Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news4.ner.bbnplanet.net!cam-news-hub1.bbnplanet.com!news.mathworks.com!newsfeed.internetmci.com!in2.uu.net!hearst.acc.Virginia.EDU!murdoch!adder.cs.Virginia.EDU!gjf2a
From: gjf2a@cs.virginia.edu (Gabriel J. Ferrer)
Subject: Re: Tournament Structure When Fitness Evaluation is Expensive
X-Nntp-Posting-Host: colossus-fo.cs.virginia.edu
Message-ID: <DwI8JC.3F9@murdoch.acc.Virginia.EDU>
Sender: usenet@murdoch.acc.Virginia.EDU
Organization: University of Virginia
X-Newsreader: TIN [version 1.2 PL2]
References: <4ud554$120i@core.bard.edu> <32106B58.7AB9@informatik.hu-berlin.de> <4v1d3p$8pb@laplace.ee.latrobe.edu.au>
Date: Wed, 21 Aug 1996 20:21:12 GMT
Lines: 36

My master's thesis investigated the application of genetic programming
to playing boardgames, and incorporated a tournament for fitness
evaluation.  It is located on the Web at:

http://www.cs.virginia.edu/~gjf2a/work/papers/thesis.ps

Here is the abstract:

----------------------------------------------------------------------------
Computer programs for playing boardgames typically utilize a static board 
evaluation function to select their moves for each turn. Devising a good 
board evaluation function is, in general, a difficult problem.  Consequently, 
substantial work has been done on devising methods for automatically 
generating such functions.

Genetic programming is a variation on the genetic algorithm paradigm wherein 
solutions to problems are encoded as computer programs. A population of 
these programs evolves over time. The evolutionary process works by
evaluating the quality of each program's solution to the problem at hand. 
This is referred to as the program's fitness. Programs with higher fitness 
are more likely to survive and propagate. Thus, over time, the overall fitness
of the population should improve.

This work shows that a fairly straightforward application of genetic 
programming results in the evolution of board evaluation functions which 
can play strategy games with an appreciable level of skill. The games used for
our experiments are the ancient Egyptian game of Senet and the modern game 
of Othello. This helps to demonstrate the general viability of the approach 
in two very different game environments.
----------------------------------------------------------------------------

--
Gabe Ferrer                       | Computer Vision Research Group
Graduate Research Assistant       | University of Virginia Computer Science
ferrer@virginia.edu               | http://www.cs.virginia.edu/~gjf2a/
Opinions expressed herein are mine, or so I'm told... 
