Snake Game AI

This project implements a Snake game with an AI agent that learns to play the game using reinforcement learning. The AI is trained using a Deep Q-Network (DQN) approach.

Project Structure

├── game.py
├── model.pth
├── requirements.txt
├── training_stats.txt
└── README.md

• game.py: Contains the main code for the Snake game and the AI agent.
• model.pth: The saved model weights for the AI agent.
• training_stats.txt: Contains training statistics such as the number of games played and the highest score achieved.
• requirements.txt: Requirements file
• README.md: This file.

Requirements

• Python 3.x
• Pygame
• NumPy
• PyTorch
• Matplotlib

Installation

1. Clone the repository:

    git clone https://github.com/pranavms13/snake-dqn
    cd snake-nn

2. Install the required packages:

    pip install -r requirements.txt

Usage

To train the AI agent, run:

python game.py

The training process will start, and the AI agent will learn to play the Snake game. The training statistics and model weights will be saved periodically.

Classes and Functions

game.py

• Direction: Enum class for the direction of the snake.

• Point: Class representing a point on the game board.

• SnakeGame: Class representing the Snake game.

  • __init__(self, w=640, h=480): Initializes the game.
  • reset(self): Resets the game.
  • play_step(self, action): Executes a game step based on the action.
  • is_collision(self, pt=None): Checks for collisions.
  • _update_ui(self): Updates the game UI.
  • _move(self, action): Moves the snake based on the action.
  • _place_fruit(self): Places a fruit on the game board.

• Agent: Class representing the AI agent.

  • __init__(self): Initializes the agent.
  • get_state(self, game): Gets the current state of the game.
  • remember(self, state, action, reward, next_state, game_over): Stores the experience in memory.
  • train_long_memory(self): Trains the agent using long-term memory.
  • train_short_memory(self, state, action, reward, next_state, game_over): Trains the agent using short-term memory.
  • get_action(self, state): Gets the action to be taken by the agent.

• Linear_QNet: Class representing the neural network for the Q-learning agent.

  • __init__(self, input_size, hidden_size, output_size): Initializes the neural network.
  • forward(self, x): Forward pass of the neural network.
  • save(self, file_name='model.pth'): Saves the model weights.
  • load(self, file_name='model.pth'): Loads the model weights.

• QTrainer: Class for training the Q-learning agent.

  • __init__(self, model, lr, gamma): Initializes the trainer.
  • train_step(self, state, action, reward, next_state, game_over): Performs a training step.

• train(): Main function to train the AI agent.

License

This project is licensed under the MIT License.

Feel free to modify the README as needed to better fit your project.