pyhf provides a pure-Python implementation of the HistFactory statistical model for multi-bin histogram-based analyses, enabling statistical inference and limit setting outside of the traditional ROOT/RooFit framework. It targets physicists and researchers in high-energy physics and related fields, offering modern computational features like autodifferentiation and GPU acceleration via backends such as PyTorch and TensorFlow.

How It Works

pyhf implements the HistFactory statistical model, which defines probability density functions (PDFs) for histogram-based measurements. It leverages computational graph libraries (NumPy, PyTorch, TensorFlow, JAX) for efficient computation and automatic differentiation, enabling advanced statistical inference techniques like hypothesis testing and interval estimation based on asymptotic formulas. This approach allows for flexible model definition and leverages modern hardware for faster analysis.

Quick Start & Requirements

• Install with NumPy backend: python -m pip install pyhf
• Install with all backends: python -m pip install pyhf[backends]
• Requires Python 3.7+.
• Official documentation: https://pyhf.readthedocs.io/
• Tutorial: https://pyhf.github.io/pyhf-tutorial/
• Binder: https://mybinder.org/v2/gh/scikit-hep/pyhf/main?labpath=docs%2Fexamples%2Fnotebooks%2Fbinderexample%2FStatisticalAnalysis.ipynb

Highlighted Details

• Supports various systematic uncertainty types: HistoSys, OverallSys, ShapeSys, NormFactor, ShapeFactor, StatError, Lumi Uncertainty.
• Integrates with ROOT/XML via uproot for model import.
• Offers multiple computational backends (NumPy, PyTorch, TensorFlow, JAX) and optimizers (SciPy, iminuit).
• Includes visualization tools for results, comparable to ROOT outputs.

Maintenance & Community

• Developed by Lukas Heinrich, Matthew Feickert, and Giordon Stark, with a list of contributors available.
• Affiliated with NumFOCUS.
• GitHub Discussions for questions: https://github.com/scikit-hep/pyhf/discussions
• GitHub Issues for bug reports: https://github.com/scikit-hep/pyhf/issues/new?template=Bug-Report.md&labels=bug&title=Bug+Report+:+Title+Here
• Announcements mailing list: https://groups.google.com/group/pyhf-announcements/subscribe

Licensing & Compatibility

• Licensed under the BSD 3-Clause License.
• Compatible with commercial and closed-source applications.

Limitations & Caveats

• While it supports importing from ROOT/XML via uproot, it is a pure-Python implementation and does not directly use ROOT libraries for its core statistical calculations.
• Non-asymptotic calculators are implemented, but the primary focus is on asymptotic formulas.