[
https://issues.apache.org/jira/browse/SPARK-8733?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joseph K. Bradley updated SPARK-8733:
-------------------------------------
Attachment: Screen Shot 2015-06-30 at 10.51.44 AM.png
> ML RDD.unpersist calls should use blocking = false
> --------------------------------------------------
>
> Key: SPARK-8733
> URL: https://issues.apache.org/jira/browse/SPARK-8733
> Project: Spark
> Issue Type: Improvement
> Components: ML, MLlib
> Reporter: Joseph K. Bradley
> Attachments: Screen Shot 2015-06-30 at 10.51.44 AM.png
>
> Original Estimate: 72h
> Remaining Estimate: 72h
>
> MLlib uses unpersist in many places, but is not consistent about blocking vs
> not. We should check through all of MLlib and change calls to use blocking =
> false, unless there is a real need to block. I have run into issues with
> futures timing out because of unpersist() calls, when there was no real need
> for the ML method to fail.
> See attached screenshot. Training succeeded, but the final unpersist during
> cleanup failed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
