Created a gym environment of the Snake game and implemented a Deep Q Network to play in it.
To view the agent in Snake with a 9x9 grid, run view.py
snake9x9.mp4
First run, training on 9x9 grid, around 3 hours with RTX 3060 Mobile:
Second run, continuing the training on 9x9 grid, around 5 hours with RTX 3060 Mobile (this took more time than the first run even though the second run played less games because the games were longer, as the snake got better at staying alive):
The grid dimensions can be changed by modifying the DeepQNetwork class' architecture in dqn.py and changing grid_size and making a new model_path with ".pt" ending in config.py. Other constants in config.py can be changed, such as the learning rate or gamma. Then, to train, run train.py, and then view.py. Note that even when running view.py, the deep q network has to have the same architecture as the corresponding ".pt" model.