/MeltingPot-Tess-v1

Tess-v1 Pure Reinforcement Learning Solution for Melting Pot Contest ( AIcrowd )

Primary LanguagePythonMIT LicenseMIT

Tess-v1: Pure RL Solution for NeurIPS Melting Pot Contest

Tess-v1 is a reinforcement learning solution developed for the NeurIPS Melting Pot Contest. The repository contains various training and model files, with models/model_v4_cont.py being applied in all the final experiments.

Table of Contents

Introduction

Tess-v1 is a reinforcement learning solution tailored for the challenges posed in the NeurIPS Melting Pot Contest. The repository encompasses various training and model files, with particular emphasis on models/model_v4_cont.py, which is employed in all the final experiments.

Trainings

  • Harvest: Trained using train_multitask_harvest.py
  • Clean Up: Trained using train_multitask.py
  • Prisoners: Trained using train_multitask.py
  • Territory Rooms: Trained using train-v6.py

Note: For reproducibility, it is essential to configure the parameters in configs.py. The necessary configuration details will be shared in the next commit.

Prerequisites

Before running Tess-v1, ensure you have the following prerequisites installed:

  • PyTorch (version >= 2.0.1)
  • Stable-Baselines3
  • Gym (version 0.21.0)
  • OpenCV-Python
  • NumPy
  • WandB (Weights & Biases)
  • MeltingPot Suite

You can install these dependencies using the following:

pip install torch>=2.0.1 stable-baselines3 gym==0.21.0 opencv-python numpy wandb dm-meltingpot

License

This project is licensed under the MIT License

Tess-v1 RL Solution

Version		: 1
Released 	: 2023-12-26
Author		:: Utku Erodoğanaras
Contacts 	: utkuearasbusiness@gmail.com

Citation

Please site this repository in this format.