Weka does not use the predefined centroids

[kno10]

But will always use random centers.

[zoq]

I think weka doesn't provide an option to set initial centroids, which I think is crucial for a valid comparison; I guess we could patch the weka buildCluster method. Maybe I missed something?

[rcurtin]

I had vague memory of handling this a long time ago (like 5 years now). I thought that Weka did allow some way to do it, but I can't remember for sure. Patching a method seems like a reasonable solution to me if we need to, but I'll dig into the docs and code for Weka a little bit more.