Difference fitrkernel and fitrsvm

Hello,

I'm looking at the different fitr-models and I'm wondering where the difference is between the default fitrkernel and fitrsvm with gaussian kernel. Both have the same hyperparameters. Fitrkernel is a gaussian kernel model, that uses an svm as a linear regression model and fitrsvm is an svm with a gauss kernel. Isn't that redundant?

Furthermore I do not understand the exact function of the hyperparameter "KernelScale" in both models. Are there any papers explaining the parameter used in Matlab?

Best regards,

Dimitri

Answer 1

[Don Mathis]

The basic difference is that fitrsvm fits an exact SVM model, in the sense that it uses the exact kernel function and solves the "dual" problem. fitrkernel solves the "primal" problem using an explicit finite-sized feature space, which results in an approximation of the kernel function. For large datasets, the kernel approximation can be much faster and give good enough results.

According to this Doc page,

"The software divides all elements of the predictor matrix X by the value of KernelScale. Then, the software applies the appropriate kernel norm to compute the Gram matrix."

[antlhem]

Could take a look into my question? https://uk.mathworks.com/matlabcentral/answers/842800-why-matlab-svr-is-not-working-for-exponential-data-and-works-well-with-data-that-fluctuates?s_tid=prof_contriblnk