possible contrib: ROOT bridge

[lukasheinrich]

Description

it could be useful to provide a pyhf-like confiig interface to a roo-hf workspace/measuremet

cc @alexander-held @cranmer @kratsg @matthewfeickert

def process_measurement(measobj):
    spec = {'channels':[],'parameters':[]}
    for c in measobj.GetChannels():
        channel_spec = {'name': None, 'samples': []}
        channel_spec['name'] = c.GetName()
        nbins = c.GetData().GetHisto().GetNbinsX()
        for s in c.GetSamples():
            sample_spec = {'name': None, 'data': [0]*nbins, 'modifiers': []}
            sample_spec['name'] = s.GetName()
            for m in s.GetNormFactorList():
                modifier_spec = {'type': 'normfactor', 'name': m.GetName(),'data': 'wha'}
                sample_spec['modifiers'].append(modifier_spec)
            for m in s.GetOverallSysList():
                modifier_spec = {'type': 'normsys', 'name': m.GetName(),'data': 'wha'}
                sample_spec['modifiers'].append(modifier_spec)
            if s.GetNormalizeByTheory():
                modifier_spec = {'type': 'lumi', 'name': 'lumi', 'data': None}
                sample_spec['modifiers'].append(modifier_spec)
            channel_spec['samples'].append(sample_spec)
        spec['channels'].append(channel_spec)

    lumi_spec = {
        "auxdata": [1.0], "bounds": [
            [
                measobj.GetLumi() - 5.0 * measobj.GetLumiRelErr(),
                measobj.GetLumi() + 5.0 * measobj.GetLumiRelErr()

            ]
        ],
        "fixed": ('Lumi' in list(measobj.GetConstantParams())),
        "inits": [measobj.GetLumi()],
        "name": "lumi",
        "sigmas": [measobj.GetLumiRelErr()]
    }

    spec['parameters'].append(lumi_spec)
    return pyhf.pdf._ModelConfig(spec,poi_name = 'SigXsecOverSM')

class roohf_bridge:
    def __init__(self,fname,mname):
        import ROOT
        self.f = ROOT.TFile.Open(fname)
        self.config = process_measurement(self.f.Get(mname))

[matthewfeickert]

@lukasheinrich in the screenshot there is print(rm.config.par_order) twice, but I'll believe that it gives the same output as print(pm.config.par_order). :+1:

it could be useful to provide a pyhf-like config interface to a roo-hf workspace/measurement

I'm possibly missing the obvious, but can you elaborate more on the use case? This is more just adding a reader utility that then is still doing all the model conversion and building. Is the idea is to add syntactic sugar so that they don't have to write the lines themselves?

[lukasheinrich]

imagine you have a

• ROOT Workspace with a HistFactory Measurement
• a ROOT RooFitResult object

and you want to plot it via cabinetry .. ideally those two things presennt a similar API to what ccabinetry uses anyways (which is inspired by pyhf) so this just provides an adapter