🐍 Reinforcement Learning for Snake Game 🐍

This project trains a Deep Q-Network (DQN) to play the classic Snake game using Reinforcement Learning (RL). The agent learns by interacting with the environment and optimizing its movements to maximize the score, as defined in the reward function.

---

Project Structure

├── snake-reinforcement-learning
│   ├── models
│   │   ├── dqn_agent.py      # Deep Q-Network agent implementation
│   ├── snake_game
│   │   ├── __init__.py
│   │   ├── snake_game.py     # Game environment (Pygame-based)
│   ├── train.py              # Training script for RL agent
│   ├── evaluate.py           # Evaluate the trained model
│   ├── dqn_snake.pth         # Saved trained model (generated after training)
│   ├── requirements.txt      # Required dependencies
│   ├── .gitignore            # Files to ignore in Git
│   ├── README.md             # Project documentation (this file)

---

Installation & Setup

1) Clone the Repository

git clone https://github.com/SergioV3005/snake-reinforcement-learning.git
cd snake-reinforcement-learning

2) Create and Activate Virtual Environment

python -m venv venv
# Activate it:
# Windows:
venv\Scripts\activate
# Mac/Linux:
source venv/bin/activate

3) Install Dependencies

pip install -r requirements.txt

---

Running the Game

Train the RL Agent

Train the agent using Deep Q-Learning:

python train.py

This script trains the agent for multiple episodes and saves the learned model as dqn_snake.pth.

Evaluate the Trained Agent

After training, you can test the agent’s performance:

python evaluate.py

This will run the trained model inside the Snake game environment and let the AI play.

---

Technical Details

• Algorithm: Deep Q-Learning (DQN)
• Frameworks: PyTorch, NumPy, Pygame
• State Representation: Distance to walls, food direction, collision risks
• Rewards System:
  • +10: Eating food
  • -10: Hitting a wall or itself
  • -0.1: Staying idle (encourages movement)

---

License

This project is open-source and available under the MIT License.

---

Contributions & Improvements

Feel free to contribute! Open an issue or submit a pull request if you have ideas for improvements.