[
https://issues.apache.org/jira/browse/SPARK-23643?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17758770#comment-17758770
]
Mitesh commented on SPARK-23643:
--------------------------------
+1 can we please document this in the ML migration guide? We use
`Dataset.randomSplit()` to come up with test/train sets, so based on this hash
change, all of our deterministic ML tests all fail (same input data, same
seed/configs, same model...produces diff results).
> XORShiftRandom.hashSeed allocates unnecessary memory
> ----------------------------------------------------
>
> Key: SPARK-23643
> URL: https://issues.apache.org/jira/browse/SPARK-23643
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 2.3.0
> Reporter: Max Gekk
> Assignee: Max Gekk
> Priority: Major
> Fix For: 3.0.0
>
>
> The hashSeed method allocates 64 bytes buffer and puts only 8 bytes of the
> seed parameter into it. Other bytes are always zero and could be easily
> excluded from hash calculation.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
