Cribbage Counterfactual Regret Minimization

The goal of this project was to research techniques that are utilized in programming agents for imperfect information two-player zero-sum games and apply them to Cribbage. Particularly, Counterfactual Regret Minimization (CFR) was applied to both the throwing and pegging stages of the game. The best CFR agent could beat the provided greedy opponent by 0.39 match points, with some simpler & faster ones beating greedy by 0.32-0.36 match points. Other methods, such as using Monte Carlo Simulations and Schell’s Discard Tables were explored with mixed results.