measure support vector clustering validation

njasaj

Hi,
I need to know if it is possible to measure validity of clustering by support vector clustering when there is no idea about the number of clusters? and also how to select the proper parameters for support vector clustering in RapidMiner? I mean that is there any operator that measure the quality of support vector clustering in RapidMiner ?

Thanks for your replies.

dudester

This is more a statistical question than a RapidMiner issue, although I am not an expert in this area.  I do know in using other mining products, that estimating initial clustering parameters is oftentimes as much art as science, and it is helpful to have content background as you explore.  There are various techniques you can use to help you with this; hope this helps

http://www.stanford.edu/~hastie/Papers/gap.pdf
http://www.ncbi.nlm.nih.gov/pubmed/18270084

phani9779

I have the similar question. I followed the links provided, but it is more in a theoritical form. Is there any operator in Rapidminer to select the optimum clusters based on parameters such as entropy, BIC, AIC etc.

 

Thanks in advance

sgenzer

hi @phani9779 - have you tried just using Auto Model? It will optimize your cluster performance very nicely.

 

phani9779

Hi Mr. Scott,

 

Thanks for your kind reply. Is it possible to conduct SVC clustering in Automodel?

sgenzer

from @IngoRM:

 

"In theory [SVC clustering] sounds good since it can handle non-linear separations better.  In practice, this does not matter that much in large number of dimensions though.  It is also terribly slow and hard to tune.  Not a big fan myself.  So it should come as no surprise that AM is not doing this."

 

Scott

phani9779

Hi Scott,

 

My thanks for your kindness. I am slowly realizing that SVC sounds better in theory but not in practice.

 

I have some issue with the license activation. Can you please get through this? I have a educational license valid till september, 2019. But after installing the license key, the activation wizard showing the trial duration.

tftemme

Hi @phani9779, 

 

Don't worry about the license. The license system always takes the license with the most privileges, which is in this case the Trial license (it corresponds to a Large license). When the trial expires it will default back to your educational license.

 

Best wishes and happy mining

Fabian

phani9779

Thank you :smileyhappy: