Is there any option in "infer.mle.fit" that can remove a certain constraint term?

[zwu0922]

In Roofit "RooAbsPdf::fitTo()", one would be able to apply constraints only to a given subset of parameters by using the "Constrain(const RooArgSet&pars)" option. I would like to know if there is a counterpart in pyhf.infer.mle.fit?

[alexander-held]

This is currently not possible. Here is a related issue: #820.

[zwu0922]

Thanks a lot, Alexander. Seems that I cannot continue my work with pyhf.

[lukasheinrich]

do you want to remove the constraint but keep the NP?

[zwu0922]

Yes, I would like to keep the NP for a "histosys" but remove the corresponding constraint term from the likelihood.

[kratsg]

Just to briefly chime in, as mentioned in the issue, it's not a technically difficult thing to implement. However, the problem is that you then enter a regime where your model is not necessarily HiFa, but is "HiFa-like". HistFactory is specific about the constraints being incorporated as part of the p.d.f.

I suspect you'll be able to progress more in the corresponding python-RooStats (if/whenever that ends up being fleshed out/developed) rather than particularly in pyhf alone. We probably could add support for this, or at least provide an easier way for the user to control or modify the underlying generated HiFa p.d.f. to do similar manipulations.

[alexander-held]

It depends a bit on how strictly you define HistFactory, but it is (partially, I think only for OverallSys and not for HistoSys) supported in the xmls (see example in #820).