[R] nls

[Prof Brian Ripley]

On Tue, 15 Aug 2006, Peter Dalgaard wrote:

> Prof Brian Ripley <ripley at stats.ox.ac.uk> writes:
> 
> > You problem is x^c for x = 0.  If you intended only c > 1, try a starting 
> > value meeting that condition (but it seems that the optimal c is about 
> > 0.27 is you increase x slightly).
> 
> Surely you mean c > 0.

I did.

> >  nls(1/y ~ a+b*x^exp(c), start=list(a=1.16122,b=0.01565,c=0))
> Nonlinear regression model
>   model:  1/y ~ a + b * x^exp(c)
>    data:  parent.frame()
>          a          b          c
>  0.9944025  0.1953168 -1.1495206
>  residual sum-of-squares:  0.03303464
> >  nls(1/y ~ a+b*x^c, start=list(a=1.16122,b=0.01565,c=exp(-1.1)))
> Nonlinear regression model
>   model:  1/y ~ a + b * x^c
>    data:  parent.frame()
>         a         b         c
> 0.9944026 0.1953165 0.3167891
>  residual sum-of-squares:  0.03303464
> 
> (but even setting c=exp(-1) triggers the error; there could be cause
> for robustifying the nls algorithm)

Well, there is an option to use a bounded-region algorithm, e.g.

x <- c(0,5,10,15,20,25,30)
y <- c(1.00000,0.82000,0.68000,0.64000,0.66667,0.68667,0.64000)
nls(1/y ~ a+b*x^c, start=list(a=1.16122,b=0.01565,c=1), trace=TRUE,
    algorithm="port", lower=c(-Inf, -Inf, 0), upper=rep(Inf, 3))

works.

>  
> > Why have you used ~~ ?  (Maybe because despite being asked not to, you 
> > sent HTML mail?)
> > 
> > On Tue, 15 Aug 2006, Xiaodong Jin wrote:
> > 
> > >   Is there anyway to change any y[i] value (i=2,...6) to make following NLS workable? 
> > >    
> > >   x <- c(0,5,10,15,20,25,30)
> > >   y <- c(1.00000,0.82000,0.68000,0.64000,0.66667,0.68667,0.64000)
> > >   lm(1/y ~~ x)
> > >   nls(1/y ~~ a+b*x^c, start=list(a=1.16122,b=0.01565,c=1), trace=TRUE)
> > >    
> > >   #0.0920573 :  1.16122 0.01565 1.00000 
> > > #Error in numericDeriv(form[[3]], names(ind), env) : 
> > > #        Missing value or an infinity produced when evaluating the model
> > > 
> > >  			
> > > ---------------------------------
> > > 
> > > 
> > > 	[[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.
> > > 
> > 
> > -- 
> > 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
> > 
> > ______________________________________________
> > 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.
> > 
> 
> 

-- 
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