[
https://issues.apache.org/jira/browse/SPARK-14408?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15230994#comment-15230994
]
Joseph K. Bradley commented on SPARK-14408:
-------------------------------------------
Not meaning to cause panic here, but I'm escalating this since it might be a
critical bug in MLlib. [~dbtsai] [~mengxr] [~mlnick] [~srowen] could you
please help me confirm that this is a bug? If you agree, then we can:
* Change this to a blocker for 2.0
* Update all failing unit tests.
** I propose to do this in a single PR. It would be great to get help with
fixing the unit tests via PRs sent to my PR.
** Alternatively, we could split up this work by creating a temporary
{{private[spark] def brokenTreeAggregate}} method to be used for unit tests not
yet ported to the fixed treeAggregate. But I'd prefer not to do this since we
will want to backport the fix.
* Backport to all reasonable versions. This will be painful because of unit
tests.
Currently, I'm testing StandardScaler a little more carefully to check its
results.
> Update RDD.treeAggregate not to use reduce
> ------------------------------------------
>
> Key: SPARK-14408
> URL: https://issues.apache.org/jira/browse/SPARK-14408
> Project: Spark
> Issue Type: Bug
> Components: ML, MLlib, Spark Core
> Reporter: Joseph K. Bradley
> Assignee: Joseph K. Bradley
> Priority: Critical
>
> **Issue**
> In MLlib, we have assumed that {{RDD.treeAggregate}} allows the {{seqOp}} and
> {{combOp}} functions to modify and return their first argument, just like
> {{RDD.aggregate}}. However, it is not documented that way.
> I started to add docs to this effect, but then noticed that {{treeAggregate}}
> uses {{reduceByKey}} and {{reduce}} in its implementation, neither of which
> technically allows the seq/combOps to modify and return their first arguments.
> **Question**: Is the implementation safe, or does it need to be updated?
> **Decision**: Avoid using reduce. Use fold instead.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
