matlab - SVM using fitcsvm with custom kernel

Question

I am training a linear SVM classifier with the fitcsvm function in MATLAB using different kernels (with flag 'kernelfunction').

My training set consists of a list of 1x3 vectors, each one of them is labels 1,-1.

At first I was using a regular rbf kernel: exp(-gamma * pdist2(x1,x2)).

Now I want to change the distance function to a geodesic distance, which requires me to have 1x3 data points in the 'kernelfunction'.

Unfortunately, the data points that are transferred to the kernel function are automatically changing to 1x2 so I can't use the required distance function.

Does any one know how can I create a rbf geodesic kernel? I can't find a way to transfer the raw data points to the kernel function.

Thank You.

question from:https://stackoverflow.com/questions/65925148/svm-using-fitcsvm-with-custom-kernel

Answer 1

Waitting for answers