Search space pruning by evolving probabilistic strategies

Abstract

A famous application of Genetic Algorithms (GA) is evolving strategies, pioneered by Axelrod's Prisoner's dilemma experiment. This paper aims to explore two new techniques for evolving playing strategies relative to that of Axelrod. The three techniques have been tested by a very challenging game: Rock-paper-scissors-lizard-spock. The proposed techniques evolve a probabilistic choice. In the first proposed technique the history is completely disregarded. The chromosome size is shortened by 99% relative to Axelrod's. The second proposed technique is a compromise between Axelrod's and the later probabilistic technique. Still the chromosome size is shortened by more than 95%. Results show that the hybrid strategy was the only one that could win all other strategies, the probabilistic technique comes next. Finally, processing requirements considering memory consumption and cpu utilization time are investigated.