[
https://issues.apache.org/jira/browse/SPARK-13952?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15198859#comment-15198859
]
Nick Pentreath commented on SPARK-13952:
----------------------------------------
[~josephkb] As far as I can see, the old MLlib GBT doesn't support a seed, so
copying it across didn't change that. Strictly speaking this is not a bug, but
an enhancement that should be made to the ML impl (that in turn will be exposed
to MLlib impl once the porting work is all done).
> spark.ml GBT algs need to use random seed
> -----------------------------------------
>
> Key: SPARK-13952
> URL: https://issues.apache.org/jira/browse/SPARK-13952
> Project: Spark
> Issue Type: Bug
> Components: ML
> Reporter: Joseph K. Bradley
> Priority: Minor
>
> SPARK-12379 copied the GBT implementation from spark.mllib to spark.ml.
> There was one bug I found: The random seed is not used. A reasonable fix
> will be to use the original seed to generate a new seed for each tree trained.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
