Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!sunic!news.chalmers.se!news.gu.se!gd-news!d6242.shv.hb.se!sa209
From: sa209@utb.shv.hb.se (Claes Andersson)
Subject: Re: Mutation, Crossover,Pop Size
Message-ID: <sa209.86@utb.shv.hb.se>
Sender: usenet@gdunix.gd.chalmers.se (USENET News System)
Nntp-Posting-Host: d6242.shv.hb.se
Organization: Department of Scocial Science
References: <3biulp$2pg@nova.umd.edu>
Date: Fri, 2 Dec 1994 17:19:21 GMT
Lines: 21

In article <3biulp$2pg@nova.umd.edu> collins@nova.umd.edu (Jim C) writes:

>I have been developing a GA for predicting movement in the OEX 
>(S&P 100 stock index) with some success (no you can't have the code)
>but I am having a problem determining some of the variables, such as Crossover rate, mutation rate, initial population size..
>Are there any widely-accepted rules of thumb for these? Esp the
>initial population size.

>Jim C.

 The most commonly used method is one-point crossover. Just let y be a 
stochastic variable, 0<y<n, n=the length of the genome. Take the y first 
genes from parent 1 and the rest from parent 2 = The offspring.

 I don't know if there is any rule for the population size.. I'd suggest 
that you try it out. A too small environment will converge fast ut then you 
can always diverge it. I'm working on a good method for divergeing 
converged GA's right now that shows considerably better performance than the 
usual heat/cool method.

Claes Andersson. University of Bors. Sweden.
