Cooperative Matrix Game Environment for GOAT

Simple environment for experimenting with cooperative matrix game.

Author: Paresh R. Chaudhary (pareshrc@uw.edu)

Overview

This repository provides a framework for creating and experimenting with cooperative game. It includes tools for generating game layouts, calculating payoffs, and simulating agent interactions.

Project Structure

The project is organized into the following directories:

• env/: Contains the game environment implementations and matrix game logic
• policy/: Houses policy implementations for different agents
• trainer/: Contains training algorithms
• utils/: Utility functions and helper modules
• runner/: Scripts for running experiments and simulations
  • goat.py: Implementation of the GOAT algorithm
  • goat_search.py: Hyperparameter search routines for GOAT
  • comedi.py: Implementation of the comedi algorithm
  • mep.py: Implementation of the MEP algorithm
  • trajedi.py: Implementation of the trajedi algorithm
  • sp.py: Implementation of the SP (Self-Play) algorithm
• results/: Directory for storing experiment results and outputs

Installation

# Clone the repository
git clone https://github.com/pareshrchaudhary/cooperative_matrix_game.git
cd cooperative_matrix_game

# Create and activate conda environment
conda create -n cmg python=3.11
conda activate cmg

# Install dependencies
pip install -r requirements.txt

# Install the package in development mode
pip install -e .

# Set the PYTHONPATH (required for imports to work correctly)
export PYTHONPATH=$PYTHONPATH:/path/to/parent/directory
# For example, if your package is at /Users/username/dev/cooperative_matrix_game:
export PYTHONPATH=$PYTHONPATH:/Users/username/dev

Usage

After installation and setting the PYTHONPATH, you can run individual algorithm files to run experiments and generate results.:

python runner/mep.py

or

1. Generate CoMeDi data to train VAE.

python runner/comedi.py

2. Train VAE.

python trainer/vae_trainer.py

3. Train GOAT.

python runner/goat.py

or

Cite

@article{chaudhary2025improving,
  title={Improving Human-AI Coordination through Adversarial Training and Generative Models},
  author={Chaudhary, Paresh and Liang, Yancheng and Chen, Daphne and Du, Simon S and Jaques, Natasha},
  journal={arXiv preprint arXiv:2504.15457},
  year={2025}
}