Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!uunet!munnari.oz.au!newsroom.utas.edu.au!pacit.tas.gov.au!aries.dpi.tas.gov.au!bhorton
From: bhorton@aries.dpi.tas.gov.au
Subject: Re: Two crossovers better than one?
Organization: Department of Primary Industry and Fisheries
Date: Thu, 23 Feb 1995 04:18:05 GMT
Message-ID: <bhorton.4.000F4DA2@aries.dpi.tas.gov.au>
Lines: 34
X-Newsreader: Trumpet for Windows [Version 1.0 Rev A]
References:  <3i11qt$8no@darkstar.UCSC.EDU>
Sender: usenet@pacit.tas.gov.au (News User)

In article <3i11qt$8no@darkstar.UCSC.EDU> mpcline@cats.ucsc.edu (Matthew Paul Cline) writes:
>From: mpcline@cats.ucsc.edu (Matthew Paul Cline)
>Subject: Two crossovers better than one?
>Date: 17 Feb 1995 02:31:57 GMT

>Does splitting crossover between two crossover operators ever generate
>better results than just using one or the other? (i.e., for genetic
>operator mixes, 35% 1-point crossover, 35% two-point crossover, and 30%
>mutation, vs. 70% 1-point crossover and 30% mutation) If anyone has any
>examples of this, what type of problem was the GA used for?  Thanks in
>advance.

I tested a system using from 1 to 20 crossovers and found no consistant effect 
on the rate of approach to the optimum, although 1 crossover was possibly 
not as good as 2 to 20 crossovers.  However, there was no difference in the 
optimum reached. This was disappointing because I had previously spent 
considerable time trying to optimize the system by putting 'genes' with 
similar effects as close as possible, so that they could stay linked when good 
sets of genes came together.  This ended up with a circular string, which 
required an even number of crossovers.  However, I also tested linear strings 
with odd and even numbers of crossovers and also scrambled strings with 
related genes as far apart as possible.  None of these changes made much 
difference to the efficiency of the system.

I use 17 genes of various sizes giving a total of 80 bits in the string, and a 
population size of 300.  The optimum is usually reached after 80 to 120 
generations.

The genetic algorithm is used to find the optimal structure for a sheep 
breeding system.  This causes considerable trouble explaining it to sheep 
breeders who confuse breeding sheep with breeding breeding-strategies.

Brian Horton

