Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!spool.mu.edu!agate!news.ucdavis.edu!sunnyboy.water.ca.gov!sunnyboy!nsandhu
From: nsandhu@venice.water.ca.gov (Nicky Sandhu)
Subject: Output range in SNNS ??
Message-ID: <NSANDHU.94Dec30102759@grizzly.water.ca.gov>
Sender: news@sunnyboy.water.ca.gov
Organization: Calif. Dept. of Water Resources
Date: Fri, 30 Dec 1994 18:27:59 GMT
Lines: 17

Hi,
	I am using SNNSv3.3 to model some time - series data. I have
been squashing the data to the range ]0,1[ and using the feedforward
net with some success.
	In order to standardize the input and output it was suggested
to me that I normalize the data to zero mean and unit variance. This
also squashes the data, though not to the range ]0,1[.
	I modified the feed-forward net in SNNS by changing the
activation function in the output unit to Act_Identity_PlusBias
function. The activation function in all other units is the
Act_Logistic function. The output functions for all units is
Out_Identity. Using only one hidden layer (2 neurons) I attempted to
train on the data set. The hidden units at once jumped to activation
of zero and the output was a constant (2.xx) over all the data points.
	Has anyone tried this out before? I would really appreciate
any help. Thanks....
-Nicky
