Playing Snake with Reinforcement Learning

This is a simple implementation of the classic game Snake, where the player controls a snake that grows in length as it eats food. The game ends when the snake collides with itself or the wall. The goal of this project is to train an agent to play the game using reinforcement learning.

Getting Started

Prerequisites

This project requires Python 3.6 or later. The following packages are also required to be installed via

pip install -r requirements.txt

Usage

To train the agent, run

python agent.py

with the line train() uncommented in the main() function. To test the agent, run the same command with play_round(filename) uncommented instead. The filename argument should be the .pth file containing the trained model.

To run the Human Playable version of the game, run

python user_playable_game.py

Graph showing the average score per episode

You can clearly see that the agent is learning to play the game. The average score per episode increases as the agent learns to play the game better. The agent is able to achieve an average score of 28.53 after 691 episodes. The max score was 79 while the min score was 0.

Example of the agent playing the game