ML Game Trainer

A machine learning game training platform with multiple classic games and AI agents.

Features

• Multiple Games: Snake, Pong, Flappy Bird, Tetris, Galaga, Dino Run, Packman
• AI Training: Deep Q-Learning with neural networks
• Real-time Visualization: Live training plots and game windows
• Speed Control: Adjustable game speed during training
• Model Persistence: Automatic model saving and loading

Games

Snake Game

• Classic snake gameplay with food collection
• Neural network learns optimal pathfinding

Pong Game

• Two-player pong with AI paddle
• Improved reward structure and state representation

Flappy Bird

• Bird navigation through pipes
• Continuous flight mechanics

Tetris

• Block stacking and line clearing
• State includes current piece and board layout

Galaga

• Space shooter with enemy waves
• Multiple enemy types and movement patterns

Dino Run

• Endless runner with obstacles
• Random obstacle generation

Packman

• Classic maze navigation with 3 ghosts
• Different ghost behaviors (Red: aggressive, Pink: conditional, Blue: random)
• 1/3 random bean placement

Installation

pip install pygame torch numpy matplotlib tkinter

Usage

python main.py

1. Select a game from the UI
2. Watch the AI train in real-time
3. Adjust speed and controls as needed
4. Models are automatically saved to model/ directory

Architecture

• main.py: Application entry point and game selection
• ui_windows.py: Tkinter UI for game selection and controls
• game_runner.py: Game loop and AI training coordination
• agent_core.py: Neural network agent implementation
• game_interface.py: Base class for all games
• utils.py: Utility functions for resource paths

Model Files

Trained models are saved as:

• model/model_[GameName]_[linear/cnn].pth

Requirements

• Python 3.7+
• Pygame
• PyTorch
• NumPy
• Matplotlib
• Tkinter