facebookresearch / nevergrad

A Python toolbox for performing gradient-free optimization
https://facebookresearch.github.io/nevergrad/
MIT License
3.96k stars 354 forks source link

missing gym in requirements #568

Open rpanai opened 4 years ago

rpanai commented 4 years ago

Inside this file https://github.com/facebookresearch/nevergrad/blob/master/nevergrad/functions/rl/envs.py you are importing gym but this is not listed as requirements.

jrapin commented 4 years ago

Hi, This is on purpose, most users only use the optimizers so the basic installation does not include gym which is only needed in the benchmarks. To install the extra requirements you can use pip install nevergrad[bench] or pip install nevergrad[all]

rpanai commented 4 years ago

The problem here is that I got an error trying to build an conda-forge package for this library. Eventually I can remove the benchmarks.

jrapin commented 4 years ago

I'm not very familiar with conda-forge, is there no way to have extra requirements only installed when needed? Most users would only use the optimizers, which do not require much, while the testbeds include way more dependencies, including huge ones like pytorch :s I would definitely not want to include this for the basic installation

rpanai commented 4 years ago

@jrapin I'm going to investigate. I think it will be great to add to conda-forge. fbprophet is already there and I recently added hiplot.

jrapin commented 4 years ago

Cool ;) let us know how it goes