count-significance 0.2.0

Accurate significance and p-values for count data

Count Significance

count-significance provides numerical tools for discovery significances and p-values in Poisson counting experiments.

It supports:

• a single Poisson count with known background;
• an on/off experiment with a Poisson control measurement;
• exact or profiled reference calculations;
• first-order profile-likelihood approximations;
• higher-order Barndorff-Nielsen corrections;
• Asimov and Monte Carlo estimates of median discovery significance.

Installation

The package requires Python 3.9 or newer.

python -m pip install count-significance

Example

from count_significance import (
    asimov_Zs_onoff,
    expected_significance_on,
    pvals_on,
    pvals_onoff_profile_sum,
)

known = pvals_on(s0=0.0, b=1.0, n=4)
onoff = pvals_onoff_profile_sum(s=0.0, n=5, m=1, tau=1.0)

known_expected = expected_significance_on(s_true=5.0, b=10.0)
onoff_expected = asimov_Zs_onoff(s_true=5.0, b=10.0, tau=2.0)

The public namespace is count_significance. The package reports one-sided discovery significances and distinguishes numerical reference calculations from asymptotic approximations.

Documentation and reproducibility

The GitHub repository contains the full mathematical overview, API notes, tests, plotting scripts, paper configurations, notebook, generated figures, and HTCondor production workflow.

This software accompanies the manuscript Discovery Sensitivity for a Counting Experiment with Background Uncertainty by Enzo Canonero and Glen Cowan.

Citation

Citation metadata are provided in CITATION.cff.

License

Count Significance is distributed under the MIT License.