This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: Problem with least-squares fitting.
On Tue, Nov 05, 2002 at 05:39:18PM -0200, Tiago de Paula Peixoto wrote:
> On Tue, 5 Nov 2002 10:16:20 +0100 (CET)
> Kenneth Geisshirt <kenneth@geisshirt.dk> wrote:
>
> > My hypothesis: fitting 1 parameter using multifit is "not nice to do".
>
> Well, it shouldn't be a problem. Suppose you want to fit something like "y
> = a * x^8" or "y = a * sin(x)". There's no other way, you must use
> multifit.
>
> I really think it may be a bug, since the fitted parameter is actually
> correct despite the error message (at least in the cases that I tested).
>
> Please someone help ! ;^)
>
himm, i agree that n=1 should be solveable, but for a single parameter, the
method might not be the most efficient. eg.
y = a*x^8 -> x=(y/a)^(1/8), y = a * sin(x) -> x = arcsin(y/a) ie. analytic
approach. or if it is not possible: y = f(x) is equivalent to y-f(x)=0 which
is solveable by root finding.
ato