[R] CVnn2 + nnet question

Manoj - Hachibushu Capital Wanzare at HCJP.com
Mon Jun 14 13:52:24 CEST 2004


Thanks for the prompt reply. I don't think I am using any non default
arguments but still...here is the exact syntax for two different case
(skip = T and default) and the corresponding error message that I
receive.


Default Case ("skip = F")

Syntax	: nn.test	<-
CVnn2(Y~X1+X2+X3+X4+X5+X6+X7+X8+X9+X10+X11,all.data.norm[1:train.set,],m
axit=500,nreps=10)

Error message: 
	Fold 1
		Size = 0 decay = 0
		Inner fold 1Error in nnet.default(x,y,w.....)  : No
weights to fit.



Case where Skip = T:

Syntax:	nn.test <-
CVnn2(Y~X1+X2+X3+X4+X5+X6+X7+X8+X9+X10+X11,all.data.norm[1:train.set,],m
axit=500,nreps=10,skip=T)

Error message:
	Fold 1 
	Size=0 decay = 0
	Inner fold 1Error in switch(type,raw=z,class= { : inappropriate
fit for class


Manoj 	

-----Original Message-----
From: Prof Brian Ripley [mailto:ripley at stats.ox.ac.uk] 
Sent: Monday, June 14, 2004 8:26 PM
To: Manoj - Hachibushu Capital
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] CVnn2 + nnet question

You are trying to fit a neural network with no connections.  That makes
no 
sense unless skip=T (as on the page you are quoting).

On Mon, 14 Jun 2004, Manoj - Hachibushu Capital wrote:

> Hi,
> 	I am trying to determine the number of units in the hidden layer
> and the decay rate using the CVnn2 script found in MASS directory
> (reference: pg 348,MASS-4). 
> 
> 	The model that I am using is in the form of Y ~ X1 + X2 + X3...
> + X11 and the underlying data is time-series in nature.
> 
> 	I found the MASS and nnet package extremely useful (many thanks
> to the contributors). 

`contributors'?  Do see the DESCRIPTION file: it is hardly anonymous
work.

> 	However I am getting an error while using the CVnn2
> function...it says 

You have not shown us the call you used, but the default arguments are
size = rep(6, 2), lambda = c(0.001, 0.01) so you have used non-default 
args without telling us what.

> Fold 1 
> Size = 0, decay = 0, inner fold 1 Error in nnet.default(x,y,w,....):
No
> weights  to fit.
> 
> 	Obliviously I am doing something wrong but am not able to figure
> it out. 

Please do re-read the basic description of neural nets and nnet().

> Do I have pass any weights?  I am bit confused since the
> documentation of nnet suggests says "Wts: initial parameter vector. If
> missing chosen at Random". Has anybody faced the same error? I am
using
> the latest R version on Linux box. 

-- 
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




More information about the R-help mailing list