This was just a fun little project to train a simple reinforcement learning model to play the classic snake game. Basically it's just +10 reward for eating an apple and -10 reward for dying. The game is made with pygame, and I ran 1000 games to see how good the model would get over that time. Here are the training results:
The best score was 85, but it seemed to average at around 35-ish. It started to figure out the zigzag technique at around 400 games or so which was pretty interesting to see.
I might to tetris next hmmmm idk
Just setup a venv and run train.py. You can change MAX_GAMES in that file if you wanna train for longer.