In #742 we switched to having a single Matern kernel for all types. This reduces code complexity but might come with some speed issues as for specific kernels we could pre-compute some elements of the spectral density (as @sbfnk did in #375).
As #742 is already complicated I don't think we should address this there but we should explore to see if the speed/technical debt trade-off is worth it.
In #742 we switched to having a single Matern kernel for all types. This reduces code complexity but might come with some speed issues as for specific kernels we could pre-compute some elements of the spectral density (as @sbfnk did in #375).
As #742 is already complicated I don't think we should address this there but we should explore to see if the speed/technical debt trade-off is worth it.