Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!howland.reston.ans.net!news.starnet.net!wupost!heino.stl.westar.com!not-for-mail
From: Frank Dries <dries@westar.com>
Subject: Re: abductive modeling and aim
Message-ID: <9197cb$11b1c.3bc@heino.stl.westar.com>
Date: Mon, 25 Sep 1995 23:11:28 GMT
X-Mailer: Mozilla 1.2b5 (Windows; I; 16bit)
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Lines: 34

Will Dwinnell <predictor@delphi.com> writes:
>David King <dgk@cmsnet.com> writes:
> 
>>I also looked at purchasing AIM a few months ago, but learned that the product
>>was very limited in the number of variables (I believe < 100) and observations
>>(~3,000) that it could accommodate.  Since we often deal with hundreds of inputs
>>and a large number of observations, I did not pursue it any further.  (I guess
>>I have a different meaning for "complex problems").
 
>.
>Well, all of this depends on what you are counting as "one input
>variable", since it's not like a variable is a variable is a variable...
>I have found that AIM (for Windows) does an adequate job at variable
>selection, given the total number of inputs it can handle as candidates.
>Remember that there is an enormous difference between 100 uncorrelated inputs
>and 100 typical real-world problem variables.  A model which used (and in
>fact required) 100 uncorellated variables would require an astronomical
>amount of training data in a worst case scenario.  It is true that AIM
>may not be able to whittle down a large number of candidate variables
>for you, and you may be stuck doing this or some other sort of
>data reduction (like PCA) on your own.

If you don't want to implement all of the feature reduction/selection
algorithms yourself, you might give PARTEK a try. PARTEK is a great
software package and includes many visual and numerical methods for
choosing/transforming input variables prior to modeling (including PCA).
Also, with PARTEK there is no limit to the number of variables or
observations.

Regards,
Frank Dries
Westar Corporation



