NLH Poker Reinforcement Learning agents

Acknowledgement

This project is a course project of ESTR3108 in CUHK. It is based on PyPokerEngine, detailed docs on env could be found at its doc site.

Preliminaries

# Game environment
pip install PyPokerEngine
# Pytorch
pip install torch
# Process the result
pip install scipy

Run

# Train AI agents
python train.py

# Test a model or just play a game
python test.py

Agent list:

• A2C Player
• DQN Player
• Q-Learning Agent
• Card Player
• Honest Player
• Random Player
• AllCall Player
• Human Player

Directory Structure

Train Example

An example figure oczraining process.

best configuration

• LR=1e-4
• gamma=0.95
• Neural Network: 8->128->8, relu
• Reward: No everage, /150