Reinforcement Machine Learning Snake Game using PyTorch framework
This is my first main large machine learning project, using the PyTorch framework.
The actual snake game is an adapted version of assets from https://github.com/patrickloeber/python-fun/tree/master/snake-pygame and it was very helpful to have a easy framework to build off. However the controls were made for a human to play and thus the game was adapted.
I'm using Deep Q-Network with PyTorch, a form of reinforcement learning, and I'm specificially implementing the Bellman Equation:
Implement Game to be controlled by AgentImplement AgentImplement ModelImplement Saving of Model- Implement loading of saved models
- Experiment with different parameters
- Create a way for the model to play without using the comparatively slow pygame
- Use cuda to use the GPU for processiong, hopefully with the last point making 1000 generations not take 4+ hours
- 10,000 generations of learning for a few different parameters?