bortk / gto-poker-bot Goto Github PK

This project is part of CSC 481 - Knowledge Based Systems at Cal Poly with Professor Rodrigo Canaan.

The autonomous part of our project is based of the work of self-operating coomputer by HyperwriteAI https://github.com/OthersideAI/self-operating-computer.

Clone the repository

git clone https://github.com/arnenoori/gto-poker-bot

Create venv

python3 -m venv env

Activate it (Mac)

source env/bin/activate

Install requirements

pip install -r requirements.txt

Add OpenAI Key

export OPENAI_API_KEY=yourkeyhere

Go on your poker website of choice. We used www.247freepoker.com and played against bots.

To run the agent with the fixed strategy agent (default), simply run:

python play.py

• agent_random.py: an agent making random decisions (used for testing and comparison)
• agent_dqn.py: a deep q agent
• fixed.py: a fixed model

You can test the performance of the 3 different agents playing each other by running:

python evaluate.py

Results after 1000 hands:

DQN Agent Wins: 51

Fixed Model Wins: 927

Random Model Wins: 6

Ties: 16

GPT-V is used to extract the game state from the screenshots. The extracted information is then used to make decisions based on the current state of the game. Here is an example of the screenshot sent to GPT-V:

Example of the screenshot sent to GPT-V:

Read more in our technical report.

• gamer/
  • adapter.py
  • api.py
  • config.py
  • fixed.py
  • operating_system.py
  • prompts.py
  • tools.py
  • utils.py
• media/
  • cowboy_bebop_poker.gif
• report/
  • proposal/
    • CSC481_Project_Proposal.pdf
    • projectproposal.md
  • 481_report.md
  • 487_report.md
  • final_presentation.pdf
• .env
• .gitignore
• README.md
• dqn_agent.py
• evaluate.py
• play.py
• random_agent.py
• requirements.txt