[
https://issues.apache.org/jira/browse/MATH-1165?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14202622#comment-14202622
]
Pashutan Modaresi commented on MATH-1165:
-----------------------------------------
I think the case that u mentioned is not the only one! What if one of the
initial random centroids is exactly the same as the one of the data points?
> Rare case for updateMembershipMatrix() in FuzzyKMeansClusterer
> --------------------------------------------------------------
>
> Key: MATH-1165
> URL: https://issues.apache.org/jira/browse/MATH-1165
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.3
> Reporter: Pashutan Modaresi
> Priority: Minor
> Labels: easyfix
> Fix For: 3.4
>
> Original Estimate: 1h
> Remaining Estimate: 1h
>
> The function updateMembershipMatrix() in FuzzyKMeansClusterer assigns the
> points to the cluster with the highest membership. Consider the following
> case:
> If the distance between a point and the cluster center is zero, then we will
> have a cluster membership of one, and all other membership values will be
> zero.
> So the if condition:
> if (membershipMatrix[i][j] > maxMembership) {
> maxMembership = membershipMatrix[i][j];
> newCluster = j;
> }
> will never be true during the for loop and newCluster will remain -1. This
> will throw an exception because of the line:
> clusters.get(newCluster)
> .addPoint(point);
> Adding the following condition can solve the problem:
> double d;
> if (sum == 0)
> d = 1;
> else
> d = 1.0/sum;
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
