poker-RL

Python poker Texas Hold'em AI through TensorFlow reinforcement learning and PyPokerEngine.

Install

Clone this repo, then run the command:

• pip3 install -r requirements.txt
  

Play vs the AI in GUI

Run this command in terminal.

• python3 -m pypokergui serve poker_conf.yaml --port 8000 --speed moderate
  

Training

To train the AI at 1v1 poker (be sure to set which player you'd like to train against in line 104 of the script).

• cd src/
  

• python3 training1v1.py
  

Pre-trained models can be found in src/cache/models/DQN2 for 1v1 or DQN9 for 1v8 poker. To visualize tf with tensorboard run this command in terminal:

• tensorboard --logdir=log/DQN2/
  

Simulating / Testing

Play a game against chosen player:

• cd src/
  

• python3 play2.py
  

Or simulate 50 games against various players and save plots:

• python3 simulate.py
  

Reports

The project was completed for a directed study in school. Feel free to check them out in reports/ (written in Latex).

Versions

• running on Mac OS X 10.15.3
• brew install pyenv
• pyenv install 3.6.0 # link pip3 and python3 aliases
• add these to .bash_profile

• alias python3=/Users/Ryan/.pyenv/shims/python3
  

• alias pip3=/Users/Ryan/.pyenv/shims/pip3
  

• python3 --version (ensure 3.6.0)
• make bug fix for pypokergui shown below

Props

• Initial setup thanks to https://github.com/VIVelev/PokerBot-AI
• Bug fix in python module pypokergui 'site-packages\pypokergui\server\templates'. See here for fix: https://github.com/chrisking/PyPokerGUI/commit/00d63bec4b2a3ee195851521362f812321fa8e4d
• RLCard Contributors for references: https://github.com/datamllab/rlcard
• EvgenyKashin @ Github for Double-Dueling-DQN architecture: https://github.com/EvgenyKashin/TensorPoker