[Rivet] histograms scaling

Andy Buckley andy.buckley at ed.ac.uk
Wed Apr 13 00:53:34 BST 2011


On 12/04/11 23:55, Anton Karneyeu wrote:
> Hi Andy,
> 
> I am looking for a way to preserve statistical information associated
> with histogram (like sum of weights, weights^2, x*weight, etc. per bin)
> after normalization/scaling step (after
> Rivet::AnalysisHandler::finalize() call). As I see, currently in Rivet
> all this information discarded - basically histogram (IHistogram1D)
> converted to an array of points (IDataPointSet).
> 
> Just to give more details - this is necessary to sum up histograms from
> different runs properly (I use Rivet as library).
> 
> Do you have an idea how to achieve this need?

Hi Anton,

There's currently no way to keep this information, and it's a hard
problem as general finalize operations can do all sorts of uncombinable
things to make the resulting histograms: it's not just Histogram1Ds, but
also Profile1Ds, and scaled/normalised/divided combinations of
histograms, which makes it *much* more complicated.

We've known that this is a problem for a long time, but the demands of
our users meant that the development effort had to go into the physics
content rather than the histogramming. We're working on a histogramming
upgrade at the moment, which means that specifically writing interface
code to use the AIDA objects would be guaranteed to break with release
2.0.0, so I wouldn't recommend that.

If you've got any suggestions on how you think we can solve the general
problem, please let us know! The histogramming upgrade will preserve
data type and fill information for standard histograms, but histo
divisions, asymmetries, etc. are probably still going to have to be
rendered into DataPointSets... maybe that will be good enough for your
purposes, though.

Andy

-- 
Dr Andy Buckley
SUPA Advanced Research Fellow
Particle Physics Experiment Group, University of Edinburgh

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.



More information about the Rivet mailing list