Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.ecn.bgu.edu!news.moneng.mei.com!howland.reston.ans.net!news.cac.psu.edu!news.pop.psu.edu!hudson.lm.com!newsfeed.pitt.edu!dsinc!ub!freenet.buffalo.edu!ah818
From: ah818@freenet.buffalo.edu (Phillip M. Fries)
Subject: Re: Diversity vs optimum answer, and time
Message-ID: <D6C98M.JrD@freenet.buffalo.edu>
Sender: nntp@acsu.buffalo.edu
Nntp-Posting-Host: freenet.buffalo.edu
Reply-To: ah818@freenet.buffalo.edu (Phillip M. Fries)
Organization: State University of New York At Buffalo, NY (USA)
References: <3lgptp$j0j@macondo.dmu.ac.uk>  
Date: Sat, 1 Apr 1995 04:17:08 GMT
Lines: 41


In a previous article, hgs@dmu.ac.uk (Hugh Sasse) says:

>
>I have tried a number of problems with GA's and have always run
>into:
>	Saturation (== very low diversity) at very suboptimal 
>		solutions.
>	Sometimes the things won't get fitter than the best of
>		the initial random selection in reasonable time.
>	increasing the population to improve accuaracy only
>		slows things down and doesn't really help
>	decreasing the selectivity to maintain diversity means
>		that results of crossovers have lousy fitness.
>	doing the opposites of the last two to speed things up
>		just makes the diversity drop to 1 really soon
>		at a poor result.
>
Hello Hugh,

I don't know if I can help you out too much, being new at this myeself. 
But one thing I've found is that tournament selection combined with 
simple crossover and mutation qued to increase as the population becomes 
more homogeneous is an approach that is considerably faster and more 
productive with the problems I've worked on so far. This approachalso 
accomplishes this for me with population sizes as low as fifty, and I 
generally run a population size of 100-150. Increasing beyond that size 
doesn't seem to make much difference. Still, sometimes I find a solution 
within 3,000 chromes (my tournament selection produces 2 chromes at a 
crack from two parents selected from a random group of 6-10 pool chromes 
which replace two chromes selected  from a random group of 3-5 pool 
chromes) and at other times I never find an answer. I generally find so 
far that I won't get much improvement after 10,000 chromes - though 
ocassionally a jump will be made higher up the chrome count. On some 
problems I find that an _extremely_ high mutation rate will keep the pool 
fed with new blood (I mutate 1 gene per child based on a similarity 
ratio of the two parents).

-- 
Phillip M. Fries
ah818@freenet.buffalo.edu
