The computer GO framework and the AI using MCTS (Monte Carlo Tree Search)
Source code that the author uses for master thesis.
- A framework that runs computer GO between AI & humans, or AI & AI.
- Computer players using Monte-Carlo Tree Search (MCTS). These AI can battle with other computer players via GTP commands.
- Machine learning frameworks using pat GO records to enforce the computer players.
You can run main computer GO framework only on Linux OS because some API (especially pthread) depends on Linux API.
- Build: run "make" in root directory (read Makefile for other make options).
- Run: read go_main_release_dynamic_runner.h or go_main_release_vs_gnugo_runner.h. These scripts run a computer player vs another computer player.
Source code
- Root dir: Entiry points and some common functions/classes.
- AI: Computer players
- Go: The main framework of computer GO.
- Gtp: GTP command wrapper and a main entry point to run the framework as GTP mode.
- ML: A machine learning framework. Source code in this directory are not incorporated with the main GO framework. You can run it as standalone.
- Record: Classes to read past records. These classes are mainly used by ML.
- utility: Utility classes, such as random, vector.
- tests: Tests run by Google Test.
Other files
- book: Books (in Japanese, "Jouseki") for Fuego ([http://fuego.sourceforge.net/])
- for_fuego_parameter_files: Parameter files to run fuego (author used these files to check performance author's AI v.s. Fuego)
- parameter_files: Parameter files to run author's AI.
- save_results: Result files of battle between AI, machine learning and so on...
Except for external libraries and resources, this software is released under the MIT License, see LICENSE.txt.
omochi (@omochi64): [https://github.com/omochi64]