Deep Connect 4
A deep reinforcement learning implementation of the classic board game Connect4.
While taking the Artificial Intelligence course with Dr. Oge Marques, a unique project idea emerged. Henry Herzfeld (GitHub) (LinkedIn), Yuri Villaneuva and I decided to investigate whether we could create a deep reinforcement learning implementation of the classic board game Connect4.
The rules are simple: Given a 6x7 board and two players, try to get four of your pieces stacked horizontally, vertically or diagonally. Players take turns choosing a column to move in on their turn. Can a reinforcement learning agent learn to play this game, and play it well? This is the question we sought to answer.
A summary of our work can be found in this notebook here.
Much of the credit for the writing in this notebook goes to Yuri Villanueva. There is also an interactive link at the bottom of the notebook to Henry's personal site, where the reader may play Connect4 with the agent we trained. Henry and I continued our work on this project past this course. The GitHub for the project can be found here.