Why Logexp transformation is done for kernel parameters gap decay, match decay and order coefficients? #1
Description
What is the intuition behind going for Logexp transformation for kernel parameters w.r.t string kernel? Does it help in avoiding nan and matrix non-invertible problems?