Snake using RL

Reinforcement Learning for the game snake using PyTorch and PyGame. Tested on Ubuntu 20.04.

Install

You can install the source using:

git clone git@github.com:TimReX-22/snake-rl.git

The required dependencies can be install by using pip:

pip install -r requirements.txt

Run

To run a model, you will have to create or modify parameters.json. The JSON file should include the following parameters:

{
  "max_memory": 100000,
  "batch_size": 1000,
  "lr": 0.001,
  "gamma": 0.9,
  "epsilon": 100,
  "model_name": "model/model_score_36_1.pth"
}

model_name should be the relative path to the model file. If you want to train a new model, do not include a model_name in the JSON file. Then you can run the agent using:

python agent.py parameters.json

If you simpy want to play the snake game using your keyboard, your can run the game by:

python snake_pygame/run.py

Incase you get a ModuleNotFoundError, you can most likely resolve it by updating PYTHONPATH with the path to the repo directory:

export PYTHONPATH=$PYTHONPATH:<path to repo>

Credits

The Inspiration for this project stems from this YouTube Video