[Rivet] Merging of nparton samples

[Andy Buckley]

Hi Alexander,

No, ratios can't be merged correctly this way --that's fundamental, not
a YODA thing, since the required information about statistical moments
in the numerator and denominato is lost by taking the ratio.

For now if you want ratios that can be reliably statistically merged,
you need to write out and merge the numerator and denominator histos
separately, and use a post-processing YODA Python script to do the
division of the merged histos.

yodamerge will attempt to guess merging of scatters like ratios, but as
you say it has to assume that there were the same number of events in
each run, and the stats are asymptotic rather than exact.

A future version of Rivet will solve this problem in general -- we know
how to do it, but it requires very significant development work to do it
in a way that is invisible to the user / analysis author.

Andy


On 13/11/14 14:43, Alexander Grohsjean wrote:
> Hi Andy,
> 
> will this properly merge histograms with fractions (like the ttbar jet
> veto analysis ) taking the number of events in
> each histogram into account? Otherwise it would just work for the
> special case where you fill each histogram
> with the same number of events, no?
> 
> Cheers, Alexander.
> 
> 
> 
> Am 13.11.2014 um 14:03 schrieb Dominic Hirschbuehl:
>> Hi Andy,
>>
>> great, this seems to be what I was looking for. I'll try it out and let
>> you know.
>>
>> Thanks
>>    Dominic
>>
>> On Thu, Nov 13, 2014 at 01:00:10PM +0000, Andy Buckley wrote:
>>> Quick comment, since I'm teaching: you can give weights to each file for
>>> merging on the yodamerge command line. Does that solve your problem, if
>>> you use the xsecs, or something proportional to them, as the merging
>>> weights?
>>>
>>> Andy
>>>
>>>
>>> On 13/11/14 12:56, Chris Pollard wrote:
>>>> Hi Dominic,
>>>>
>>>>      Yes and no. I need to weight the different histograms during the
>>>>      merging.
>>>>      What I have are for example 5 samples:
>>>>      tt+0p   9.97 pb
>>>>      tt+1p  10.83 pb
>>>>      tt+2p   7.72 pb
>>>>      tt+3p   3.95 pb
>>>>      tt+4+p  2.58 pb
>>>>      Now these 5 subsampels need to be merged to one common sample,
>>>> with
>>>>      their given xs.
>>>>
>>>>
>>>> Right, but if you normalize each histogram to the appropriate xs in
>>>> finalize(), this shouldn't be a problem. Is this not what you're doing?
>>>>
>>>> Chris
>>>>
>>>>
>>>> _______________________________________________
>>>> Rivet mailing list
>>>> Rivet at projects.hepforge.org
>>>> https://www.hepforge.org/lists/listinfo/rivet
>>>>
>>>
>>> -- 
>>> Dr Andy Buckley, Royal Society University Research Fellow
>>> Particle Physics Expt Group, University of Glasgow / PH Dept, CERN
> 


-- 
Dr Andy Buckley, Royal Society University Research Fellow
Particle Physics Expt Group, University of Glasgow / PH Dept, CERN