[R] 'singular gradient matrix? when using nls() and how to make the program skip nls( ) and run on

[Joerg van den Hoff]

On Wed, Sep 05, 2007 at 04:43:19PM -0700, Yuchen Luo wrote:
> Dear friends.
> 
> I use nls() and encounter the following puzzling problem:
> 
> 
> 
> I have a function f(a,b,c,x), I have a data vector of x and a vectory  y of
> realized value of f.
> 
> 
> 
> Case1
> 
> I tried to estimate  c with (a=0.3, b=0.5) fixed:
> 
> nls(y~f(a,b,c,x), control=list(maxiter = 100000, minFactor=0.5
> ^2048),start=list(c=0.5)).
> 
> The error message is: "number of iterations exceeded maximum of 100000"
> 
> 
> 
> Case2
> 
> I then think maybe the value of a and be are not reasonable. So, I let nls()
> estimate (a,b,c) altogether:
> 
> nls(y~f(a,b,c,x), control=list(maxiter = 100000, minFactor=0.5
> ^2048),start=list(a=0.3,b=0.5,c=0.5)).
> 
> The error message is:
> 
> "singular gradient matrix at initial parameter estimates".
> 
> 
> 
> This is what puzzles me, if the initial parameter of (a=0.3,b=0.5,c=0.5) can
> create 'singular gradient matrix', then why doesn't this 'singular gradient
> matrix' appear in Case1?
> 
> 
> 
> I have tried to change the initial value of (a,b,c) around but the problem
> persists. I am wondering if there is a way out.
> 
> 
> 
> My another question is, I need to run 220 of  nls() in my program with
> different y and x. When one of the nls() encounter a problem, the whole
> program stops.  In my case, the 3rd nls() runs into a problem.  I would
> still need the program to run the remaining 217 nls( )! Is there a way to
> make the program skip the problematic nls() and complete the ramaining
> nls()'s?

?try

> 
> 
> 
> Your help will be highly appreciated!
> 
> Yuchen Luo
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.