This is a course project of SJTU AI3617.
The group name is: 璃月究极赌怪之光
Members include: Wenzhuo Zheng, Mingyang Jiang, Jinghao Feng
Limited Texas Hold'em is a world-wide popular card game, and also a typical imperfect information multi-agent game. It's a compound of fortune and strategy. This project studies application of MARL algorithms on this game. We first study performance of existing RL algorithms, and propose some customized modifications based on experimental observations. Then, we integrate our ideas to propose a brand new value-aware and rival-aware MARL algorithm that outperforms all baselines.
-
Check dependencies in
requirements.txt. -
If you want to reproduce our results in our paper, run
train.py.
Example:
python train.py --player1 cheater --player2 random- If you want to load pretrained model and evaluate, run
evaluate.py.
Example:
python evaluate.py --player1 cheater --player2 random --load_path1 path_to_your_model- If you want to train the cheating sheet, run
train_table.py.
Example:
python train_table.pyDefault output directory is networks.
- If you want to play with your model, run
Human_vs_AI.py.
Example:
python Human_vs_AI.py --agent cheater --load_path path_to_your_modelHave fun!