Hyper-parameter optimization for a custom kernel SVR with Bayesian optimization?

Hi everyone. I want to optimize hyper-parameters for a SVR in Matlab using Bayesian optimization toolbox, but for a custom Kernel not for the default kernels. Because in Matlab help it says that for a custom kernel you have to define kernel scale within kernel. Has anybody experience with that problem? I want to define my own kernel and then to optimize hyper parameters for a regression problem using support vector machines. With default kernels it works very well, but since there is any example it is a bit hard to understand.

Answer 1

[Don Mathis]

You'll need to use the bayesopt function to do that. There is an example of support vector classification on this page: http://www.mathworks.com/help/stats/bayesian-optimization-case-study.html. Maybe you can adapt it to your regression problem.