Using Genetic Algorithms to Play Rock-Paper-Scissors
Complex Adaptive Systems
CS 591, Section 8
Assignment 1
February 16, 2006
1 Introduction
The idea of this assignment is to use the genetic algorithm to evolve strategies for playing a sim-
ple two-person game known as Rock-Paper-Scissors (RPS). In RPS, each player has a choice of
three alternatives: Rock, Paper, and Scissors. After each player has made her choice, the winner
is determined as follows: Rock beats Scissors (by smashing the scissors); Paper beats Rock (by
wrapping the Rock); Scissors beats Paper (by cutting the paper). For one round of the game, all
choices have equal probability of winning, so no strategy is favored. However, the game becomes
more interesting when there are repeated interactions between the players. In this assignment we
will be as interested in the dynamics of a population of RPS players as we are in evolving a single
rule that plays the game well.
2 Assignment
You are free to choose any publicly available genetic algorithm software or to write your own using
the language of your choice. One possible choice is a very simple genetic algorithm written in C by
Gary Flake (see URLs below). If you choose to use Flake’s code, please note that it generates FSAs
for playing the Prisoner’s Dilemma. You will be using a different representation to play a different
game.
1. Due: March 1 Modify an existing GA (or write your own) to learn strategies for RPS.
(a) Test the GA against a simple Nash strategy (the strategy that chooses among the three
alternatives with equal probability). That is, evaluate an individual’s fitness by measur-
ing its success playing against the Nash strategy
(b) Change the fitness function so that an individual iplays against xother individuals
on each generation to determine i’s fitness. The best value for xwill depend on your
population size—-you will need to experiment with different values of xto learn which
values produce the most interesting dynamics. Each ivs. xinteraction consists of a
repeated RLP game, say for 51 rounds.
2. Due: March 13 Extend the model to include other representations and space.
(a) Design a representation that includes the Nash strategy. e.g., by encoding the strategy
as a set of probabilities, implement the strategy, and compare its performance with the
performance of the representation from Part I.
(b) Add space to the model, so that the population occupies a 2-Dimensional array, with one
individual (or a small number of individuals) occupying every cell in the array. Study
the effects of adding spatial structure to the population.
1
