Snake-RL

A self-learning pygame implementation of Snake using Reinforcement Learning.

Below you can find an example of an episode performed by a trained agent.

This repository consists of two major components:

• A python package consisting of a basic Snake game built on pygame, completed with OpenAI gym wrapper.
• Reinforcement learning experiments using a simple DQN (Deep Q-Network) in which agents learn to play this Snake game

Setup

Start by creating the conda environment required to run the application.

conda env create -f environment.yml

Now that the environment has been build, you can install the snake_gym package by running:

pip install snake_gym/

Play the game

Now that you have built the application, you can play it simply by using the following command:

python -m snake_gym

Reinforcement Learning experiment

You can start the training of the RL experiment by running:

python RL/train.py

Alternatively, you can simply run inference of the trained agent by running:

python RL/inference.py