Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!news.alpha.net!uwm.edu!caen!hearst.acc.Virginia.EDU!murdoch!uvacs.cs.Virginia.EDU!mjd4c
From: mjd4c@uvacs.cs.Virginia.EDU (Michael J. Daniel)
Subject: Re: A Question
X-Nntp-Posting-Host: cobra-fo.cs.virginia.edu
Message-ID: <D4IFy2.90G@murdoch.acc.Virginia.EDU>
Sender: usenet@murdoch.acc.Virginia.EDU
Organization: University of Virginia Computer Science Department
References:  <1995Feb19.153912.24031@dcs.warwick.ac.uk>
Date: Fri, 24 Feb 1995 15:20:26 GMT
Lines: 38

In article <1995Feb19.153912.24031@dcs.warwick.ac.uk>, P.N.Randell@dcs.warwick.ac.uk (Paul N Randell) writes:
> 
> I am currently working on a preprocessing element for the GENITOR TSP package.
> It produces an enriched initial population without loosing the diversity
> that allows the GA to explore the problem space.
> At the moment i'm producing tours of equivalent or shorter length than
> those produced by GENITOR on its own and in a reduced number of generations.
> 

What is GENITOR TSP package and where can I get it?

Ive considered altering the initial populations genetic structure
and other techniques for improving the search, 
but my thesis advisor suggested that I would have 
to show that they were of general applicability, 
(ie either theorically sound analysis, 
or lots of experimental results). 
Neither was appropriate for a master's thesis about
communication among subpopulations.

There is definite room for improvement, by altering 
the populations genetic makeup in some way.
Considering 1) that one reason improvement stops 
is allele fixing, ie a loss of 
necessary "good" alleles from the population,
and 2) that mutation is essentially the same idea.

> One possible interpretation of this is that natural selection is not the
> best method for producing small scale building blocks (or sub tours in 
> this case) ,what ever small scale may be taken as meaning.

Certainly. Natural selection is just a first good guess at how to find
good answers without intelligent supervision.



Michael

