[
https://issues.apache.org/jira/browse/MAHOUT-564?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12979572#action_12979572
]
Sean Owen commented on MAHOUT-564:
----------------------------------
I think this is due to the fact that the patch changed another test method to
use a different threshold, and changing it back works. Frank do you know if
that change from 0.001 to 1.5 in the test class was unintentional?
I would like to do a bit more with this patch. KMeansClusterer was indeed
slightly confused since it is in some cases initialized from a Configuration
and sometimes not. In the former case it is supposed to get its threshold
internally but in the other case it isn't, and there was a problem there. Might
be more straightforward to have one approach, so I did that locally.
> KMeansClusterer does not use distanceThreshold parameter in
> testConvergence(Iterable<Cluster> clusters, double distanceThreshold) method
> ----------------------------------------------------------------------------------------------------------------------------------------
>
> Key: MAHOUT-564
> URL: https://issues.apache.org/jira/browse/MAHOUT-564
> Project: Mahout
> Issue Type: Bug
> Components: Clustering
> Affects Versions: 0.4
> Reporter: Frank Scholten
> Fix For: 0.5
>
> Attachments: MAHOUT-564-without-star-imports.patch, MAHOUT-564.patch
>
>
> While running KMeansClusterer#runKMeansIteration sequentially I noticed that
> the distanceThreshold parameter is not used. The convergenceDelta field is
> used instead. However, it's initialized at 0 and only set when creating a
> KMeansClusterer with a Configuration object, which is only used in a
> MapReduce setting.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
