findthegap 0.0.8

Tools for finding gaps and valleys in data distribution with a twice-differentiable density estimator with finite support.

FindTheGap

This package (also unformally known as "Gappy") provides tools for geometric data analysis, targeted at finding gaps and valleys in data distribution. It provides a (twice-differentiable) density estimator (Quartic Kernel Density Estimator) relying on pytorch for auto-differentiation, a routine to approximate critical points in the density ,and various statistics to identify and trace `gaps' and valleys in the distribution.

This package can be installed through pip (https://pypi.org/project/findthegap/):

pip install findthegap 

See https://github.com/contardog/findthegap for demo and usecase notebook in the folder 'examples'.

Notebook requirements: sklearn, matplotlib

The folder 'examples' contains a notebook showcasing how to use those tools on 2D data (available in the folder data).

Disclaimer: this code is work in progress and might go through some changes especially for higher (>2!) dimension...

Contributors: Gabriella Contardo (CCA at Simons Foundation), David W. Hogg(CCA/NYU/MPIA), Jason S.A. Hunt (CCA)

You can find more information about the methods in the paper "The emptiness inside: Finding gaps, valleys, and lacunae with geometric data analysis" https://arxiv.org/abs/2201.10674

---

Dependencies:

• numpy >= 1.19.5

• torch >= 1.10.1

• scipy >= 1.5.4

Update version 0.0.5: fix path computation with gradient descent.

Update version 0.0.6: changed score_samples in quarticKDE for speed and memory.

Update version 0.0.7: Fix bug in changed score_samples

Update version 0.0.8: Make the description reappear?!