remove baselines from ctoybox

toybox-rs / Toybox

The Machine Learning Toybox for testing the behavior of autonomous agents.

http://toybox.rs

27 stars 12 forks source link

remove baselines from ctoybox #78

Open etosch opened 5 years ago

etosch commented 5 years ago

TODO: call toybox from baselines, not baselines from toybox. I'd like to completely excise baselines from our repo.

jjfiv commented 5 years ago

https://github.com/toybox-rs/baselines

jjfiv commented 5 years ago

New plan: use a pip install of "stable-baselines". Challenge: all our hacks to get Toybox environments inside require code editing or hotpatching.

[ ] hotpatching atari-step (This is really an openai/gym hack).
[ ] get_turtle(env) to find the toybox or the innermost env.
[ ] SampleEnvs needs a place to live.
[ ] Toybox outputs greyscale by default; need to skip cv2.cvtColor stuff, or switch to RGB by default
[ ] Need to add a TimeLimit environment (This is really an openai/gym hack).

jjfiv commented 5 years ago

Stable Baselines' implementation of acktr is inefficient; getting ~30FPS on gypsum. https://github.com/hill-a/stable-baselines/issues/196

kclary commented 3 years ago

@etosch judgement: we can remove baselines etc but the sample tests depend on this for regress models in unit tests. sample tests already live over in https://github.com/toybox-rs/openai-baselines-envs, so we can pull out both baselines and the sample tests.