Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!news.sprintlink.net!in1.uu.net!tank.news.pipex.net!pipex!uknet!newsfeed.ed.ac.uk!ainews!andrewt
From: andrewt@aisb.ed.ac.uk (Andrew Tuson)
Subject: Re: RE: Representation of a problem before a GA search
Message-ID: <DDzAHz.86D@aisb.ed.ac.uk>
Sender: news@aisb.ed.ac.uk (Network News Administrator)
Organization: Dept AI, Edinburgh University, Scotland
References: <1995Aug23.150447.1@otago.ac.nz> <000019d1+0000036a@msn.com> <41ohju$3tq@steel.interlog.com>
Date: Sun, 27 Aug 1995 16:44:22 GMT
Lines: 41

In article <41ohju$3tq@steel.interlog.com> Bruno Degazio
<bruno@accesspt.north.net> writes:

>I am actually interested in more complicated data as the 'genes', where 
>each gene indicates an operation on a set of musical data. Do I treat 
>the data structures as 'raw bit strings' or should my GA operators 
>(crossover, mutation, inversion) 'know' something about the structure of 
>the data?

I would take the latter approach - quite simply because the GA depends upon
fitness correlations in the representation/operator set in order to optimise
effectively. These are more likely to be present in a high level
representation.

Or to put it another way - the representation, operators and fitness function
all provide information to guide the search. Why have the GA work harder than
it has to when you can give it explicit information on how to perform the
search?

I also recommend the following paper which argues that GA representations
should represent the problem space:

@TechReport{Radcliffe92,
  author = 	 "N. J. Radcliffe",
  title = 	 "Non-Linear Genetic Representations",
  institution =  "Edinburgh Parallel Computing Centre",
  year = 	 "1992",
}

You should be able to get this off EPCC`s web site (at the Univ. of Edinburgh).

[This may have been published elsewhere - anyone?]

Hope this helps.

-- 
Andrew Tuson (andrewt@aisb.ed.ac.uk)

Department of Artificial Intelligence, University of Edinburgh, Scotland, U.K.
An expert is a person who avoids the small errors while sweeping on to the
grand fallacy..........:-)
