Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!uunet!hearst.acc.Virginia.EDU!murdoch!uvacs.cs.Virginia.EDU!mjd4c
From: mjd4c@uvacs.cs.Virginia.EDU (Michael J. Daniel)
Subject: Re: About Hollands book 'Adaptation in ...'
Message-ID: <D027DE.M41@murdoch.acc.Virginia.EDU>
Sender: usenet@murdoch.acc.Virginia.EDU
Organization: University of Virginia Computer Science Department
References:  <71639.s9101799@mail.student.utwente.nl>
Date: Wed, 30 Nov 1994 02:39:14 GMT
Lines: 27

In article <71639.s9101799@mail.student.utwente.nl>, "Albert van Breemen" <s9101799@mail.student.utwente.nl> writes:
> Introducing me: Albert van Breemen, student with the Department of Electrical 
>                 Engineerinf, University of Twente, the Netherlands
> Interest:       AI, at the moment specially GA
> 
> Problem:
> 
> I've read the most of Hollands book 'Adaptation in natural and 
> artificial systems'. While reading the book the following question 
> did arise: GA works only good if their is some correlation between f(c) and
> c, with c a structure of A. If their is no correlation then the search is
> just random (This is just mine opinion which I have not proved or neither 
> I've read any prove, but I want this to prove mathematicaly soon). 

There is a tsp problem on softlib.rice.edu in /pub/tsp/source/eil101.tsp.
It seems that when you get close to the solution many paths have the 
same cost. So you end up with a population, all the same fitness,
but all different paths.  The selection pressure is gone.
It's just a random search at that point.

We've also gotten to within 3 distance units of the optimal and
still are ~25 edges off.

Michael



