[
https://issues.apache.org/jira/browse/SPARK-6004?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14337025#comment-14337025
]
Joseph K. Bradley commented on SPARK-6004:
------------------------------------------
The point of the previous PR introducing validation was to allow early
stopping. We should keep early stopping as an option, but I do think your
JIRA/PR bring up one good point: Even if we never stop early, it may make more
sense to return the best model, rather than the last model. (However, some
users may want the full model since they went to all of that trouble to train
it.)
Ping [~MechCoder] --- what do you think?
I'd vote for:
* allow early stopping based on validationTol
* return the best model instead of the full model (if we do not stop early
while doing validation)
> Pick the best model when training GradientBoostedTrees with validation
> ----------------------------------------------------------------------
>
> Key: SPARK-6004
> URL: https://issues.apache.org/jira/browse/SPARK-6004
> Project: Spark
> Issue Type: Improvement
> Components: MLlib
> Reporter: Liang-Chi Hsieh
> Priority: Minor
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
