R help - Oct 2003 - nnet behaving oddly

Hi,
  I was trying to use the nnet library and am not sure of whats going
on. I am calling the nnet function as:

 n <- nnet(x,y,size=3,subset=sets[[1]], maxit=200)

Where x is a 272x4 matrix of observations (examples) and y is a 272x1
matrix of target values. However when I look at nnet$residuals they are
off by two orders of magnitude (compared to the output from neural
network code that I already have). Looking at nnet$fitted.values shows
all the values to be 1 (whereas my target values range from 0 to 150).

Am I making an obvious mistake in the way I'm calling the function? Is
the fact that n$fitted.values is all 1's indicating that the NN is doing
a classification? If so how can I make it do quantitation?

The man page mentions that if the response is a factor then it defaults
to quantitation. However my y matrix just contain numbers - so it
should'nt be doing classification.

Any pointers would be appreciated.

Thanks,

-------------------------------------------------------------------
Rajarshi Guha <rxg218 at psu.edu> <http://jijo.cjb.net>
GPG Fingerprint: 0CCA 8EE2 2EEB 25E2 AB04 06F7 1BB9 E634 9B87 56EE
-------------------------------------------------------------------
Psychology is merely producing habits out of rats.

On 20 Oct 2003, Rajarshi Guha wrote:
> Hi,
>   I was trying to use the nnet library and am not sure of whats going
> on. I am calling the nnet function as:
> 
>  n <- nnet(x,y,size=3,subset=sets[[1]], maxit=200)
> 
> Where x is a 272x4 matrix of observations (examples) and y is a 272x1
> matrix of target values. However when I look at nnet$residuals they are
> off by two orders of magnitude (compared to the output from neural
> network code that I already have). Looking at nnet$fitted.values shows
> all the values to be 1 (whereas my target values range from 0 to 150).
> 
> Am I making an obvious mistake in the way I'm calling the function? Is
> the fact that n$fitted.values is all 1's indicating that the NN is
doing
> a classification? If so how can I make it do quantitation?
Yes, so please do read the help page accurately.
> The man page mentions that if the response is a factor then it defaults
> to quantitation. However my y matrix just contain numbers - so it
> should'nt be doing classification.
That's incorrect reading of the help page, which actually says

     If the response in 'formula' is a factor, an appropriate
     classification network is constructed; this has one output and
     entropy fit if the number of levels is two, and a number of
     outputs equal to the number of classes and a softmax output stage
     for more levels.  If the response is not a factor, it is passed on
     unchanged to 'nnet.default'.

and you did not give a formula.

As someone else said recently. those who write the manuals don't expect to 
either read them for you nor re-write them here, so please show more 
consideration for their work.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

> From: Rajarshi Guha [mailto:rxg218 at psu.edu] 
> 
> Hi,
>   I was trying to use the nnet library and am not sure of 
> whats going on. I am calling the nnet function as:
> 
>  n <- nnet(x,y,size=3,subset=sets[[1]], maxit=200)
Please give us output of something like:

str(x)
summary(y)

Also, I believe the subset argument is only meant for calls via 
formula; e.g.,

nnet(y ~ x, ...)

and needs to be a vector of logicals the same length as the number
of rows in x, indicating which rows to include in the fitting.
Please also tell us what sets[[1]] is.

Andy
 
> Where x is a 272x4 matrix of observations (examples) and y is 
> a 272x1 matrix of target values. However when I look at 
> nnet$residuals they are off by two orders of magnitude 
> (compared to the output from neural network code that I 
> already have). Looking at nnet$fitted.values shows all the 
> values to be 1 (whereas my target values range from 0 to 150).
> 
> Am I making an obvious mistake in the way I'm calling the 
> function? Is the fact that n$fitted.values is all 1's 
> indicating that the NN is doing a classification? If so how 
> can I make it do quantitation?
> 
> The man page mentions that if the response is a factor then 
> it defaults to quantitation. However my y matrix just contain 
> numbers - so it should'nt be doing classification.
> 
> Any pointers would be appreciated.
> 
> Thanks,
> 
> -------------------------------------------------------------------
> Rajarshi Guha <rxg218 at psu.edu> <http://jijo.cjb.net>
> GPG Fingerprint: 0CCA 8EE2 2EEB 25E2 AB04 06F7 1BB9 E634 9B87 56EE
> -------------------------------------------------------------------
> Psychology is merely producing habits out of rats.
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list 
> https://www.stat.math.ethz.ch/mailman/listinfo> /r-help
>