tensorflow/minigo

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

3542 stars · 561 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Minigo is an open-source, minimalist implementation of the AlphaGo Zero algorithm for the game of Go, built on TensorFlow. It provides a readable reinforcement learning pipeline for self-play training, evaluation, and serving of Go models, with tooling for Kubernetes and Google Cloud Platform.

Use cases

• reproduce the AlphaGo Zero algorithm from scratch
• train a neural network Go engine via self-play reinforcement learning
• learn how to build RL pipelines with TensorFlow and Kubernetes
• run a Go AI engine to play against or analyze games
• study a readable implementation of AlphaZero-style training
• benchmark RL training on GPUs and TPUs

When to choose

• you want a clear, educational implementation of AlphaGo Zero rather than a maximally competitive engine
• you want to experiment with self-play RL pipelines on cloud infrastructure
• you need open training data and models for Go research
• you want to extend or adapt an AlphaZero-style architecture for other games

When to avoid

• you need the strongest possible Go AI for serious play or analysis
• you want a lightweight engine without TensorFlow or GPU dependencies
• you need actively maintained software with recent releases
• you want a general-purpose game AI framework rather than a Go-specific one

Facets

library · maturity maintenance

machine-learning deep-learning reinforcement-learning game-engine simulation artificial-intelligence machine-learning reinforcement-learning education python cloud go-engine alphago-zero self-play tensorflow board-games mcts neural-network game-development linux docker gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="tensorflow/minigo")