AI Hoops!

Teach agents to play Double Dunk using DDQNs! (Paper)

Table of Contents

1. About AI Hoops
2. Getting Started
3. Installation
4. Usage
5. Contact
6. Acknowledgments

About AI Hoops

AI Hoops is a respository for training and executing models on the Dobble Dunk Atari Enviroment. It supports both single and multi-agent enviroments. It provides an api to easily train the model architecture on various hyperparamaters. This is fully compatible with other Atari games, and should just need replacing the environment name that is hard coded to be DoubleDunk.

Here is the Paper documenting our findings

(back to top)

Getting Started

Installation

Install the following prerequisites. Be sure to have auto rom install roms for both Gymnasium and PettingZoo if you want to use both.

• Pytorch
• Gymnasium
• PettingZoo
• Arcade-Learning-Environment
• AutoROM
• Installation

or, use the requirements.txt we provide by doing:

pip3 install -r requirements.txt

Note: This installs everything except Torch.

• In order to install Torch with cpu, do pip3 install torch
• In order to install Torch with CUDA you have to follow the instructions on the PyTorch website.

After installing, you must execute the following command to accept the AutoROM license if you want to get the DoubleDunk ROM from Autorom (do this in a regular terminal)

AutoROM -y

NOTE: Gymnasium doesn't work with Python 12.x, please use Python 11.x for the best performance.

(back to top)

Usage

To Train the model from scratch using preset hyperparameters (you can modify the hyperparameters inside this file)

   python3 train.py

To test the trained model and generate results: (update the file name inside to test out different models)

    python3 test.py

To to train a multi agent model:

    python3 train_multi_agent.py

(back to top)

Links

Project Link: https://github.com/anish-sahoo/AI-Hoops

Paper PDF - Final_Report_CS4100.pdf

Paper Repository - https://github.com/anish-sahoo/AI-Hoops-Paper

(back to top)

Contact

Anish Sahoo - [email protected]

Ethan Carpenter - [email protected]

Sana Ali - [email protected]

(back to top)

Acknowledgments

• A Deeper Look at Experience Replay
• Large Batch Experience Replay
• Atari Documentation
• Atari Double Dunk Environment
• All the journal articles are referenced in our paper

(back to top)

Thank you for reading!