Closed willwerscheid closed 3 years ago
using a mixture of normals for npmle is probably slower than just using a discrete grid of point masses? probably worth coding up the discrete grid directly? (in ash i only used the mixture of normals/uniforms to avoid doing the special case....)
good point! i don't think the gains would be terribly significant but it shouldn't be too hard to implement (really 2-3 lines I think). and i think i have to make a couple of changes anyway because ebnm_normal_mix_workhorse
previously assumed a scalar "mode" parameter whereas the way this is implemented is like a vector of modes.
right now it calls ebnm_ash but why not use the faster function