Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!gatech!news.mathworks.com!news.kei.com!news.texas.net!news.sprintlink.net!news-fw-6.sprintlink.net!mv!usenet
From: "Scott Russell" <queuetoo@emrys.mv.com>
Subject: Re: Q: More than one fitness parameter
Message-ID: <01bb8461$1165d0c0$420116cf@Emrys>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Organization: QC
Date: Wed, 7 Aug 1996 13:03:22 GMT
References: <01bb8343$2091e3a0$LocalHost@Emrys> <phwwzc2cc4.fsf@dude.mcs.anl.gov> <4u9kv5$3qt@bignews.shef.ac.uk>
X-Newsreader: Microsoft Internet News 4.70.1141
X-Nntp-Posting-Host: emrys.mv.com
Content-Transfer-Encoding: 7bit
Lines: 35

I said
>> Now comes the problem.  How can I make the fitness function
>> multidimensional?  I had a gut feeling that I could simply calculate
>> a TotalCost for each property of each component, compute fitness for
>> each and average them together. I.E.,

Unfortunately, when I originally posted, I had my bone-head hat on.
crink@netonecom.net was nice enough to point out to me that by simply
scaling my  parameters so each one targets 0 (not 1), my first approach
works.  But now, there seems to be some question as to how solid an
implementation this is.

Gordon D. Pusch (pusch@mcs.anl.gov) said:
>Multicriterion optimization is intrinsically ill-defined. Many
>approaches have been proposed, e.g., a weighted sum of criteria
>(which is sort of what you're doing), sum of squares of criteria,
>etc.; however, each of these approaches contains elements of
>arbitrariness or ambiguity (e.g., with a different set of weights,
>one gets a different solution; what does one do if the various
>``costs'' have differing units, etc.).
>

Different sets of weights resulting in a different solution...  Isn't that
the goal?  If I simply wanted _a_ best solution, I'd be using linear
functions, or some kind of a recursive function.  The reason I'm using GA
is so I can find _all_ of the best solutions - or at least a set big enough
to satisfy whoever asked for the list.  The costs having different units...
Isn't that another simple scaling problem?

As I've mentioned in mail to a few of you, this is my first foray into the
worls of GA, so please don't be offended by my ignorance, but I'm surprised
at the reaction this question has raised. Don't all of you have to deal
with multiple fitness parameters?  I'm not sure how you would make a GA
that actually _did_ anything without them...

