Hi Antonio,
On Wed, 26 Apr 2006, Antonio Petrella wrote:
> Hi Heiko,
>
> >
> > 1) What is the difference between Thorsten's signal function you have
> > used in your fits documented at April 4th compared with what is
> > used now?
>
> The difference is that in the latest fit on data (the one posted on 19th
> of April) I have fixed the Crystal ball parameters with ones found
> fitting combinatorial background, and signal pdf's parameters with ones
> found fitting signal MC (except the parameters related to the angular
> point, which you suggested to float), while the fit on data posted on
> 4th of April was an attempt to use the new signal pdf with all
> parameters free to float.
I think we need to fix the combinatorial BG parameters on MC.
However, it would be interesting to see if the fit works out
when letting all signal parameters to float.
BTW: It is maybe not a real surprise that, when fixing all signal
parameters on MC, that thefit does not work so well on data.
For lower purity it would be a surprise if the different purity
classes are populated in the same way in data and MC.
One could reweight the MC on the tag side with data such that
one gets the same population in the different purity classes.
Given the tight time schedule this is not possible and we need
to find a reasonably working fit configuration.
Cheers,
Heiko
> > In that configuration the fit worked reasonably well also on data
> > and since we are behind the time schedule I would vote to use this
> > configuration.
> >
> > 2) Concerning the chi^2 scaling:
> > a) I'm not sure if I really understood how the errors on the fit
> > parameters have been calculated. Did you do it by yourself?
> > Could you please be more specific?
>
> The fitted parameters and their error are calculated by RooFit
> automatically. I have used the procedure described on
>
> http://roofit.sourceforge.net/docs/classref/examples/fitgen3.cc.html
> to perform the nll and chi^2 fits. Maybe this is not the appropriate
> procedure for extended fits.
>
> > b) A chi^2 fit can be screwed up if there are bins with very small
> > statistics or even zero entries.
> > Is the range in the histogram constrained?
>
> The datasets used to make the various fits have all the same
> charateristics (range, binning). The only thing I did was to rescale the
> error for each bin.
>
> ciao,
> Antonio
> > Cheers,
> > Heiko
> >
> >
> > On Fri, 21 Apr 2006, Antonio Petrella wrote:
> >
> >>
> >> Hi Kerstin,
> >>
> >>> since numbers are so different I am still wondering whether I compare
> >>> the correct tables... Is the table from today and the table from April
> >>> 12 (not using any links but the tables directly on the main page) the
> >>> correct ones to compare?
> >> Yes that's correct. I would have expected that the yields for the All MC
> >> fit from April 12 (NLL fit) and the yields for the fit with no error
> >> scaling from April 20 (Chi^2 fit) were the same.
> >>
> >>> If so, the yields are so different that it should be enough to just
> >>> compare them to the number of events in the histograms to tell us
> >>> which fit obviously gives us weird numbers... have you looked at that?
> >>>
> >> Yes, since the dataset is the same for both fits, and has 1561024
> >> entries, the wrong one is certainly the minimum chi^2 fit.
> >>
> >>
> >> Antonio
> >>> Cheers,
> >>> Kerstin
> >>>
> >>>
> >>> On Thu, 20 Apr 2006, Antonio Petrella wrote:
> >>>
> >>>> Hi Kerstin,
> >>>> that is exactly what I meant. The sample is right the same but yields
> >>>> are different. This happened just by switching to chi^2 fit.
> >>>>
> >>>> Probably I'm doing something wrong... but it's not clear to me where: I
> >>>> have followed the example on roofit web site
> >>>> (http://roofit.sourceforge.net/docs/classref/examples/fitgen3.cc.html)
> >>>> to make the chi^2 fit.
> >>>>
> >>>> Do ou have any suggestion?
> >>>>
> >>>> Bye...
> >>>> Antonio
> >>>>
> >>>> Kerstin Tackmann ha scritto:
> >>>>> Hi Antonio,
> >>>>>
> >>>>> can you be a little more clear what you mean by "numbers look quite
> >>>>> strange"? Do you fit the same samples as on April 12th? The yields
> >>>>> seem to be very different. But maybe I am just not comparing the
> >>>>> correct numbers...
> >>>>>
> >>>>> Thanks,
> >>>>> Kerstin
> >>>>>
> >>>>> On Thu, 20 Apr 2006, Antonio Petrella wrote:
> >>>>>
> >>>>>> Hi all,
> >>>>>>
> >>>>>> on the web page
> >>>>>> http://www.slac.stanford.edu/~petrella/mesfits/mesfits.html
> >>>>>>
> >>>>>> I posted the result of the chi^2 scaling using a minimum chi^2 fit.
> >>>>>> This time we can see a variation on the yields and errors, but numbers
> >>>>>> look quite strange...
> >>>>>>
> >>>>>> Antonio
> >>>>>>
> >>>>>>
>
