change the rbf in SVM

Hello,

I am using the command

cl = fitcsvm(X,Y,'KernelFunction','rbf',...
  'BoxConstraint',Inf,'ClassNames',[-1,1]);

in my program. How do I go about changing the 'rbf'?

Answer 1

[Walter Roberson]

Just change it to your desired value.

https://www.mathworks.com/help/stats/fitcsvm.html#input_argument_namevalue_KernelFunction

[monkey_matlab]

I tried

cl = fitcsvm(X,Y,'KernelFunction',2, 'BoxConstraint',Inf,'ClassNames',[-1,1]);

but I got the error:

Error using classreg.learning.modelparams.SVMParams.make (line 217) You must pass 'KernelFunction' as a string.

[Walter Roberson]

That is correct, the parameter you give after the 'KernelFunction' keyword needs to be a string naming the function that will be called by the kernel.

What function are you trying to use as your kernel?

[monkey_matlab]

Hello, I wanted to use 'rbf' with different sigma values. I also tried:

cl = fitcsvm(X,Y,'KernelFunction','rbf', 'rbf_Sigma',2, 'BoxConstraint',Inf,'ClassNames',[-1,1]);

but that also did not work.

Thanks for your time and help!

[monkey_matlab]

Hello, the fitcsvm help does not tell me how to vary the sigma. I tried

cl3 = fitcsvm(X,Y,'KernelFunction','rbf', ...
'Standardize',true,'BoxConstraint',10,'ClassNames',[-1,1]);

but do not yet know how to add and then vary sigma. Thanks!

[Walter Roberson]

Create your own RBF routine that has a sigma built in to it.

Or possibly the KernelScaling might be what is needed, but the documentation for how that all worked is long and it was not clear to me how the various aspects work together.

[Hassan Raza]

How to make the code implementation Multiquadric RBF on 1d heat equation.

Answer 2

[Jhon Pilataxi]

Here someone said how to change:

https://es.mathworks.com/matlabcentral/answers/328796-gaussian-kernel-scale-for-rbf-svm