0
$\begingroup$

From another answer here, I was linked to a paper:

Linear kernel and non-linear kernel for support vector machine?

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.141.880&rep=rep1&type=pdf

So to clarify, I think that the linear cannot be better than rbf if you have the right parameters and that linear is just a worse version of rbf. Linear comes from rbf. Is this correct? Also, in Matlab I see sigma for kernel scale but not gamma. Are gamma and sigma the same?

$\endgroup$
1
$\begingroup$

I think that the linear cannot be better than rbf if you have the right parameters

Yes, that is shown in the linked paper.

and that linear is just a worse version of rbf. Linear comes from rbf. Is this correct?

Linear kernel is the simple case of not using the kernel trick and you can get to it as a special case of the RBF kernel.

Also, in Matlab I see sigma for kernel scale but not gamma. Are gamma and sigma the same?

The Gaussian (RBF) kernel should have 1 parameter (called gamma or scale). The linear kernel has no parameters.

$\endgroup$
  • $\begingroup$ Re: the last point, there are unfortunately many parameterizations of the Gaussian RBF kernel in relatively widespread use: the ones I know of are all of the form $\exp( - A \lVert x - y \rVert^2 )$ where $A$ can be $\gamma$, $\frac{1}{2 \sigma^2}$, $\frac{1}{\sigma^2}$, $\frac{1}{\tau}$, among other variations. $\endgroup$ – Dougal Aug 31 '17 at 20:29
  • $\begingroup$ Thanks! Can we say that there is 1 real-valued parameter? $\endgroup$ – Matteo T. Aug 31 '17 at 20:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.