yhyu13/AlphaGOZero-python-tensorflow

What it solves

This project provides a Python and TensorFlow implementation of the AlphaGo Zero architecture, enabling the training of an AI agent capable of playing the game of Go without requiring human knowledge or data.

How it works

The system implements the architecture described in DeepMind's "Mastering the Game of Go without Human Knowledge" paper. It supports supervised training, a self-play pipeline for reinforcement learning, and distributed learning to improve the agent's performance.

Who it’s for

Developers and researchers interested in reinforcement learning and the AlphaGo Zero architecture who want to experiment with training their own Go-playing AI.

Highlights

• Self-Play Pipeline: Implements the core mechanism for the agent to learn from its own games.
• GTP Support: Includes the Go Text Protocol, allowing the agent to be used as an engine for Go software like Sabaki.
• Distributed Learning: Supports training across multiple resources to accelerate the learning process.
• Supervised Training: Provides the ability to perform initial training using existing game datasets.