I've included an interface to Minuit as a new module PDL::Minuit in the
PDL distribution, so if you get fresh source from the cvs you can get it.
Alternatively for those who would like to try this without recompiling
PDL at this point, I have put a standalone tarball at:
http://www.eso.org/~ajordan/PDL-Minuit-1.0.tgz
Enjoy (assuming the compilation goes smoothly....).
-Andres
Rahman Amanullah wrote:
Hi Andres,
I want to do exactly what Karl explicitly wrote in his email, solve a
non-linear Chi2 problem with an arbitrary weight matrix. Minuit is
perfect for this, and if you think you will have a PDL interface ready
within a few days, that would just be fabulous!
Cheers,
Rahman
On Thu, 2007-06-14 at 22:33 +0200, Andres Jordan wrote:
I have indeed written a PDL interface to Minuit which I've been
intending to post here for a while (for a long time I was chasing a
small mem leak in the interface but finally solved that some months
ago). Minuit is very flexible and robust minimization package. I will
clean up the code a bit, finish the documentation and finally post this
in the next few days.
Rahman: I am bit confused as to what you mean by inputting the
covariance matrix. There is a Minuit manual at:
http://wwwasdoc.web.cern.ch/wwwasdoc/minuit/minmain.html which might
help you in determining if Minuit will help you with what you want.
-Andres
Karl Glazebrook wrote:
I would have thought that one could easily modify any of these fairly
easier by just changing
the way it computes the goodness of fit metric.
i.e. something like chi^2 = D^T I D where I is the inverse of the
covariance matrix and D = Data - Model matrix
Andreas Jordan has written a PDL interface to MINUIT. Andreas please
speak up!
Karl
On 14/06/2007, at 2:14 PM, Rahman Amanullah wrote:
Hi,
I'm looking for a fitter in PDL that takes an *input* covariance matrix.
My data is heavily correlated and I need to take this into account in my
fit.
So far I've been looking at PDL::Fit::LM and PDL::Fit::Levmar but none
of these appear to have this feature. It's possible to pass weights to
PDL::Fit::LM, but not a full covariance matrix so the input data is
still treated as uncorrelated in the fit.
If there is no existing fitter that supports this, I would very much
appreciate any thoughts on how I could either modify any of the existing
packages to support this, or any external library for which it would be
fairly straight forward to write a PDL:PP interface. Perhaps CERNLIB
MINUIT is a good option? Has anyone ever written a PDL interface to
this?
Cheers,
Rahman
--__________________________________________________________
Rahman Amanullah, +46 8 553 786 61
Stockholm University ICQ: 62860758
Stockholm Centre for Physics, 17d 50m E
Astronomy & Biotechnology 59d 26m N
PGP/GPG key ID: 0x3C229C80
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
