Shape only systematics

[Tomoya-Iizawa]

Hi,

I would like to implement systematics only with shape, since normalization uncertainty is included as overall uncertainty.
Is there any option for this?
In the document, I can find only Normalization and NormPlusShape.
Is it necessary to normalize variation distribution to nominal distribution by hand?

Best,
Tomoya

[alexander-held]

Hi Tomoya, there is currently no option for this. I think this should be in scope for cabinetry though and not too difficult to implement. I added #447 to track this.

The easiest way at the moment would probably be to have a custom post-processing script that removes normsys modifiers from the workspace for all the cases where you would like to use such systematics. Below is an example which uses the example workspace created by example.py:

import cabinetry

ws = cabinetry.workspace.load("workspaces/example_workspace.json")

MOD_NAMES_TO_PRUNE = ["Modeling"]  # modifiers name to prune
MOD_TYPE_TO_PRUNE = ["normsys"]  # modifier types to prune

# the ws object is just a dictionary, iterate through that and drop pieces as needed
# only prune modifiers matching all requirements (name and type)
for channel in ws["channels"]:
    for sample in channel["samples"]:
        sample["modifiers"] = [
            mod
            for mod in sample["modifiers"]
            if not (
                mod["name"] in MOD_NAMES_TO_PRUNE and mod["type"] in MOD_TYPE_TO_PRUNE
            )
        ]

model, data = cabinetry.model_utils.model_and_data(ws)

You can check e.g. with cabinetry.visualize.modifier_grid(model) before and after this pruning that the edits work as intended.

In pyhf there is a function to perform pruning, however I think it does not work too well for this specific case: you want to prune only if multiple requirements are fulfilled (modifier type is normsys and name matches that of the systematic you want to prune). At the moment, pyhf prunes with a logical or (see scikit-hep/pyhf#2374).

[Tomoya-Iizawa]

Hi,
Thank you for your answer and sorry for the late reply.
I will refer to that, and for the time being, I made a custom script to remove the effect of normalization from variation distribution.