fastprof 0.4.1

'A fast profiling tool for likelihood-based statistical analyses'

fastprof: A fast profiling tool for likelihood-based statistical analyses

Presentation

The tool is intended to provide a simple way to reuse, reinterpret and combine public likelihoods, in particular the ones used in high-energy physics experiments. It implements a simplified likelihood format, *Simplified Likelihoods with Linearized Systematics (SLLS), which provides good approximations to full likelihoods, but is faster to evaluate.

The SLLS format is similar to the HistFactory format implemented for instance in pyhf: it defines measurement bins with yields expressed in terms of model parameters that be be either parameters of interest (POIs) or nuisance parameters (NPs). The simplification occurs in the treatment of the NPs, which are considered at linear order only. This in turns provides a simple way to maximize the likelihood with respect to the NPs using matrix algebra, which is typically much faster than the maximization of the full likelihood.

SLLS models are intended to be used to perform statistical inference using classical frequentist techniques: computing p-values, discovery significances and confidence intervals, and setting limits on new phenomena. Along with the likelihood itself, the package includes a set of frequentist statistical tools, based on the techniques described in arXiv:1007.1727

The tool is written in python, with most of the work done by numpy, and models are stored in JSON or YAML markup files. Models implemented within pyhf or roofit can be automatically converted to the SLLS format using tools provided in the package.

Setup

The package can be set up as follows:

git clone ssh://[e-mail]:7999/nberger/fastprof.git
cd fastprof
git checkout v0.4.1 -b v0.4.1
source ./setup-env.sh

This sets up the latest stable version,v0.4.1. (skipping this line sets up the latest master version instead, but this is not recommended). The last command sets up a python3 pyenv working environment.

Alternatively, the package can be installed directly using

pip install fastprof

Documentation

Detailed documentation can be found in the package itself in build/sphinx/html, or on the documentation website. SLLS likelihoods and examples of their use are described in detail in arXiv:2301.05676.