Re: [R] retrieving coefficients from nls() when it fails to converge

[Spencer Graves]

     Have you considered "optim"?  That has produced something for me 
when I got nothing from "nls".  You may wish to call "optim" a second 
time with the output from the first;  you might actually get different 
answers.  Please specify "hessian=TRUE", as follows: 

     fit0 <- optim(..., hessian=TRUE)

     Then "eigen(fit$hessian, symmetric=TRUE)" may tell you why "nls" 
did not converge. 

     The standard advice in such cases is to reconsider what you really 
want and then fewer parameters and / or reparameterize;  see e.g., Bates 
and Watts (1988) Nonlinear Regression Analysis and Its Applications 
(Wiley). 

     hope this helps.  spencer graves

Potter, David M wrote:

Hi --

I am fitting four-parameter logistic regression models using nls(),
typically within the body of a loop.  Sometimes a fit fails to converge and
yet I would like to pull off the coefficients from the last iteration and
use them to compute backfitted (i.e., inverse prediction) values.
Note: To ignore the fit completely, but have the loop continue, I have
successfully used try(), but in this case the fits are sufficient for
getting backfitted values (even though the coefficients are poorly
estimated), so I need the actual coefficients.
R information:

platform i386-pc-mingw32
arch     i386           
os       mingw32        
system   i386, mingw32  
status                  
major    1              
minor    7.1            
year     2003           
month    06             
day      16             
language R  

Thanks for your help.

David

__
David M. Potter
Groton Nonclinical Statistics
Pfizer Global Research and Development


LEGAL NOTICE\ Unless expressly stated otherwise, this messag...{{dropped}}

______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
 

______________________________________________
[EMAIL PROTECTED] mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help