A reinforcement learning implementation of the classic Snake game.
This project implements a reinforcement learning agent that learns to play Snake. The agent is trained to maximize its score by collecting food while avoiding collisions with walls and its own body. Created a small application using streamlit to showcase the agent as well as a visualisation of the neural network in action.
- Framework: Streamlit
- Algorithms: Deep Q-Learning
- State Representation: Grid-based environment state
- Action Space: Discrete (Left, Right, Forward)
pip install -r requirements.txt
To run the application:
streamlit run app.py
The agent was trained using an epsilon-greedy strategy with the following parameters:
- Learning rate: 0.001
- Discount factor: 0.99
- Epsilon decay: 0.995
- Replay memory size: 10000
MIT