R help - Jul 2003 - question about formulating a nls optimization

Dear list, 

I'm migrating a project from Matlab to R, and I'm
facing a relatively complicated problem for nls.  My
objective function is below:
>> objFun <- function(yEx,xEx,tEx,gamma,theta,kappa){
     yTh <- pdfDY(xEx,tEx,gamma,theta,kappa)
     sum(log(yEx/yTh)^2)
   }

The equation is yTh=P(xEx,tEx) + noise.
I collect my data in:
>> data <- data.frame(xEx,tEx,yEx)
And I do the nls:
aux <- nls( ~ objFun(yEx,xEx,tEx,gamma,theta,kappa), 
           data=data,
   start=list(gamma=0.085, theta=8.62*10^(-5),
kappa=4.45*10^(-3)), trace=TRUE)

I get the following error: 
179.8911 :  8.50e-02 8.62e-05 4.45e-03 
Error in sum(rr[-(1:npar)]^2) : subscript out of
bounds

One problem that I identified is that the algorithm
does not change the values of the parameters at all. 
It calculates the objective function OK, but the
parameters are not updated.  

The function pdfDY is a Fourier integral with
parameters xEx, tEx, gamma, theta, kappa, that I 
calculate with a Filon algorithm.  The calculation of
pdfDY is correct.  

Can anybody help me with this?  Thank you very much, 

Adrian

Adi Humbert <adrian_humbert at yahoo.com> writes:
> Dear list, 
> 
> I'm migrating a project from Matlab to R, and I'm
> facing a relatively complicated problem for nls.  My
> objective function is below:
> 
> >> objFun <- function(yEx,xEx,tEx,gamma,theta,kappa){
>      yTh <- pdfDY(xEx,tEx,gamma,theta,kappa)
>      sum(log(yEx/yTh)^2)
>    }
> 
> The equation is yTh=P(xEx,tEx) + noise.
> I collect my data in:
> 
> >> data <- data.frame(xEx,tEx,yEx)
> 
> And I do the nls:
> aux <- nls( ~ objFun(yEx,xEx,tEx,gamma,theta,kappa), 
>            data=data,
>    start=list(gamma=0.085, theta=8.62*10^(-5),
> kappa=4.45*10^(-3)), trace=TRUE)
> 
> I get the following error: 
> 179.8911 :  8.50e-02 8.62e-05 4.45e-03 
> Error in sum(rr[-(1:npar)]^2) : subscript out of
> bounds
> 
> One problem that I identified is that the algorithm
> does not change the values of the parameters at all. 
> It calculates the objective function OK, but the
> parameters are not updated.  
> 
> The function pdfDY is a Fourier integral with
> parameters xEx, tEx, gamma, theta, kappa, that I 
> calculate with a Filon algorithm.  The calculation of
> pdfDY is correct.  
> 
> Can anybody help me with this?  Thank you very much, 
Your objFun() is returning a scalar where nls is expecting a vector of
residuals. Try making it return just log(yEx/yTh), or rewrite as 

log(yEx) ~ log(pdfDY(xEx,tEx,gamma,theta,kappa))

or use optim() to minimize general functions.

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907

Adi Humbert <adrian_humbert at yahoo.com> writes:
> I'm migrating a project from Matlab to R, and I'm
> facing a relatively complicated problem for nls.  My
> objective function is below:
> 
> >> objFun <- function(yEx,xEx,tEx,gamma,theta,kappa){
>      yTh <- pdfDY(xEx,tEx,gamma,theta,kappa)
>      sum(log(yEx/yTh)^2)
>    }
> 
> The equation is yTh=P(xEx,tEx) + noise.
> I collect my data in:
> 
> >> data <- data.frame(xEx,tEx,yEx)
> 
> And I do the nls:
> aux <- nls( ~ objFun(yEx,xEx,tEx,gamma,theta,kappa), 
>            data=data,
>    start=list(gamma=0.085, theta=8.62*10^(-5),
> kappa=4.45*10^(-3)), trace=TRUE)
> 
> I get the following error: 
> 179.8911 :  8.50e-02 8.62e-05 4.45e-03 
> Error in sum(rr[-(1:npar)]^2) : subscript out of
> bounds
Could you run traceback() at this point so we can see where the error
message is originating?  I would hope that npar is 3 and my guess is
that the error message is originating in a call to

           conv = function()
           {
               rr <- qr.qty( QR, resid ) # rotated residual vector
               sqrt( sum( rr[1:npar]^2 ) / sum( rr[ -(1:npar) ]^2 ) )
           },

How many rows are there in your data frame?
> One problem that I identified is that the algorithm
> does not change the values of the parameters at all. 
> It calculates the objective function OK, but the
> parameters are not updated.  
> The function pdfDY is a Fourier integral with
> parameters xEx, tEx, gamma, theta, kappa, that I 
> calculate with a Filon algorithm.  The calculation of
> pdfDY is correct.

No need to respond to my suggestion about traceback().  Peter has
detected the problem.