search

jeremiecoullon / SGMCMCJax

Lightweight library of stochastic gradient MCMC algorithms written in JAX.
https://sgmcmcjax.readthedocs.io/en/latest/index.html
Apache License 2.0
95 stars 8 forks source link

Requirement clarification #36

Closed ColCarroll closed 2 years ago

ColCarroll commented 2 years ago

I'm not sure if it is a design decision, but if the library's only "real" dependency is jax and jaxlib (you also use tqdm, but I doubt that's "structural"). If you plan to keep it like this, I might call that out in the installation instructions -- when looking for minimum and maximum python versions, I ended up in your setup.py, then in jax's (it is Python>=3.7!)

Anyways, if you might bring on more, feel free to ignore, but that might at least deflect any installation problems over to JAX.

jeremiecoullon commented 2 years ago

I might call that out in the installation instructions

I don't quite understand how to do this. Do you mean changing in setup.py the line install_requires=["jax", "jaxlib", "tqdm"], to install_requires=["jax", "jaxlib"], ?

ColCarroll commented 2 years ago

Oops, sorry, this was very unclear. I meant that it is cool that jax is the only real requirement, and I might highlight that in the README. Something like (my addition in bold)

SGMCMCJax is a lightweight library of stochastic gradient Markov chain Monte Carlo (SGMCMC) algorithms. The aim is to include both standard samplers (SGLD, SGHMC) as well as state of the art samplers while requiring only JAX to run.

ColCarroll commented 2 years ago

(this is not at all a requirement for the paper review either -- just a suggestion!)

jeremiecoullon commented 2 years ago

Done! :)