CODEWITHSUNDEEP
matlabmachine-learningparametersstatisticssvm
ganninu93
ganninu93

Reputation: 1601

Setting gamma value for RBF kernel in templateSVM

According to this document (and several others), the RBF kernel is defined by the equation below, and can be adjusted by varying the parameters C and gamma. enter image description here

The documentation for svmTemplate defines the RBF kernel without a gamma value as shown below.

enter image description here

Why does svmTemplate not use gamma? Is there a way of specifying it?

Upvotes: 0

Views: 622

Answers (1)

Raff.Edward
Raff.Edward

Reputation: 6524

I'm guessing it is because it is essentially rolled into the "KernelScale" option in the documentation. It indicates that it divides all the values in the data by a constant, which allows you to essentially set the gamma parameter that way. You can see that ||xz-yz|| = ||x-y||*z, and so obtains the equivalent impact (so long as you select the KernelScale value appropriately). The documentation says the scale is selected heuristically by default, and there do exist heuristics for the RBF kerne's parameters.

It's not the way I would have coded such a system, but it should work.

Upvotes: 1

Related Questions