Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!newsgate.duke.edu!news.mathworks.com!newsfeed.internetmci.com!in2.uu.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: Question on Encoding Input Data
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <DuAv18.yA@unx.sas.com>
Date: Tue, 9 Jul 1996 23:39:08 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <1996Jul4.134853.1@otago.ac.nz> <31E0FB14.F35@gauss.crmpa.unisa.it>
Organization: SAS Institute Inc.
Lines: 29


In article <31E0FB14.F35@gauss.crmpa.unisa.it>, Marsella Marco <marmar@gauss.crmpa.unisa.it> writes:
|> xiaodong@otago.ac.nz wrote:
|> > 
|> > Hi, there. I have a problem regarding encoding input data. For example in the
|> > Breast cancer data set, there is an attribute, breast-quad, which can possibly
|> > have following values:left-up, left-low, right-up, right-low, central. What is
|> > the best way encoding them into numerical values?
|> 
|> I think that the better way to encode your information is a fuzzy set.
|> If you think the breast-quad as fuzzy set, you can use a membership function
|> like guaussian to represent the value left-up , left-low etc. 
|> You need to define 5 node each of them with a gaussian like e^( (x-m )^2/s^2). 
|> For each node you must define m and s, that are average and variance of gaussian.
|> To fix these value you need Neural-Gas or Kohonen algorithm.

And Marsella Marco must need a sledgehammer to kill a fly. If 
spatial position is relevant, do what Greg Heath said:
|> [Use] two nodes: (-1,1), (-1,-1), (1,1), (1,-1), and (0,0).

If spatial position is not relevant, use 4 binary inputs as
explained in the FAQ ("How should categories be coded?" in
ftp://ftp.sas.com/pub/neural/FAQ2.html).

-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
