[From nobody Sat Mar 18 16:42:07 2006
Message-ID: <408D4058.1080007@colorado.edu>
Date: Mon, 26 Apr 2004 11:01:12 -0600
From: Erik Johnson <ebjohnso@colorado.edu>
User-Agent: Mozilla Thunderbird 0.5 (X11/20040208)
X-Accept-Language: en-us, en
MIME-Version: 1.0
CC: r-help@stat.math.ethz.ch
Subject: [R] nnet question
References: <Pine.LNX.4.44.0404260702530.22842-100000@gannet.stats>
In-Reply-To: <Pine.LNX.4.44.0404260702530.22842-100000@gannet.stats>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

I am using R 1.8.0, and am attempting to fit a Neural Network model of a 
time series (here called Metrics.data).  It consists of one time series 
variable run on its lag (AR(1)).  Basically, in an OLS model it would 
look like
Metrics.data$ewindx ~ Metrics.data$ewindx.lag1
However, I am trying to run this through a neural network estimation.  
So far, I have been getting convergence very quickly, and do not believe 
it too be true.
Here is the code and output.  Please note that I am using all of the 
values for training and testing in one matrix, as I do not care about 
the testing results right now, I only want to capture weights.  Here is 
the code and output

 > nnet(metrics.data$ewindxlag1,metrics.data$ewindx,size=2, entropy=FALSE)
# weights:  7
initial  value 78858370643.085342
final  value 78841786515.212158
converged
a 1-2-1 network with 7 weights
options were -

When I run the iris3 example, the convergence looks much nicer 
(consisting of more than one iteration).  Am I missing some fundamental 
understanding of this example?  Thanks for any input.

]