Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!oitnews.harvard.edu!purdue!lerc.nasa.gov!magnus.acs.ohio-state.edu!math.ohio-state.edu!cs.utexas.edu!news.sprintlink.net!tank.news.pipex.net!pipex!uknet!newsfeed.ed.ac.uk!ainews!andrewt
From: andrewt@aisb.ed.ac.uk (Andrew Tuson)
Subject: Re: Advice on GA methods.
Message-ID: <DD5F96.Bny@aisb.ed.ac.uk>
Sender: news@aisb.ed.ac.uk (Network News Administrator)
Organization: Dept AI, Edinburgh University, Scotland
References: <808140731snz@mayfair.demon.co.uk>
Date: Fri, 11 Aug 1995 13:39:05 GMT
Lines: 63

In article <808140731snz@mayfair.demon.co.uk> thasan@mayfair.demon.co.uk
writes:

Hi,

>I have recently developed a GA based optimizer for technical analysis.

[snip]

>For crossover, I used simple roulette whell selection and then simple
>mutation.

I think that you forgot to mention what crossover method you used...:-)

What was it BTW?

I would advise against using roulette wheel selection, as it is prone to
sampling errors. Baker`s SUS algorithm would be a better choice. The use of
elitism is usually a win in my experience too. I would advise you to read the
following:

@incollection{Hancock94,
   author = "Peter J. B. Hancock",   
   title = "An empirical comparison of selection methods in
                  evolutionary algorithms",
   booktitle = "Selected Papers: AISB Workshop on Evolutionary
                  Computing, Lecture Notes in Computer Science No 865",
   publisher = "Springer Verlag",
   year = 1994,
   editor = "Terry C. Fogarty",
   pages = {80-94}}

>Is there any better technique? i.e any alternative crossover methods,
>any way of mutating dynamically e.t.c.

Without knowing more about your problem I would not be able to help. One thing
to bear in mind is that binary strings are often not the appropriate
representation to use for practical problems. Try representing your problem
at a higher level - that might help.

One way of looking at this is that the GA exploits the information about the
that you provide it. This information is contained in the representation,
operators and fitness function, the most convenient place to put the
information will vary; the trick is to give it in a form that the GA can use..

>Though Goldberg suggests other techniques such as :
>        Deterministic Sampling, stochastic sampling with or without
>replacement e.t.c.
>I am really not sure which one to go for.

I hate to say that many questions such as these have not been definitely
answered yet! Which makes it more fun for us GA researchers, but harder for
those who have to solve real problems with GAs..:-|

Anyway, back to my MSc dissertation..:-(

Hope this helps..
-- 
Andrew Tuson (andrewt@aisb.ed.ac.uk)

Department of Artificial Intelligence, University of Edinburgh, Scotland, U.K.
An expert is a person who avoids the small errors while sweeping on to the
grand fallacy..........:-)
