spey-pyhf

About

spey-pyhf is a pyhf plug-in for the spey statistical inference interface. It exposes pyhf's full likelihood-building machinery — including HistFactory-style JSON workspaces and signal patches — through Spey's backend-agnostic API, so any analysis that ships a pyhf likelihood can be consumed by Spey alongside other backends. Once installed, Spey automatically detects the plug-in and registers three backends: pyhf (full statistical model from a background-only workspace plus a signal patch), pyhf.uncorrelated_background (a lightweight uncorrelated-background helper) and pyhf.simplify (workspace simplification).

Installation

The plug-in is published on PyPI and can be installed with:

python -m pip install spey-pyhf

To install directly from the main branch:

python -m pip install --upgrade "git+https://github.com/SpeysideHEP/spey-pyhf"

Note that the main branch may not be the stable version.

Quick example

Given a pyhf background-only workspace and a signal patch, you can build a Spey statistical model and compute the exclusion confidence level in a few lines:

import spey

background_only = {
    "channels": [
        {
            "name": "singlechannel",
            "samples": [
                {
                    "name": "background",
                    "data": [50.0, 52.0],
                    "modifiers": [
                        {"name": "uncorr_bkguncrt", "type": "shapesys", "data": [3.0, 7.0]}
                    ],
                }
            ],
        }
    ],
    "observations": [{"name": "singlechannel", "data": [51.0, 48.0]}],
    "measurements": [{"name": "Measurement", "config": {"poi": "mu", "parameters": []}}],
    "version": "1.0.0",
}

signal_patch = [
    {
        "op": "add",
        "path": "/channels/0/samples/1",
        "value": {
            "name": "signal",
            "data": [12.0, 11.0],
            "modifiers": [{"name": "mu", "type": "normfactor", "data": None}],
        },
    }
]

wrapper = spey.get_backend("pyhf")
model = wrapper(
    analysis="simple_pyhf",
    background_only_model=background_only,
    signal_patch=signal_patch,
)

print(model.exclusion_confidence_level())  # [0.9474850259721279]

Because Spey is fully backend-agnostic, every method of spey.StatisticalModel (likelihood scans, upper limits, combinations, etc.) works the same way on top of pyhf. For the full API and tutorials, see the online documentation.