Support for multiple masses in datacards

[IzaakWN]

I am trying to write a single datacard for multiple mass points:

cb.WriteDatacard(f"output/datacard.txt",f"output/datacard.inputs.root")

instead of

for mass in masses:
  cb.cp().mass([mass,'*']).WriteDatacard(f"output/datacard_M{mass}.txt",f"output/datacard_M{mass}.inputs.txt")

However, currently the writing routine throws an error:

CombineHarvester/CombineTools/src/CombineHarvester_Datacards.cc

Lines 669 to 675 in ba8d9b8

	if (masses.size() != 1) {
	throw std::runtime_error(FNERROR("Process " + sig_proc + " in bin " +
	bin +
	" has multiple entries with multiple "
	"mass values, this is not supported"));
	}

I am wondering if there is a canonical way of doing this, or otherwise, we can add support for this? The inputs ROOT file would contain the histograms for all signal mass points, while writing the $MASS keyword in the datacard (which it already does, actually), and setting the signal rate to -1 to make Combine take the rate from the histogram's integral, see https://cms-analysis.github.io/HiggsAnalysis-CombinedLimit/latest/part2/settinguptheanalysis/?h=rate#rates-for-shape-analyses

Currently, an ugly workaround to get the histograms of all signal mass points written to the same ROOT file is the following:

outfname_txt  = "output/datacard.txt"
outfname_root = "output/datacard.inputs.root"
outfile_root  = TFile(outfname_root,'RECREATE')
for mass in masses: # write each signal mass point to ROOT file
  cb.cp().signals().mass([mass]).WriteDatacard(outfname_txt,outfile_root) # cannot handle multiple masses $MASS
cb.cp().signals().ForEachProc( lambda x: x.set_rate(-1) ) # tell Combine to take the rate from the histogram
cb.cp().mass([masses[0],'*']).WriteDatacard(outfname_txt,outfile_root)
outfile_root.Close()

But it's a bit ugly, and it overwrites the text file for each signal point, but all signal histograms in the ROOT file seem to retain the integral of their initial value.

Tagging @anigamova

[ajgilbert]

Hi @IzaakWN - this code was written quite a while ago now, but I have the feeling this use case wasn’t envisaged at the time. In principle I think it could be supported - the main thing is that CH would need to verify that in the presence of multiple signal masses each one has the same set of systematics associated (and exact same values for lnNs). So I would say possible, but not a trivial amount of work

[IzaakWN]

Hi @ajgilbert, ah yes. I guess the use case might be a bit more narrow.

(This question was triggered in context of datacard validation & preservation, but it is not the end of the world to have one datacard per mass point now that we can configure which datacard the validation pipeline runs on.)