sharedcare / mahjong_rl

Reinforcement learning approaches for Mahjong AI

Mahjong

The objective of this project is to build a reasonable AI agent to play Mahjong with human players.

The enviroment of Mahjong is built by RLCard.

Requirements

• PyTorch
• RLCard

Usage

Train PPO Agent

python main.py --log_dir 'experiments/mahjong_ppo_result/'

Play Demo with trained agent

python demo.py

Todo

• PPO Agent
• Demo
• Distributed PPO Agent
• MAPPO Agent
• SAC Agent
• Add other types of Mahjong games