🎲 GNUBG for Python3

Advanced neural network backgammon evaluation engine, now in Python.

This project brings the full power of GNU Backgammon’s neural net evaluator into a modern Pythonic workflow, enabling game developers, researchers, and AI engineers to use state-of-the-art backgammon logic in their own applications.

🔧 Features

• 🧠 Neural Net Evaluation: Direct bindings to GNUBG’s six trained networks.

• 🎲 Best Move Finder: Get top plays with customizable ply-depths and move listings.

• 📈 Rollouts & Probabilities: Simulate thousands of games for accurate statistics.

• 🎯 Position Classification: Automatically detect race, bearoff, contact positions.

• 💡 Resign Logic: Integrated resignation decisions based on win probability.

• 🧬 Equities Tables: Built-in support for GNUR, Snowie, Jacobs, and more.

• 🔍 C-level Speed: Evaluations run via compiled C++ backend using Meson for fast builds.

📦 Installation

pip install gnubg

Built for Linux, macOS, and Windows (via MinGW). Requires Python 3.8+.

📚 Usage Example

from gnubg import best_move, board_from_position_id  
board = board_from_position_id("4HPwATDgc/ABMA")  # GNUBG Position ID dice = (3, 5) 
moves = best_move(board, *dice, n=2, s='X') 
print(moves) 

Want to simulate match outcomes? Run rollouts or cube decisions using built-in methods like rollout() and evaluate_cube_decision().

👨‍💻 Built With

• C++ bindings via pybind11 and Meson

• Direct port of GNUBG internals

• Python 3 extension module gnubg with complete API coverage

⚖️ License

GNU GPLv2 — same as the original GNU Backgammon project.

📥 Source Code & Contributions

📂 GitHub: github.com/reayd-falmouth/gnubg-nn-pypi 

🐍 PyPI: gnubg · PyPI

Contributions and PRs welcome!