GNU Go

GNU Go is a classical Go engine that does not use neural networks. It is a lighter alternative to KataGo, useful when GPU resources are limited or when a weaker opponent is desired for curriculum training.

Installation

# Ubuntu / Debian / WSL
sudo apt-get install -y gnugo

# macOS
brew install gnugo

Service Module

The GNU Go integration lives in:

  • gym_gui/services/go_ai/gnugo_service.py

  • gym_gui/services/go_ai/gtp_engine.py (shared GTP base class)

GNU Go communicates via the GTP (Go Text Protocol), sharing the same base class as KataGo.

Usage in MOSAIC

  • Human vs GNU Go: Play Go against a classical engine using the GUI.

  • Curriculum training: Use GNU Go as an easier opponent before graduating agents to play against KataGo.