Load model, without having "mu" parameter

[ssaha1234]

Dear pyhf community,
I am performing a Control Region only fit for my analysis. I have the likelihood with all the regions, but I do not have mu parameter defined under channels, as I am assuming Signal = 0 in the CRs. I have mu defined under measurements as the poi.

I am processing the following lines for the mle fit -

wspace = pyhf.Workspace(json.load(open(sys.argv[1])))
model = wspace.model()
data = wspace.data(model)
init_pars = model.config.suggested_init()
fixed_params = model.config.suggested_fixed()

result, result_obj = pyhf.infer.mle.fit( data, model, maxiter=1000000, init_pars=init_pars, fixed_params=fixed_params, return_uncertainties=True, return_result_obj=True )

At the moment, the code crashes when I try to load the model in this line - model = wspace.model() as mu is not defined. I am thinking if I can forcefully set mu to 0, to do the fit, without the crash ? The fit works fine when include all the regions (SR+CR).

Thanks !
Best,
Shreya

[alexander-held]

Hi @ssaha1234, setting mu to a constant value of 0 will not be enough to avoid that crash I think. You would additionally need to keep some sample in the model on which mu acts. Since mu will be constant 0, that sample effectively does nothing and this should result in the model you are looking for I believe.

A more convenient method has been implemented after pyhf version 0.6.3 was released, see #1602 / #1636 / #1638. It allows you to set "poi": "" in the config, or use wspace.model(poi_name=None) for fits without POI. You can already use poi_name=None now (thanks to #950), however you will not get through model validation without the workaround shown at the bottom of #1602 (comment). That workaround may be the most convenient thing to use with pyhf version 0.6.3, and 0.7.0 will bring the new options.

[ssaha1234]

Hi @alexander-held,
Thank you very much for the detailed workarounds, I am looking through the tickets and comments, and I will try it out after !
Best,
Shreya