Aleo is a flexible Mahjong AI development toolkit written in C++. It was developed to facilate the design of high-level Mahjong algorithms. With Aleo, you can easily implement AI for Mahjong under GuoBiao(aka Chinese Official) rules, and deploy it to Botzone.
- Easy composition and decomposition of functions
- Extremely fast search of winning hands
- Simulator for self play
- Clear Python interface for machine learning
Aleo won the first place in the competition of 2018 Game AI course, with a remarkable margin over the second competitor. It has an average winning rate of 46.8%, as well as excellent winning scores.
Aleo can search over all possible combinations of 5 waiting tiles in 0.5s on a single i7-4720HQ CPU.
For Windows, open Aleo.sln in Visual Studio. There are three main projects in Aleo:
- Bot
- Data generator
- Faan calculator
It is recommended to compile each project in Release mode to achieve full speed.
Aleo has only tested with data generator and faan calculator on Linux.
For data generator, run make generator. For faan calculator, run make fan_calculator.
Note that make clean is required between two different make commands.
- Visual Studio 2013 or later
- G++ 4.8.4 or later
- Python 2/3 with Tensorflow and Keras
Aleo is authored by Zhaocheng Zhu and Fangyin Wei.
Licensed under an Apache-2.0 license.