Comparison of Expectimax and Monte Carlo algorithms in Solving the online 2048 game

• Autores: Noa Yarasca, Efrain, Nguyen, khoi
• Formato: artículo
• Fecha de Publicación: 2018
• Institución: Universidad Nacional Mayor de San Marcos
• Repositorio: Revista UNMSM - Pesquimat
• Lenguaje: español
• OAI Identifier: oai:ojs.csi.unmsm:article/15069
• Enlace del recurso: https://revistasinvestigacion.unmsm.edu.pe/index.php/matema/article/view/15069
• Nivel de acceso: acceso abierto
• Materia: 2048 game; Expectimax algorithm; Monte Carlo algorithm; heuristics; Juego 2048; Algoritmo Expectimax; Monte Carlo; heuristicas

In this work, two search algorithms Expectimax and Monte Carlo Tree Search (MCTS) were developed to solve the well-known “2048" puzzle online-game and compare their results. In both cases, five heuristics were employed to obtain favorable tile positions within the game. These heuristics were combined to maximize the game-score in all possible board positions. As a result, the game-score, the maximum value of tile obtained, and the computing time employed in solving the game are shown. In addition, the efficiency of each algorithm and its sub-cases are presented. This research concludes by arguing that Monte Carlo Tree Search was more efficient in higher score than Expectimax algorithm, although in a longer time. Increments in level of depth-search in Expectimax and number of moves in MCTS do not necessarily resulted in obtaining higher score.