search

glmcdona / LuxPythonEnvGym

Matching python environment code for Lux AI 2021 Kaggle competition, and a gym interface for RL models.
MIT License
73 stars 38 forks source link

[feature request] use legacy kind of agents to run on the new environment #97

Open vitoque-git opened 2 years ago

vitoque-git commented 2 years ago

Use legacy kind of agents to run on the new environment.

It seems that what you want to achieve is not supported, the structure is different. The main problem as I see it is that the observation state is constructed differently, so the "new" state can not be directly given to the "legacy" agent. But even if this is resolved, still need to write wrappers to fit the structure. Maybe it is worth doing this work. You can open an issue on the github.

https://www.kaggle.com/c/lux-ai-2021/discussion/276419#1535771

glmcdona commented 2 years ago

Agreed. Writing a wrapper to the traditional agent.py is a great idea. That way these agents can be ported over without any work.

I'm working on the replay output feature right now, and I'll work on this next unless someone else takes on this work.

vitoque-git commented 2 years ago

Any news on this? I know you guys are really busy, but it would be awesome to have something

glmcdona commented 2 years ago

@vitoque-git , sorry for the delay. Could I learn a bit more about what you are looking for? Do you have an official format python agent, and you'd like to simulate games faster to measure success rate faster or similar? Or do you have an official format python agent structured for RL with the kaggle environment wrapper, and you'd like to port it over?