SVM learner parameter

[degapifa]

Hello,

I am trying to optimize the performance of the SVM learner using a polynomial kernel

• Polynomial (has three parameters)
  • Power
  • Bias
  • Gama

could someone kindly explain to me, the concept of these three terms?

[aworker]

Hello @degapifa

Welcome to the KNIME forum.

This previous post by @ScottF may help to answer your question:

What do "SVM Learner" parameters mean?

Hmmm, it seems the links in our node description for the SVM learner are out of date. Sorry about that. Instead, you might check out this article, which describes some of the commonly used kernel functions in SVMs (and their parameters). It also touches briefly on what happens to the boundaries when those parameters are adjusted, in the case of RBFs.

Hope this is of help

Best regards,

Ael

[degapifa]

thanks for the answer, I can’t understand the parameter: power corresponds to p in the frist formula?

[ScottF]

p is the power, yes. You can read more about the effects of gamma on various kernels (including polynomial) in this recent paper:

https://www.researchgate.net/publication/344458945_The_effect_of_gamma_value_on_support_vector_machine_performance_with_different_kernels

[degapifa]

thanks for the answer. It is a really good job to increase understanding. One last thing that i have noticed: i should optimize the cost C. In the SVM node in knime would it be equivalent to overlapping penalty?

[ScottF]

I believe so, yes (although I should verify that internally).