R help - Nov 2007 - How can I fitting this function from values.

Hi all, I'd like fit this function:

G(t)=A*1/(1+t/B)*1/sqrt(1+t/C)

where A, B and C are fitting constants that one like search. I have
got a fcs<-(t,G(t)) value table in a data frame.

Any one can help me? Tahnks Horacio.

On Mon, Nov 12, 2007 at 04:27:26PM -0300, Horacio Castellini
wrote:
> Hi all, I'd like fit this function:
> 
> G(t)=A*1/(1+t/B)*1/sqrt(1+t/C)
> 
> where A, B and C are fitting constants that one like search. I have
> got a fcs<-(t,G(t)) value table in a data frame.
> 
> Any one can help me? Tahnks Horacio.
if your data are in `df' and the columns are labeled `t' and `G'
this seems what you want (cf. `?nls'):

nls(G ~ A*1/(1+t/B)*1/sqrt(1+t/C), start =list(A=1, B=2, C=3), data = df) 

where `start' should be set to reasonable start values.

and I would think, it is better to use p1 = 1/B, p2= 1/C during the fit
(no singularities of G for certain parameter values...).

joerg

On 13/11/2007, at 8:27 AM, Horacio Castellini wrote:
> Hi all, I'd like fit this function:
>
> G(t)=A*1/(1+t/B)*1/sqrt(1+t/C)
>
> where A, B and C are fitting constants that one like search. I have
> got a fcs<-(t,G(t)) value table in a data frame.
>
> Any one can help me? Tahnks Horacio.

I ***thought*** that nls would solve your problem in a trice.
But I tried a toy example before replying to you, and it messed up
mightily:

foo <- function(t,ccc){
ccc[1]*1/(1+t/ccc[2])*1/sqrt(1+1/ccc[3])
}
ttt <- seq(1,10,length=100)
yyy <- foo(ttt,c(5,3,2))
set.seed(42)
dat.test <- data.frame(t=ttt,y=yyy+rnorm(100,0,0.1))
with(dat.test,plot(t,y))
with(dat.test,lines(t,foo(t,c(5,3,2)))) # Looks cool.
fit.test <- nls(y~A*1/(1+t/B)*1/sqrt(1+t/C),start=list 
(A=5,b=3,C=2),data=dat.test)
# Gives error; no idea what the implications are.

The error message:

Error in `[[<-.data.frame`(`*tmp*`, var, value = c(0.70934153524875,  
0.761288405463172,  :
   replacement has 65 rows, data has 100
In addition: Warning message:
In t/B : longer object length is not a multiple of shorter object length

I don't think I have *ever* had a good experience with nls(); it is  
always pretty flaky.
(Well, maybe it worked for me once ..., can't remember for sure. :-) )

		cheers,

			Rolf Turner


######################################################################
Attention:\ This e-mail message is privileged and confid...{{dropped:9}}