"Support Vector Clustering"

ftgo

Hi fellows,

Im trying to use SVClustering algorithm to perform some clustering tasks. However, this operator seems to not work as expected. Some issues I have noticed:

1) If we try to cluster the Iris dataset, the output is quite different from the expected one (for example, try this with KMeans with 3 clusters).

2) If we try with different parameter combinations, it does not seems to work either.

3) Trying to switch the kernel type it does not work. Looks like other than default one are not implemented.

4) What parameter represents the 'q' (the scale parameter of the Gaussian kernel) cited on the original paper (link: http://portal.acm.org/citation.cfm?id=944807) ?

Thanks in advance,
Felipe.

ftgo

Hi all,

Does anyone can help me on this issue?

land

Hi,
I will help you as soon as you explain what you exactly mean by

1) Incorrect clustering? Whats a correct clustering?

2) What does it mean "does not work"? Crashes the program?

3) What does it mean "does not work"? Crashes the program?

4) It's the gamma I guess.

Greetings,
  Sebastian

brianbaker

I don't have any trouble running it - it doesn't crash.  I have been able to get 1 or 2 clusters, but not 3, from the iris set.

It would be helpful to know how the parameters p, q, c map into RapidMiner.  (It looks like gamma is kernel specific??)