Snake Game AI

Snake AI is a Python-based implementation of the classic Snake game with an AI agent that learns to play the game using Q-learning. This project is designed to showcase how reinforcement learning techniques can be applied to train an AI agent to perform tasks.

Table of Contents

• Getting Started
  • Prerequisites
  • How to Use
• Project Structure
• Training the AI Agent
  • Hyperparameters
  • Training Progress

Getting Started

Prerequisites

• Pygame (for the game interface)
• PyTorch (for the neural network)
• Matplotlib (for plotting graphs)

How To Run

• Open cmd in current directory
• run 'python agent.py'

Project Structure

• agent.py: Entry Point of the program which call all other classes
• game.py: Contains Game logic and Snake game class
• model.py: Defines the neural network used for Q-learning.
• plotting.py: Provides functions for plotting game statistics.
• model.pth: Pre-trained model weights './model./model.pth'

Training The AI Agent

Hyperparameters

You can adjust the AI agent's learning behavior by modifying hyperparameters in the agent.py file. Experiment with different values to see how they affect learning speed and performance.

Training Progress

• During training, the AI agent's progress is displayed in the console.
• The training progress can be visualized using the provided plotting functions.
• At Learning Rate = 0.001, Batch Size = 1000, Discount Rate = 0.8, Epochs = 1000 and 255 hidden layers model attained a score of 84 (at around 600ish epoch) and average score was ~23

Acknowledgments

• FreeCodeCamp for inspiration for Reinforcement learning model