Playing the game "Rock, Paper, Scissors, Lizard, Spock" is not difficult since the game is both simple and, at first glance, subject to chance. It is more challenging to develop a winning strategy that will guarantee a win or many wins in the long run. In this thesis, we used artificial intelligence to try to discover a pattern in the moves of the opponent, focusing specifically on the history of moves in a game to secure a positive win/loss ratio. We did not include other factors, such as observing the movement of the opponent’s hand as they choose which move to use.
For this purpose we developed the following algorithms: history string matching, markov chains, reinforcement learning based on moves and meta-classificator with reinforcement learning. We tested our developed algorithms through various test scenarios, which included machine learning with the help of simple auxiliary algorithms and move sequences that were generated in advance.
After conducting all experiments, we compared the algorithms and analyzed their performance. The results have shown that in most test scenarios the best performing algorithms were markov chains and meta-classificator with reinforcement learning.
|
