This is a general purpose statistical analysis framework that uses an adaptive bandwidth kernel density estimation technique to construct smooth probability distribution functions with finite data, which are then used to statistically extract parameters from a dataset using either binned or unbinned maximum likelihood fits.

Where possible, and if available, calculations are accelerated on a GPU.

See my blog post on kernel density estimation for an overview of the mathematics underpinning this implementation of kernel density estimation, as well as brief description of how to use this package.

The examples directory also contains some toy fits in Jupyter notebooks.

Currently kdfit is pre-alpha. API is subject to change at any time.

This project depends on features present in Python 3.9+. The included setup.py will install all necessary dependencies:

• CuPy
• NumPy
• SciPy
• Uproot4 - optional
• h5py - optional

First clone this repository:

git clone https://github.com/BenLand100/kdfit

And then install with pip:

pip install --user -e kdfit

This will create an editable installation that will link to the cloned git repo. This means the kdfit code installed on your system will automatically be updated if the repo is updated or edited Reinstallation may still be necessary if there are new dependencies.

Copyright 2021 by Benjamin J. Land (a.k.a. BenLand100). Released under the GPLv3 license.