Dots and Boxes DQN

An interactive visualization of Deep Q-Network agents learning to play Dots and Boxes through self-play.

Features

• Real-time visualization of game board and training progress
• Interactive speed controls (0.2x - 5.0x)
• Training metrics visualization:
  • Win rates
  • Evaluation scores
  • Q-values
  • Move quality
• Play against trained models
• Checkpoint system for saving and loading models

Installation

pip install dots-and-boxes-dqn

Usage

python -m dots_and_boxes_dqn

Then open http://localhost:5000 in your browser.

Training Controls

• Speed: Adjust training visualization speed
  • 0.2x: Slow motion for analysis
  • 1.0x: Normal speed
  • 3.0x: Fast visualization
  • 5.0x: Training focus
• Pause/Resume: Temporarily halt training
• Play vs AI: Test your skills against saved models

Technical Details

• Uses Dueling DQN architecture
• Prioritized Experience Replay
• Self-play training methodology
• Periodic model checkpointing
• Flask + SocketIO for real-time updates

Development

git clone https://github.com/yourusername/dots-and-boxes-dqn
cd dots-and-boxes-dqn
pip install -e ".[dev]"