Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!news.kei.com!simtel!col.hp.com!news.dtc.hp.com!hpscit.sc.hp.com!hpax!news1.boi.hp.com!hp-pcd!hpcvsnz!news
From: 5LG3 (Terrill Snyder)
Subject: Re: Classifier: Coding an environment represented by continuous variables
Sender: news@hpcvsnz.cv.hp.com (News )
Message-ID: <D4EG4L.C1n@hpcvsnz.cv.hp.com>
Date: Wed, 22 Feb 1995 11:33:57 GMT
References: <3htr5t$6dm@newsbf02.news.aol.com>
Nntp-Posting-Host: seanm.cv.hp.com
Organization: InkJet Business Unit (IJBU)
X-Newsreader: WinVN version 0.82
Lines: 33

In article <3htr5t$6dm@newsbf02.news.aol.com>, stephynb@aol.com (StephynB) says:
>
>I am interested in developing a classifier system which uses some
>continuous variables in the environmental message such as price level
>(121.3), interest rates (3.2 percent ), GDP growth (2.9 percent), and
>unemployment (5.2 percent) rate as well as inequality relations ( less
>than 4.2)
>
>I have been unable to find examples for coding such types of variables in
>the literature referred to in the FAQ. Almost 100% of the time, the
>environmental variables have been mapped onto various discrete valued
>states: on or off. Holland's "insect eater" comes to mind. Goldberg uses a
>multiplexer as an example.
>
>Intuitively, I realize that when we see a number "7.2" and it refers to
>unemployment, we assign it into the group "high unemployment." This would
>seem to suggest the use of a multivalued logic and sets. I am somewhat
>resistant to this track because it would seem that the membership
>functions would have change over time (remember when double digit
>inflation was high? now we think 4.0 percent will ruin the country) as a
>result of increased experience and changes in expectations.
>
>Any suggestions of how I might go about it? I need some help along the
>lines of a recipe -- most journal articles seem to skip the section on
>combining the ingredients. 
>
>Thank - you very much in advance.
>
>stephyn.butcher@interramp.com

I would just use percent change from the previous time period, since change
is very important, and still has all of the information.

