Revert slight change in return types of `suggested_fixed`

[alexander-held]

Summary

#1639 very slightly changed the return type of model.config.suggested_fixed(). It used to be List[bool], but now some elements can be numpy.bool_, which behaves differently (see numpy/numpy#9646):

>>> import numpy as np
>>> isinstance(np.bool_(True), bool)
False

This makes typing / type checking in libraries using pyhf more complicated (spotted via scikit-hep/cabinetry#301). Would it be possible to consistently return List[bool] again instead?

Additional Information

The example below shows how numpy.bool_ entries appear for some, but not all models.

import pyhf

spec = {
    "channels": [
        {
            "name": "SR",
            "samples": [
                {
                    "data": [25, 5],
                    "modifiers": [
                        {
                            "data": [5, 2],
                            "name": "staterror_SR",
                            "type": "staterror",
                        },
                    ],
                    "name": "Signal",
                }
            ],
        },
    ],
    "measurements": [{"config": {"parameters": [], "poi": ""}, "name": "fit"}],
    "observations": [
        {"data": [35, 8], "name": "SR"},
    ],
    "version": "1.0.0",
}


def check_types(model):
    for i, par in enumerate(model.config.suggested_fixed()):
        if not isinstance(par, bool):
            print(f"{model.config.par_names()[i]} has type {type(par)}")


model = pyhf.simplemodels.correlated_background([5], [10], [11], [9])
print("checking correlated_background model")
check_types(model)

model = pyhf.Workspace(spec).model()
print("checking custom model")
check_types(model)

Output for 9fbbbf9 and later (including current master):

checking correlated_background model
checking custom model
staterror_SR[0] has type <class 'numpy.bool_'>
staterror_SR[1] has type <class 'numpy.bool_'>

Output for 5ea4e0a:

checking correlated_background model
checking custom model

Code of Conduct

• I agree to follow the Code of Conduct

[alexander-held]

This only seems to affect staterror modifiers. shapesys.required_parset uses a slightly different way to construct the information and returns a list of bools. shapefactor.required_parset does not seem to support per-bin settings at the moment and is also not affected, as are the remaining modifier types that don't have per-bin settings.

The staterror behavior can be changed to return List[bool] again at a few different levels, I will push a PR with a suggestion and some comments in case you'd like to have the old behavior restored.