There doesn't seem to be any reason why we couldn't make an RFF or ORF-based Gaussian spectral mixture basis, and it would probably be faster for small to moderate input dimension than the Fast Food based implementation. We could share quite a bit of code between the implementations as well (?).
There doesn't seem to be any reason why we couldn't make an RFF or ORF-based Gaussian spectral mixture basis, and it would probably be faster for small to moderate input dimension than the Fast Food based implementation. We could share quite a bit of code between the implementations as well (?).