In this project, we have implemented several reinforce learning methods to solve the Mahjong game. Our environment is based on the project of https://github.com/clarkwkw/mahjong-ai. We used its Game.py , Player.py, HKRules.py, Tile.py, MoveGenerator.py to build the pipeline of this complicated game.

In the folder of MCTS_method, we implemented the Monte Carlo Tree Search method for the agent to play the Mahjong game. The MonteCarloTree_move.py is the implementation of interface MoveGenerator in MoveGenerator.py. You can download and run the test.ipynb to check out the game test with RandomGenerator and our implementation MCTS_Generator.

In the folder of DeepQLearning, we inplemented the Deep Q-Learing algorithme. To train the model, please run the scricpt Training.py. Network.py is the deep network we utilize; Generator.py is the implementation of MoveGenerator. Dependency: Tensorflow

In the folder of PolicyGradient_ActorCritic, we have implemented the method of Policy Gradient and Actor Critic. To run the code, please open the jupyter notebook respectively. Dependency: Tensorflow