[R] Error with nls
Douglas Bates
bates at stat.wisc.edu
Wed Mar 27 17:46:55 CET 2002
1-27206531-0-90000491 <domi at sun11.ukl.uni-freiburg.de> writes:
> I use:
>
> platform i386-pc-mingw32
> arch x86
> os Win32
> system x86, Win32
> status
> major 1
> minor 4.1
> year 2002
> month 01
> day 30
> language R
Thank you for providing that information.
> I try to fit a 2 compartment model. The compartments are open, connected
> to each other and are filled via constant input and a time depended
> function as well. Data describes increasing of Apo B after dialysis. Aim
> of the analysis is to test the hypothesis whether the data could described
> by two simple disconnected one compartment modes ore the "saturated
> model" holds? The first order differential equation for the saturated
> model:
>
> db5 = - (k50+k56)*b5 + k56*b6 + c*g(t) + h
> db6 = + k65*b5 - (k60+k65)*b6 + d
>
> db5, db6 are the first derivatives, b5, b6 are the functions to be
> fitted. The remaining parameters are unknown and should follow from the
> fit.
>
> assuming that g(t) has the functional form: b4i + (b40-b4i)*exp(-k4*t)
>
> (after calculations of 2 papers of A4) follows the solution:
>
> L5L6 <- function(b40, b4i, k4, t, p50, p56, p60, p65, pc, ph, pd, pb50,
> pb60) {
>
> k50 <- exp(p50)
> k56 <- exp(p56)
> k60 <- exp(p60)
> k65 <- exp(p65)
> c <- exp(pc)
> h <- exp(ph)
> d <- exp(pd)
> b50 <- exp(pb50)
> b60 <- exp(pb60)
> a <- (k50+k56)
> b <- k65
> e <- k56
> f <- (k60+k65)
> z1 <- (-(a+f)/2 - sqrt((a+f)^2/4 - a*f + b*e))
> z2 <- (-(a+f)/2 + sqrt((a+f)^2/4 - a*f + b*e))
> K <- ((z1+a)/(z2-z1))
> B1 <- (b/(z2-z1)*b60 - K*b50)
> A1 <- (b50-B1)
> X1 <- (b*d/(z2-z1)-K*(c*b4i+h))
> X2 <- (K*c*(b4i-b40))
> X3 <- (c*b4i + h - X1)
> X4 <- (c*(b40-b4i)- X2)
> C1E <- (X3/(-z1)*(1-exp(z1*t)) +
> X4/(-(k4+z1))*(exp(-k4*t)-exp(z1*t)))
> C2E <- (X1/(-z2)*(1-exp(z2*t)) +
> X2/(-(k4+z2))*(exp(-k4*t)-exp(z2*t)))
> b5 <- (A1*exp(z1*t) + B1*exp(z2*t) + C1E + C2E)
> b6 <- ((z1+a)/b * A1*exp(z1*t) + (z2+a)/b * B1*exp(z2*t) +
> (z1+a)/b * C1E + (z2+a)/b * C2E)
> y <- f5*b5 + f6*b6
> return(y)
> }
>
> I am in the lucky circumstances having starting values, because a nlr-fit
> succeeds, the graphical presentation of the fits looks quite nice. The nlr
> function is part of Lindsey's library(gnlm), but now I would like to apply
> Pinheiro and Bates library(nlme) and I have got an error:
>
> > m2 <- nls(y ~ L5L6(b40, b4i, k4, t, p50, p56, p60, p65, pc, ph, pd,
> > pb50, pb60),
> > + data=help, start=c(p50=0.008678954, p56=-0.595153967,
> > + p60=-4.602990518, p65=-0.625732096,
> > + pc=-0.128657978, ph=0.708033556, pd=1.140357461, pb50=1.311141424,
> > + pb60=1.270852258))
> > Error in numericDeriv(form[[3]], names(ind), env) :
> > Missing value or an Infinity produced when evaluating the model
> >
> If somebody feel that he can help me, I could send him my R- code and
> data file as well.
It is likely that the iterative algorithm is progressing to values of
the parameters that don't make sense physically. I suggest that you
add trace = TRUE to your call to nls. This will provide a record of the
parameter values, the residual sum of squares, and the convergence
criterion throughout the iterations.
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
More information about the R-help
mailing list