Home »
Public Forums »
archive »
Re: constrained least-square fit
| Re: constrained least-square fit [message #5942] |
Fri, 08 March 1996 00:00 |
ROsborn
Messages: 2
Registered: March 1996
|
Junior Member |
|
|
In article <4hq1qb$1f2e@itssrv1.ucsf.edu>, aki@itsa.ucsf.edu wrote:
> I am to do some modeling of data. The modelfunction I am using is nonlinear.
> Therefore I am looking for a Levenberg-Marquard type of optimization routine.
> However I need to apply constraints to the fit (it becomes unstable otherwise).
> so curvefit.pro provided by IDL doesn't do it.
>
> Is anyone aware of a general least-square fit that allows constraints
written in
> IDL?
> Any hints are appreciated,
>
> Andreas
You could always transform the constrained parameters into circular
coordinates before calling the procedure. This has the advantage of
ensuring that chi^2 varies smoothly at the parameter limits. It doesn't
always work but is a better compromise than putting an infinite barrier in
the chi^2 v p space.
To transform the parameters, use
pt = arcsin((2*p-pmax-pmin) / (pmax-pmin))
Then you can restore them again using
p = 0.5*(pmax+pmin) + 0.5*(pmax-pmin) * sin(pt)
Hope this helps.
--
Ray Osborn Tel: +1 (708) 252-9011
Materials Science Division Fax: +1 (708) 252-7777
Argonne National Laboratory E-mail: ROsborn@anl.gov
Argonne, IL 60439-4845
|
|
|
|
Current Time: Wed Oct 08 15:56:13 PDT 2025
Total time taken to generate the page: 0.00572 seconds
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
