Closed simeoncarstens closed 1 year ago
I'm not sure this makes sense (anymore). With the (albeit, rudimentary) support for Stan and PyMC, no one would ever want to use binf
. Furthermore, in the meantime, there was a hand-crafted, minimal HMC sampler added to Chainsail.
I once had my own subpar shot at a probabilistic programming library when I didn't even know that that's a thing. Anyways, I wrote
binf
(https://github.com/simeoncarstens/binf) for my research, which implements interfaces for posterior distributions, likelihoods etc., but also a Gibbs and a HMC sampler. We could think about using the latter in rexfw by depending on thebinf
library instead of whipping up our own sampler (which would essentially be a copy-and-paste of thebinf
one.Discussion, because I'm not sure we want yet another dependency with all of its luggage (
csb
and thusscipy
) for a class which is essentially just 20 lines of code.