Cross-Validation returns different model

mark42

Hello,

I am using RapidMiner Studio 7.2.002 and recognized something strange after training libSVM (nu-SVC, linear kernel) classifier within X-Validation and outside X-Validation (with the same parameters): the output models are different! According to the documentation of the X-Validation operator the output model is trained on the whole example set, which would be the same as just using the libSVM training.

 

Do I miss anything or why are these models different although the operators, parameter and data are the same?

 

Best
Mark

 

 

MartinLiebig

Hi Mark,

 

the only way i can imagine that this one happens is that their is a randomness in the algorithm or the starting points of the optimization make a difference. I've tested it on sonar and saw no difference - but that might not be that meaningful.

 

Are you doing some kind of preprocessing inside x-val which might be different?

 

 

~martin

mark42

Hi Martin,

 

I don't use any preprocessing inside the X-Validation: only libSVM, Apply Model and Performance Classification (Accuracy).  The data and parameters are the same.

 

Best

Mark