Pachi can refer to: a simple modular framework for programs playing the game of Go/Weiqi/Baduk, and a reasonably strong engine built within this framework.
|pachipachi | minipachi|
The default engine plays by Chinese rules and should be about 7d KGS strength on 9×9. On 19×19, it can hold a KGS 2d rank on modest hardware (Raspberry Pi, dcnn) or faster machine (e.g. six-way Intel i7) without dcnn. When using a large cluster (64 machines, 20 cores each), it maintains KGS 4d and has won e.g. a 7-stone handicap game against Zhou Junxun 9p.
By default, Pachi uses the UCT engine that combines Monte Carlo approach with tree search; UCB1AMAF tree policy using the RAVE method is used for tree search, while the Moggy playout policy using 3×3 patterns and various tactical checks is used for the semi-random Monte Carlo playouts. MM patterns and deep learning are used to guide tree search. Technical information about Pachi's architecture and algorithms can be found in Petr Baudis' Master's Thesis (more papers).
Current stable Pachi version is 12.84 (codename Jowa), released on September 10, 2023.
You can download it or explore the source code online.
Pachi implements only the playing algorithm and does not have any builtin user-friendly interface. In order to play with Pachi, you will want to also install a "GTP frontend" software, e.g. gogui; you can also use e.g. kgsGtp to connect your Pachi instance to an internet Go server (example config).
Pachi's source code is maintained within the
Pachi git repository.
Patches and help is encouraged, as well as reuse of the source code base for your own research.
We use github for discussions, coordinating development and providing user support.
In the past email@example.com mailing list has been used as well.
Pachi is developed by: