Re: NEC-LIST: Genetic Algorithms

From: Bill Comisky <bcomisky_at_email.domain.hidden>
Date: Wed, 20 Oct 1999 20:52:46 -0700 (PDT)

On Wed, 20 Oct 1999, Keith Lysiak wrote:

> How many real variables will your matlab routine optimize
> simutaneously?

I've used it for anywhere from 2 to >20 variables. Some of the
variables may have had little affect on the solution though (sometimes
you get "vestigial" structures that have no obvious benefit, but which
do not hurt either). For large numbers of variables you may have to
tune the GA parameters to increase exploration of your search space
(larger population, higher mutation, etc), and it will take longer to
converge.

There is no inherent limit in the code on the number of optimization
variables. I've tried using some of matlab's constrained optimizers,
and the GA seems much more robust as you increase the number of
variables. It's worked well for me for a large variety of problems.

BTW, the GA is trivially parallel; that is you can spread the figure
of merit evaluations across N processors and achieve a nearly N
reduction in total execution time. I've used up to 20 processors in a
Beowulf cluster with this GA (the job/host bookkeeping code you need
to do this is not built into the GA itself, it was separate).

Bill

--
Bill Comisky
bcomisky_at_pobox.com
Received on Thu Oct 21 1999 - 01:44:37 EDT

This archive was generated by hypermail 2.2.0 : Sat Oct 02 2010 - 00:10:39 EDT