OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Chess Alpha Zero

Chess Alpha Zero is an open-source research codebase that applies AlphaGo Zero-style reinforcement-learning methods to chess, with documented supervised learning from PGN game files, self-play, model training and evaluation workflows.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Chess Alpha Zero — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it to study or adapt a chess-learning codebase, especially when experimenting with PGN-based supervised learning or examining the documented training-worker design.

Who it is for

  • Researchers and students studying reinforcement learning in chess.
  • Developers exploring supervised training from PGN records or self-play training workflows.
  • Chess-engine users interested in connecting the documented UCI interface to a GUI.

Documented uses

  • Inspect an implementation of AlphaGo Zero-inspired methods applied to chess.
  • Prepare PGN game records for the documented supervised-learning pipeline.
  • Understand how self-play, model optimization and evaluation are divided into workers.
  • Connect the UCI interface to a chess GUI.
Before you choose

Requirements & limits

The repository is public and MIT-licensed. Its README lists Python 3.6.3, TensorFlow GPU 1.3.0 and Keras 2.0.8 as the environment.

  • The README describes self-play as too costly for a single machine, while reporting that supervised learning worked for the authors.
  • The README recommends at least 3,000 games and no more than three to four epochs for the supervised-learning step.
  • The documented UCI interface supports a chess-GUI workflow separate from the model-training workers.
  • A Binder demo notebook is linked, but the documentation does not establish a complete current setup path for reproducing training.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare supervised-learning data

Prepare PGN game files for the repository’s supervised-learning workflow.

Before you start

  • Chess PGN files
  • A copy of the repository
  • The README’s documented play-data folder: data/play_data

Steps

  1. Download PGN game files and place them in data/play_data.

  2. Use the documented supervised-learning worker in place of the self-play worker: python src/chess_zero/run.py sl.

  3. The README recommends datasets of at least 3,000 games and at most three to four epochs to reduce overfitting.

Expected outcome

PGN files are staged for the documented supervised-learning process, which can train a model from recorded games.

Guide limitations

  • The README does not provide a complete account of dependencies or a full training setup procedure.
  • The project authors report that self-play was too costly on a single machine; this guide covers supervised-data preparation, not a successful training run.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
Zeta36/chess-alpha-zero · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

Leela Chess Zero — The README points readers to this separate project for distributed chess-engine development and participation; it is presented as a moved-on project, not as a component of this rep