What is the description of the algorithm that MATLAB uses to determine 'auto" Kernel scale in SVM Classification Learner?

I am looking for a pointer (maybe a paper?) that describes the "heuristics subsampling" algorithm that MATLAB uses to determine the KernelScale in their Classification Learner app -- specifically SVMs using an RBF Kernel with the KernelScale = 'auto' option.

[sh10101]

Hi,

I am wondering if you had any luck finding an explanation for the 'auto kernel scale' setting in the "templateSVM" function documentation, as suggested by David?

thank you,

[PR]

Hi, I am also looking for this information. Any luck?

Answer 1

[David Ding]

Hi,

Perhaps you can check out the six papers listed under the "templateSVM" function in MATLAB documentation:

https://www.mathworks.com/help/stats/templatesvm.html#References

[Zeynab Mousavikhamene]

I have the same question as r00sk0. The answer you provided are refernces for templateSVM which I dont think is related to the question asked above. We wan to know what ar the "heuristic subsampling" steps? description to this procedure.