This repository gathers a bunch of scripts and tools to convert the HF output from an analysis into pyhf workspaces (BkgOnly.json and patchset.json), perform statistical inference and plot the results. It also allows to create simplified versions of the likelihood using simplify. It further allows to cross-check the results of the simplified likelihood by comparing them with the full results from the analysis. Needs both python 2.7 and python 3 because of ROOT, HistFitter and pyhf.
Creating a signal patch is as easy as running
python3 make_signalpatch.py --bkg-only <BkgOnly.json> --output-file <output-file>
A simplified signal patch only contains nominal signal yields and no uncertainties. You can construct such a signal patch file from any full signal patch file using
python3 make_simplified_signalpatch.py --group <group> --output-file <output-file>
Fitting all workspaces can be done using
python3 run_cls.py --group <group>
Alternatively, once can also use run_patchset.py to run over an existing set of BkgOnly.json and patchset.json files built e.g. using make_signalpatch.py from above.
Running a fit over the entire patchset using a provided likelihood can be done using:
python3 run_patchset.py --group <group>
This defaults to using the full likelihood. You can also specify --simplified in order to run using the simplified likelihood.
The results of the fits need to be harvested and converted into the right format for the HistFitter plotting tools to be able to use them. This can be done by running
python3 harvest.py --group <group>
For this step, you'll need a more or less recent ROOT version as well as python2.7 (sorry). No need to setup Histfitter, as all necessary classes are included in this repository. Create the usual HF-style TGraph using
python harvestToContours.py -i <input_harvest> -o <output_tgraph> -x <x> -y <y> -l <forbidden_function>
Plotting the resulting graphs can be done with
python plot_contours.py --group <group> --xmin <xmin> --xmax <xmax> --ymin <ymin> --ymax <ymax> [--draw-CLs]