Skip to main content

statistics tools and utilities

Project description

scikit-stats: statistics tools and utilities

Build Status Azure DevOps tests Azure DevOps coverage Binder PyPI PyPI - Python Version DOI


Install scikit-stats like any other Python package:

pip install scikit-stats

or similar (use --user, virtualenv, etc. if you wish).

Getting Started

The scikit-stats module includes modeling and hypothesis tests submodules. This a quick user guide to each submodule. The binder examples are also a good way to get started.


The modeling submodule includes the Bayesian Block algorithm that can be used to improve the binning of histograms. The visual improvement can be dramatic, and more importantly, this algorithm produces histograms that accurately represent the underlying distribution while being robust to statistical fluctuations. Here is a small example of the algorithm applied on Laplacian sampled data, compared to a histogram of this sample with a fine binning.

>>> import numpy as np
>>> import matplotlib.pyplot as plt
>>> from skstats.modeling import bayesian_blocks

>>> data = np.random.laplace(size=10000)
>>> blocks = bayesian_blocks(data)

>>> plt.hist(data, bins=1000, label='Fine Binning', density=True, alpha=0.6)
>>> plt.hist(data, bins=blocks, label='Bayesian Blocks', histtype='step', density=True, linewidth=2)
>>> plt.legend(loc=2)

bayesian blocks example


This submodule provides tools to do hypothesis tests such as discovery test and computations of upper limits or confidence intervals. scikit-stats needs a fitting backend to perform computations such as zfit. Any fitting library can be used if their API is compatible with scikit-stats (see api checks).

We give here a simple example of a discovery test, using zfit as backend, of gaussian signal with known mean and sigma over an exponential background.

>>> import zfit
>>> from zfit.loss import ExtendedUnbinnedNLL
>>> from zfit.minimize import Minuit

>>> bounds = (0.1, 3.0)
>>> obs = zfit.Space('x', limits=bounds)

>>> bkg = np.random.exponential(0.5, 300)
>>> peak = np.random.normal(1.2, 0.1, 25)
>>> data = np.concatenate((bkg, peak))
>>> data = data[(data > bounds[0]) & (data < bounds[1])]
>>> N = data.size
>>> data = zfit.Data.from_numpy(obs=obs, array=data)

>>> lambda_ = zfit.Parameter("lambda", -2.0, -4.0, -1.0)
>>> Nsig = zfit.Parameter("Ns", 20., -20., N)
>>> Nbkg = zfit.Parameter("Nbkg", N, 0., N*1.1)
>>> signal = Nsig * zfit.pdf.Gauss(obs=obs, mu=1.2, sigma=0.1)
>>> background = Nbkg * zfit.pdf.Exponential(obs=obs, lambda_=lambda_)
>>> loss = ExtendedUnbinnedNLL(model=signal + background, data=data)

>>> from skstats.hypotests.calculators import AsymptoticCalculator
>>> from skstats.hypotests import Discovery
>>> from skstats.hypotests.parameters import POI

>>> calculator = AsymptoticCalculator(loss, Minuit())
>>> poinull = POI(Nsig, 0)
>>> discovery_test = Discovery(calculator, [poinull])
>>> discovery_test.result()

p_value for the Null hypothesis = 0.0007571045424956679
Significance (in units of sigma) = 3.1719464825102244

The discovery test prints out the pvalue and the significance of the null hypothesis to be rejected.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

scikit-stats-0.1.2.tar.gz (16.7 kB view hashes)

Uploaded Source

Built Distribution

scikit_stats-0.1.2-py2.py3-none-any.whl (24.6 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page