Project Overview
PokerBot is a limit Texas Hold'em game where players compete against a machine learning model. It is built around a MVC format seperating model, controller, and view components for easier game development and expansion.
Gameplay and Interface
For a full breakdown of the PokerBot architecture, visit the project overview on GitHub: PokerBot Architecture Overview. You can also view the game being played starting at 3:20 in the video above.
Machine Learning Approach
The bot uses a Q-learning framework that allows it to improve over iterations. At the start of the game, the bot will make completely random decisions. However, we created a command line that allows you to make it play against itself however many times you specify. After maybe several thousand iterations it becomes quite good, and as you play it, it continues to learn using reward based outcomes. This allows it to decide when best to raise, stand, or fold based on the expected long-term reward given with what hand they have.
