Re: [ROOT] Bug in likelihood fitting?

From: Victor Perevoztchikov (perev@bnl.gov)
Date: Fri Apr 28 2000 - 23:47:03 MEST


> > It seems that log-likelihood fitting is integerizing bin
> > contents before fitting.  Is this intended behavior?

Hi Burt,

likelyhood fitting is based on known function of distribution. In current case
it is binomila or poisson. It supposed that bins contain number of entries.
So they should be integers. If you use weighted histogramm, then distribution
of sum of weights in bin is not poisson, and anyway result of fit is wrong.

Victor


Dave Morrison wrote:
> 
> Hi Burt,
> 
> Yep.  This has been a topic of discussion in PHENIX computing recently.
> Here are the lines that do the deed (H_H1.cxx:1546-1550):
> 
>          if (fu < 1.e-9) fu = 1.e-9;
>          icu  = Int_t(cu);
>          fsub = -fu + cu * TMath::Log(fu);
>          fobs = hFitter->GetSumLog(icu);
> 
> The bin contents are floor()'ed.  I suppose one could argue that since
> the log likelihood procedure in TH1::Fit happens to assume a poisson
> probability for sampling the observed bin contents from the model you're
> on somewhat sketchy ground if the bins don't contain integer values, but
> I think the last line above should be changed to
> 
>          fobs = hFitter->GetSumLog(cu);
> 
> if only to avoid needless surprises.
> 
> Cheers,
> Dave
> 
> Burt Holzman wrote:
> 
> > It seems that log-likelihood fitting is integerizing bin
> > contents before fitting.  Is this intended behavior?
> 
> --
> David Morrison  Brookhaven National Laboratory  phone: 631-344-5840
>                 Physics Department, Bldg 510 C    fax: 631-344-3253
>                           Upton, NY 11973-5000  email: dave@bnl.gov

-- 
Victor M. Perevoztchikov   perev@bnl.gov  perev@vxcern.cern.ch       
Brookhaven National Laboratory MS 510A PO Box 5000 Upton NY 11973-5000
tel office : 631-344-7894; fax 631-344-4206; home 631-345-2690



This archive was generated by hypermail 2b29 : Tue Jan 02 2001 - 11:50:24 MET