[Rivet] ATLAS analysis to be included into Rivet

[James Monk]

Hi Hendrik,

I tend to agree that including histograms that only give correct results when un-weighted events are used is inviting a bit of a cock-up at some later time, albeit this is a soft QCD analysis and *usually* wouldn't have weighted events.

At the very least the analysis should throw an exception and a clear error message if it ever receives an event weight different from unity.  That ought to protect against wrong results, even if it is not the most ideal solution.  I'd ask the question first about how long it would take to make it work properly though.

cheers,

James

On 15 Apr 2013, at 16:58, Hendrik Hoeth wrote:

> Any opinions? This affects 14 out of 27 histograms. I really don't feel
> comfortable taking the analysis as it is, but I also don't feel like
> taking the heat on this list after rejecting the analysis (hey, this one
> doesn't even crash! I can't even imaging how strong a fire protection
> I'd need!) -- been there, done that way too often. So what do you guys
> think?
> 
> Thus spake Roman Lysak (lysak at fzu.cz):
> 
>> On 04/12/2013 11:41 AM, Hendrik Hoeth wrote:
>>> Hi Roman, hi Kiran,
>>> 
>>> I'm having a brief look at the code right now. Have you tested it with
>>> weighted events? You do fill a weight vector _vecWeight, but never use
>>> it. The regression seems to assume equal weights for all events. Is that
>>> correct?
>> That is correct.
>> 
>> The azimuthal part of the analysis includes the weights.
>> The fwd-bkwd part of the analysis does not include the weights.
>> 
>> We tested this only with unweighted events (Herwig++ & Pythia6) and
>> compared to the results in the paper (there are no plots with MC
>> having weighted events, so we would not be able to validate the code
>> on weighted events anyway).
>> 
>> I'm not sure what is the preference here, but it seems to me that
>> including the weights would require quite some changes to the code
>> (I'm not the author of the analysis who wrote the analyzing part of
>> the code and so it may take some time to do it).
> 
> _______________________________________________
> Rivet mailing list
> Rivet at projects.hepforge.org
> http://www.hepforge.org/lists/listinfo/rivet