@nblumoe: I think the definitive answer to that is a listen to this episode http://www.thetalkingmachines.com/blog/2016/10/1/anglican-and-probabilistic-programming

I also think that https://github.com/uncomplicate/bayadera may be useful there in the future as well

@aaelony @nblumoe Bayadera roughly has similar use cases as Stan. Much faster on the models that I've tried. Needs to be documented, released, and polished...

cool thanks!

just checking in here after a felt millenium to ask a silly question and directly bump into a useful tip (about bayadera), cool

anyways, I'm tinkering around with some code that implements a multinominal nb classifier which is currently doing a lot of stuff based on simple clojure vectors

in particular, I'm sort of slicing my data matrix manually

this is taking a lot of time and I'm wondering if a change to matrix/select (vectorz-clj) is likely to be faster

I know it's a very broad question that is hard to answer without seeing concrete code