Re: [ROOT] minimization class within ROOT

From: Edmond Offermann (edmondoffermann@yahoo.com)
Date: Wed Dec 01 2004 - 17:18:11 MET


Hi Sebastien,

Try to formulate your fitting function so that you minimize the
correlation between the parameters . After the fit you can always
transform into the shape you desire . Strong correlated parameters
will just lead to near-singular matrices in the solution engine.

I only know of the usage of Lagrangian Multipliers to enforce
parameter equality constaints . If you know how to use them
in correlated parameter searches , please enlighten us :)

Eddy

--- Sébastien Gadrat <gadrat@clermont.in2p3.fr> wrote:

> Hi Rene, hi rooters,
> 
> I am trying to fit several spectra obtained with different set of
> cuts. 
> Most of them can be fit easily with one or both minimizers (I usually
> 
> use both as cross check), i.e. TMinuit and TFumili. But sometimes,
> the 
> fit failed for both. This seem to be due to some little bit
> correleted 
> parameters of my user defined functionnal. I could try to rewrite it
> but 
> I put it in an easy way for later use. So I would like to keep it as
> it 
> is and I was thinking about new minimizers... Especially, I heard
> some 
> minimizers can overcome correlation within parameters (Lagrange 
> multiplicators). So I wondered whether we have such minimizers in
> ROOT.
> Thanks in advance,
> Best regards,
> 
> Sebastien
> 
> Rene Brun wrote:
> > Hi Sebastien,
> > 
> > Currently, we have only two classes of minimizers (The Minuit
> family
> > and TFumili).
> > We are currently working on a linear fitter that could be used
> > to fit distributions with a combination of linear functions.
> > Instead of a conventional minimisation, this implies only a matrix
> > inversion.
> > A RobustFitter algorithm is also in the pipeline.
> > 
> > Could you indicate the rationale behind your question?
> > 
> > Rene Brun
> 
> 



This archive was generated by hypermail 2b29 : Sun Jan 02 2005 - 05:50:10 MET