Incorrect computation of wavelets

[daniel-unyi-42]

First of all, thank you for sharing your code, it was extremely useful for my project.

I believe the function 'weight_wavelet_inverse' does not work as intended. Since you alter the eigenvalues in 'weight_wavelet' permanently, in 'weight_wavelet_inverse' you compute the exponential on the altered eigenvalues (basically you do an exp of exp thing). The product of the wavelet and its inverse should always be identity.

Funny enough, your half-incorrect code gives better results than the corrrect one. I guess it's because other kernels can perform better than the heat kernel?