Document Type

Thesis

Publication Date

2000

Disciplines

Computer Sciences | Physical Sciences and Mathematics

Advisor

Andrew Holey

Abstract

The goal of this project was to apply unsupervised machine learning algorithms to the board game othello. We used eight genetic algorithms to develop eight different othello strategies. Each genetic algorithm played 2.5 million games to develop one of the eight strategies. Each strategy was then tested against the other seven strategies, as well as other computer players and human players. The results show that most of the strategies play at the level of a beginning player, with the best strategies on par with a slightly experienced player. We also implemented a reinforcement learning algorithm in order to improve on the weaknesses of the genetic algorithm. This strategy worked by storing many of the board configurations the computer encountered while playing othello, as well as the results of the games it played. The computer could then make decisions based on past experiences. This method did not produce any significant results, but possibly could with some improvements to the algorithm.

Share

COinS