This repository provides a Japanese Riichi Mahjong AI agent, suitable for researchers and developers looking to build or test their own Mahjong AI. It offers a framework for real-time testing against human players and includes a development library for processing game logs.

How It Works

The AI employs heuristics and an enhanced opponent model using a "bagging of multilayer perceptrons" approach. This strategy aims to improve performance by combining multiple predictive models, specifically for predicting dangerous tiles and opponent behavior, outperforming previous state-of-the-art agents in experiments.

Quick Start & Requirements

• Install/Run: Execute Python scripts from the client directory, e.g., python main.py or python gui_main.py.
• Prerequisites: Python, requires registration and a User ID from tenhou.net for online play.
• Resources: Running the AI involves connecting to tenhou.net.
• Docs: Game rules: https://en.wikipedia.org/wiki/Japanese_Mahjong. Development library: https://github.com/erreurt/MahjongKit.

Highlighted Details

• Outperforms state-of-the-art agents in experiments on tenhou.net.
• Includes a framework for developing custom AI agents with defined interfaces.
• Offers a GUI for debugging and real-time testing.
• Future work includes Random Forest for meld/Riichi decisions and LSTM for waiting tile prediction.

Maintenance & Community

The project is a master's thesis project; updates and documentation are infrequent due to the author's availability. Contact via email is suggested for detailed algorithm information.

Licensing & Compatibility

The repository does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project's README indicates infrequent updates and a need for better documentation. Tenhou.net may ban IPs for excessive parallel testing. The AI's performance is benchmarked against specific prior works, and its effectiveness in diverse environments is not detailed.