/Pommerman-RLlib

A repo to train pommerman agents with RLlib

Primary LanguagePython

Pommerman-RLlib

Environment Setup

  • pommerman: download the playground and install the lib
cd playground
conda env create -f env.yml

To update pommerman environment

pip install .
  • RLlib & libs:
cd Pommerman-RLlib
pip install -r requirements.txt

Structure

  • models: build customize PyTorch models
  • demo: deploy the trained models for visualization
  • env: a pommerman wrapper to incorporate with RLlib
  • reward: an interface to build custom reward functions

Running Script

! Make sure you are in the root of this project

  • First set up the environment path
conda activate pommerman
export PYTHONPATH=$PYTHONPATH:.
  • then running the script
python script/A2C/Train_curriculum.py game_training
  • running the evaluation script to see your result
  • Do not forget to modified your model's path in script
python script/A2C/Train_curriculum.py game_training