You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The text was updated successfully, but these errors were encountered:
dylanhmorris
changed the title
Should pyrenew's sampler backend be numpyro's bundled NUTS sampler
Should pyrenew's default MCMC sampler backend be numpyro's bundled NUTS sampler?
Oct 1, 2024
My preference stands with numpyro over blackjax. Not sure where I am with bayeux given that I have not done much with it. Having other JAX based sampling options available seems good, but for the default I presently think deferring to numpyro seems best (this could ofc change if I learned bayeux was superior in some manner).
I did a bunch of digging into within chain parallelization. My current takeaway is that jax can do it and maybe some of the other backends can but numpyro can't. If that actually is the case it puts more weight on being backend agnostic at least sometime in the future IMO.
I really can't see how or why that would be the case though so very keen to see evidence to the contrary from people who have actually fit a numpyro model and so might be better informed.
Or should the standard pattern be to specify the model in
numpyro
but pass it off tobayeux
orblackjax
etc for inference?@CDCgov/pyrenew-devs
See also #361 #151
The text was updated successfully, but these errors were encountered: