Skip to content

Conversation

@jakegrigsby
Copy link
Collaborator

@jakegrigsby jakegrigsby commented Jun 4, 2025

A messy alternate timeline version of poke-env that maintains the original gymnasium API:

  • OpenAIGymEnv (+ ability to swap in custom Players)
  • Rewards functions that take last_battle and current_battle as input (+ faster deepcopy fix)
  • Removes "observation" system that slows down deep copy and could be handled on the RL side

Starts from poke-env v0.8.3 (the last to include an OpenAIGymEnv), goes backwards to restore the working version of the reward function, and then tacks on some very minor changes that come up in early generations.

Tries to bring key fixes/improvements since v0.8.3 that weren't related to gymnasium (new protocol-->request order, leaving rooms after the battle)... but I'm probably missing some of them.

I stopped short of changing sim protocol (AbstractBattle) to support early gen edge cases, or fixing the currently broken PP counter. These are left as they were during the Metamon experiments, and fixes will be done in custom AbstractBattle/Players.

jakegrigsby and others added 30 commits June 3, 2025 15:15
…own deepcopy; revert to old save_replay system
… keeping, and makes reward functions much easier to write
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants