[
https://issues.apache.org/jira/browse/SPARK-5695?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14335239#comment-14335239
]
Joseph K. Bradley commented on SPARK-5695:
------------------------------------------
This is likely from re-computing the model predictions on each iteration (and
communicating larger and larger models). I've made a JIRA for that.
I don't see any bugs in the caching logic (but it is a bit complex). I think
I'll close this JIRA for now.
> Check GBT caching logic
> -----------------------
>
> Key: SPARK-5695
> URL: https://issues.apache.org/jira/browse/SPARK-5695
> Project: Spark
> Issue Type: Task
> Components: MLlib
> Reporter: Xiangrui Meng
> Assignee: Joseph K. Bradley
>
> A user reported that `t(n) = t(n-1) + const`, which may caused by cached RDDs
> being kicked out during training. We want to check whether there exists space
> to improve the caching logic.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
