Multi-label (Support Vector Clutering) cite BenHur/etal/2001a.

Legend

Dears,

Happy New Year.
One of great masterpiece rapidminer.

When I have tested Support Vector Clustering operator in several ways, however, two-labeled clusters are always produced by rapidminer 5.0.

As you know, according to the BenHur's paper, there are multi-labeled results I've seen.
How can I make it? Is it possible with rapidminer 5.0 (community version)?


Kindly Regards,
Danny Seo.

land

Hi,
this should be possible, but it depends on your parameter settings. Perhaps you have to change some of the parameter values defining which clusters are outliers. Seems to me, that your operator only divides into noise and the rest...

Greetings,
  Sebastian

Legend

Hi, Mr. Sebastian Land.

Thanks for your reply.

Yes, you right. I had only cluster_1 and noise labels as a resut, although I was trying several parameter sets.
As my understanding Ben-Hur's SVC work, there are some parameters: C, q and p.
In RapidMiner, each parameter, C, q and p are related to convergence epsilon(KKT), kernel gamma and p respectively.

I hope that please give me any comments, expertises to improve SVC result.

Best Regards,
Danny.

Legend

Does anyone have an idea for this thread?

BR.