[
https://issues.apache.org/jira/browse/SPARK-13226?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15136118#comment-15136118
]
holdenk commented on SPARK-13226:
---------------------------------
Further poking shows that they did both depending on the kmeans algorithm used,
so will use the tolerance for tolerance based.
> MLLib PowerIteration Clustering depends on deprecated KMeans setRuns API
> ------------------------------------------------------------------------
>
> Key: SPARK-13226
> URL: https://issues.apache.org/jira/browse/SPARK-13226
> Project: Spark
> Issue Type: Improvement
> Components: MLlib
> Reporter: holdenk
> Priority: Trivial
>
> The current MLLib PowerIteration clustering implementation sets the number of
> runs inside of the kmeans call to 5 (apparently arbitrary). This should
> likely be replaced with a specific tolerance.
> The reference implementation also appears to use a tolerance, so this would
> also be moving closer to the reference implementation (
> http://www.cs.cmu.edu/~wcohen/ )
--
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
