How to customize SVM kernel parameters in Matlab

Hi

I want to ask about how to customize SVM kernel parameters in Matlab.

For ex: I have linear kernel, polynomial kernel, and RBF kernel with:

C = 0.2;
gamma = 0.8;
r = 0.05;
d = 3;

My question is how do I customize it using svmstruct.

1. linear: ??

2. polynomial: ??

3. RBF:

svmstruct = svmtrain(data, groups, 'Kernel_Function', 'rbf', 'RBF_Sigma', 0.2, 'BoxConstraint', 0.8);

CMIIW about how to customize RBF kernel.

I'd really appreciate if anybody could help my research. Thanks.

Regards

Ikra

Answer 1

[Ikra89]

anybody can help? :)

Answer 2

[Ahmed]

Documentation given by

   help svmtrain

should give you all information you need. What is the meaning of your variable r?

1. the linear kernel has only parameter C

   svmstruct = svmtrain(data, groups, 'Kernel_Function', 'linear', 'BoxConstraint', 0.2);

2. polynomial has parameters C and polyorder

   svmstruct = svmtrain(data, groups, 'Kernel_Function', 'polynomial', 'polyorder',3,'BoxConstraint', 0.2);

3. RBF

you might have mixed up the constants C and gamma in your code

[Ikra89]

and what about "r parameter" ?

as i know:

linear has C parameter

polynomial has C, gamma, r, d

rbf has C and gamma

[Ahmed]

It looks like Matlab's svmtrain function only supports homogeneous polynomial kernels. You can however, define an arbitrary kernel function and pass a handle of it to svmtrain. See section kernel_function -> @kfun in the documentation for description and example.

[Ikra89]

can you give me an example how to use @kfun

i have tried many times but still error

[Ahmed]

Using an anonymous function of a linear kernel would be:

svmstruct = svmtrain(data, groups, ...
'Kernel_Function',@(x1,x2) x1*x2','BoxConstraint', 0.2);

Note, that your kernel function must lead to a valid kernel.

[behruz]

hi the combination of C and gamma how is implemented? in fitcsvnm or svmtrain we dont have any gamma factor how is the impact of it apllied to svm?

[behruz]

any comment?