Benjamin Tillius: Yatzy som en Markovbeslutprocess: optimala strategier anpassade utifrån spelarens mål

Abstract: This paper investigates the Swedish version of the dice game Yahtzee. The paper focuses on modelling the game as a Markov Decision Process (MDP) and uses backward induction to calculate different optimal strategies (policies) depending on the player's objective, as well as the expected total reward given the strategy. The main comparison is between the optimal strategy for maximizing the final score and the optimal strategy for maximizing the probability of reaching some specific score, such as a former record. The optimal expected final score was calculated to be 248.4400.