Utility belt for spey-pyhf plug-in

Working with full likelihoods from HEPData

The spey-pyhf plug-in includes a set of helper functions to manipulate the input data in a spey-ready fashion.

Show code cell content

Hide code cell content

import spey
from spey_pyhf.helper_functions import WorkspaceInterpreter
import json

We will use an example background-only JSON file which is downloaded from HEPData, but you can use any background-only JSON file.

Load the background file. You can use any background only JSON file for this example. Simply adjust the rest of the code accordingly.

with open("/PATH/TO/BkgOnly.json", "r") as f:
    bkg_only = json.load(f)

spey_pyhf.helper_functions.WorkspaceInterpreter scans the background-only dictionary and extracts relevant information such as channel names:

interpreter = WorkspaceInterpreter(bkg_only)
print(list(interpreter.channels))

['WREM_cuts', 'STCREM_cuts', 'TRHMEM_cuts', 'TRMMEM_cuts', 'TRLMEM_cuts', 'SRHMEM_mct2', 'SRMMEM_mct2', 'SRLMEM_mct2']

and information about bin sizes in each channel

print(interpreter.bin_map['SRHMEM_mct2'])

3

we can inject signal to any channel we like

Attention!

Notice that the rest of the channels will be added without any signal yields. If some of these channels need to be removed from the patch set, they can be added to the remove list via the remove_channel() function. Note: This behaviour has been updated in v0.1.5. In the older versions, the channels that were not declared were removed.

interpreter.inject_signal('SRHMEM_mct2', [5.0, 12.0, 4.0])

Notice that we only added 3 inputs since the "SRHMEM_mct2" region has only 3 bins. One can inject signals to as many channels as one wants, but for simplicity, we will use only one channel. Now we are ready to export this signal patch and compute the exclusion limit

pdf_wrapper = spey.get_backend("pyhf")
statistical_model = pdf_wrapper(
    analysis="simple_pyhf",
    background_only_model=interpreter.background_only_model,
    signal_patch=interpreter.make_patch(),
)
print(statistical_model.exclusion_confidence_level())

[0.9856303068018685]