You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@flink.apache.org by "Fan Hong (Jira)" <ji...@apache.org> on 2023/03/27 08:45:00 UTC

[jira] [Updated] (FLINK-31625) Possbile OOM in KBinsDiscretizer

     [ https://issues.apache.org/jira/browse/FLINK-31625?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Fan Hong updated FLINK-31625:
-----------------------------
    Description: 
In KBinsDiscretizer, the main computation `findBinEdgesWithXXXStrategy` is accomplished in a single subtask. While data sampling is used to decrease memory usage, the memory overhead can still be prohibitive for large input vectors, potentially resulting in OOM errors.

A potential solution is to implement parallel computation, distributing the data evenly among all workers.

  was:
In KBinsDiscretizer, the main computation `findBinEdgesWithXXXStrategy` is put into a single subtask. While data sampling is used to decrease memory usage, the memory overhead can still be prohibitive for large input vectors, potentially resulting in OOM errors.

A potential solution is to implement parallel computation, distributing the data evenly among all workers.


> Possbile OOM in KBinsDiscretizer
> --------------------------------
>
>                 Key: FLINK-31625
>                 URL: https://issues.apache.org/jira/browse/FLINK-31625
>             Project: Flink
>          Issue Type: Bug
>          Components: Library / Machine Learning
>            Reporter: Fan Hong
>            Priority: Major
>
> In KBinsDiscretizer, the main computation `findBinEdgesWithXXXStrategy` is accomplished in a single subtask. While data sampling is used to decrease memory usage, the memory overhead can still be prohibitive for large input vectors, potentially resulting in OOM errors.
> A potential solution is to implement parallel computation, distributing the data evenly among all workers.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)