Optimal Yahtzee

Optimal Yahtzee: A Comparison between Different Algorithms for Playing Yahtzee
Daniel Jendeberg & Louise Wikstén
KTH Royal Institute of Technology
2015

Abstract

This study investigates the game of Yahtzee, a semi-strategic, luck-based dice game, to determine how different algorithms perform in maximizing the game’s score. The primary focus is on comparing an optimal algorithm, derived from probability theory and graph-based analysis, with other heuristic-based algorithms. Additionally, human trials were conducted to assess how well humans, through experience and reinforcement learning, can approximate optimal play. The optimal algorithm was found to outperform other strategies in terms of scoring, but at a significant computational cost, requiring approximately 13.8 GB of RAM and 22 hours to compute. The study concludes that while the optimal algorithm offers superior performance, its high resource demands make heuristic-based approaches more practical for real-time play. Human players, with sufficient practice, also demonstrated near-optimal performance, suggesting that reinforcement learning can effectively guide strategy in Yahtzee.

More Resources

3-D Dynamics of a Dice Roll

Optimal Multiplayer Yahtzee

Game and Variance