# heuristics for gamma in rbf kernel

My question is a follow-up to this question: SVM rbf kernel - heuristic method for estimating gamma.

Basically, I want to find interesting values for gamma by first calculating the pairwise distance between a large number of samples with different labels. Then, I use the 1st, 2nd and 5th percentile as sigma and calculate gamma as: $$\gamma = \frac{1}{2\sigma^2}$$

Given that gamma changes how far each sample reaches and so the number of samples that influence the hyperplane, is the gamma that corresponds to the nth percentile going to change the hyperplane based on n% of the samples?