Template histogram visualization from workspace input

[alexander-held]

cabinetry.visualize.templates relies on the cabinetry configuration file and input histograms to visualize pre-/post-modification histograms and include histogram axis information. It would be useful to have the same visualization also accessible from a pure workspace input.

[alexander-held]

Small example with basic functionality:

import pathlib

import cabinetry
import numpy as np


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


for channel in ws["channels"]:
    for sample in channel["samples"]:
        sample_yield = np.asarray(sample["data"])
        nominal_histo = {"yields": sample_yield, "stdev": np.zeros_like(sample_yield)}  # ignore uncertainty

        for modifier in sample["modifiers"]:
            if modifier["type"] == "normsys":
                up_histo_orig = {"yields": sample_yield*modifier["data"]["hi"], "stdev": np.zeros_like(sample_yield)}
                down_histo_orig = {"yields": sample_yield*modifier["data"]["lo"], "stdev": np.zeros_like(sample_yield)}

            elif modifier["type"] == "histosys":
                up_histo_orig = {"yields": modifier["data"]["hi_data"], "stdev": np.zeros_like(sample_yield)}
                down_histo_orig = {"yields": modifier["data"]["lo_data"], "stdev": np.zeros_like(sample_yield)}

            else:
                continue

            figure_name = (
                    f"{channel['name']}_{sample['name']}_{modifier['name']}.pdf"
                ).replace(" ", "-")
            figure_path = pathlib.Path("figures") / figure_name

            figure_label = (
                f"region: {channel['name']}\nsample: {sample['name']}"
                f"\nsystematic: {modifier['name']}"
            )

            cabinetry.visualize.plot_model.templates(
                nominal_histo,
                {},
                {},
                up_histo_orig,
                down_histo_orig,
                np.linspace(0, len(sample_yield), len(sample_yield)+1),  # workspace does not contain binning information
                "observable",  # workspace does not contain observable name
                figure_path=figure_path,
                label=figure_label
            )

For the implementation we may want to consider using pyhf API functionality of the model to read all this data instead in a more generic way.