Closed alexander-held closed 1 year ago
model_utils.prefit_uncertainties
can be fixed by using the suggested_fixed
API. model_utils.unconstrained_parameter_count
is not affected by constrained parameters, but needs in principle to be able to handle shapefactor
modifiers that are partially fixed. This is not currently supported by pyhf
on a workspace specification level (https://github.com/scikit-hep/pyhf/issues/2053), but it would be good to already implement this in cabinetry
such that it automatically works if support arrives in pyhf
.
The support for partially fixed parameters (such as
staterror
with zeros in some bins) inpyhf
0.7 can result in scenarios where the behavior ofsuggested_fixed_as_bool
is undefined, see https://github.com/scikit-hep/pyhf/issues/1944.This currently breaks
model_utils.prefit_uncertainties
incabinetry
(and also affectsmodel_utils.unconstrained_parameter_count
). Working around this should be possible by iterating bin by bin for affected parameters and using thesuggested_fixed
API instead.