Connect 4 is a popular board game in which two players alternately take turns dropping their respective tokens on a 6 × 7 board. It has been solved, and it can be shown that the player going first can always win. ConnectX is a generalizaton of this game. Our algorithm has to learn to play across variable board sizes and might have to connect more than 4 tokens, and this makes the task significantly more challenging.
I teamed up with Shubham Gupta to attempt to create an agent to play ConnectX as the course project in an advanced Reinforcement Learning course under Prof. Shivaram Kalayankrishnan.
I thoroughly enjoyed the process of formal research, involving a literature survey and so much trial and error. I must also give credit to Prof. Kalayankrishan, who was a terrific guide. The report, with all the technicalities of the project, can be found at arXiv:2210.08263 [cs.AI].