search

salesforce / warp-drive

Extremely Fast End-to-End Deep Multi-Agent Reinforcement Learning Framework on a GPU (JMLR 2022)
BSD 3-Clause "New" or "Revised" License
465 stars 78 forks source link

support gymnasium >= 0.26? #86

Closed cboettig closed 1 year ago

cboettig commented 1 year ago

Thanks for sharing this impressive resource! I see the config currently restricts installation to before breaking changes were introduced into the gymnasium package in its version 0.26 release. Are there plans to migrate to supporting the latest version of the gym/gymnasium package? You are probably already familiar with the offical migration guide

Emerald01 commented 1 year ago

In the next release 2.5.0, we will remove this restriction. Our codebase is very weakly dependent on gym module. We used it to define our environment attributes and random seeding in our outer Python wrapper. Internally our code has no dependence on gym.

cboettig commented 1 year ago

Thanks @Emerald01 ! yup, noted that the gym module isn't really used here anyway; its just nice not to have to maintain a separate virtualenv for my warp-drive examples when doing any benchmark them against gym-based RL just avoid the conflict :-). sorry for the trouble and have really been enjoying getting my teeth into warp-drive so far!