--- Diff: mllib/src/main/scala/org/apache/spark/ml/tree/treeParams.scala ---
@@ -538,7 +538,7 @@ private[ml] object GBTClassifierParams {
Array("logistic").map(_.toLowerCase(Locale.ROOT))
}
-private[ml] trait GBTClassifierParams extends GBTParams with
TreeClassifierParams {
+private[ml] trait GBTClassifierParams extends GBTParams with
TreeRegressorParams {
--- End diff --
I looked more into this and see that setImpurity is deprecated. I presume
the point is to use set(impurity, ...) instead. Yeah, that's no longer possible
to override in subclasses, scratch that; overriding setImpurity would have been
just fine IMHO but that's not going to last anyway.
I do agree this is why there are different traits for classifiers and
regressors, but I don't think that means a classifier should extend
TreeRegressorParams because its parameters happen to match.
One option is to let the definition of `impurity` itself be overridden.
That seems OK. Or we could make a new 'VarianceClassifier' or something that
defines this variance-only impurity parameter and let TreeRegressionParams and
GBTClassifierParams extend it. It is a little wacky, but quite reasonable from
an OOP perspective. I think.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
