Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!princeton!flagstaff.princeton.edu!bagalman
From: bagalman@flagstaff.princeton.edu (Michael D. Bagalman)
Subject: commercial software
Message-ID: <1995Mar20.203047.606@Princeton.EDU>
Summary: seeking advice on software purchase 
Originator: news@hedgehog.Princeton.EDU
Sender: news@Princeton.EDU (USENET News System)
Nntp-Posting-Host: flagstaff.princeton.edu
Organization: Princeton University
Distribution: usa
Date: Mon, 20 Mar 1995 20:30:47 GMT
Lines: 32

The last question I asked brought me a good number of responses, for 
which I thank the members of this newsgroup.  If I may,  would like to 
seek advice on yet another question.

I need to model a response (output) for which I have over 100 inputs 
(which I hope to pare down to 10 or 20) and thousands of observations.  I 
need to acquire some software which runs on a PC to do this and I've been 
trying to do my "homework" so as to make the correct decision.  So far 
I've narrowed my field to

MATLAB's Neural Network Toolkit (along with MATLAB itself)

SAS

NeuroShell 2

NeuralWare (Pro II or Predict)

and maybe Brainmaker.

Can anyone tell me if any of these is especially good or especially bad?  
NeuralWare costs much more than NeuroShell 2 but I can't tell why.  I'm 
reluctant to invest too much $ in anything which might not solve my 
problem due to limitations on speed or memory on a PC.  I'm not concerned 
about the variety of nnets available, in the sense that I know I want to 
use backprop (feedforward) so I don't need options like RBF or Hopfield, 
although maybe someday I'll have a new problem for which I will want them.

I've got the literature on most of this software, but I'm hoping for some 
independent unbiased views.  Your help is much appreciated.

Michael
