Sébastien Ballesteros wrote:
hello, I have got some trouble with R functions nlm(),
nls() or optim() : I would like to fit 3 parameters
which must stay in a precise interval. For exemple
with nlm() :
fn-function(p) sum((dN-estdata(p[1],p[2],p[3]))^2)
out-nlm(fn, p=c(4, 17, 5),
I have had my best luck by re-parametrizing so that I no longer needed
restrictions. For example, if parameters must be positive, then I optimize
over parameters in log space, taking the exponential within my function.
This requires small changes to the function I'm optimizing (and the
gradient,
maybe you could re-parameterize the problem, e.g.,
p \in (a, b)
p^* = (p - a) / (b - a) \in (0, 1)
x = qlogis(p^*) \in \Re
I hope it helps.
Best,
Dimitris
Dimitris Rizopoulos
Ph.D. Student
Biostatistical Centre
School of Public Health
Catholic University of Leuven
Address: Kapucijnenvoer 35,
After some experimentation, I have had very good luck with the rgenoud
package. It handles box constraints well and for small problems its slow
speed has not been much of a problem. It has worked for me on Windows,
Macintosh, and Linux.
Chris
JPMorgan Asset Management
