Browsing by Author "Lang, Sean Ryan"

Cribbage is a card game involving multiple methods of scoring which each receive varying emphasis over the course of a typical game. Reinforcement learning is a machine learning strategy in which an agent learns to accomplish a task via direct experience by collecting rewards based on performance. In this thesis, reinforcement learning is applied to the game of cribbage, improving an agent’s policy of combining multiple basic strategies, according to the needs of the dynamic state of the game. From inspection, a reasonable policy is learned by the agent over the course of a million games, but an increase in performance was not demonstrated.