segreg is a Python module for segmented regression.
Segmented regression models are defined as having breakpoints where the functional form changes. This project is currently limited in scope to the case of piecewise-linear and continuous univariate models with at most two breakpoints.
The segreg code fits segmented regression models to data using an exact algorithm due to Hudson.
The primary implementation is based on cython. Alternative implementations (of the same
algorithm)
are also provided in pure python, with or without numba.
To accomodate bootstrap and other monte carlo routines, it helps for the model fitting to be fast and exact. By exact, we mean that the optimal parameter values minimizing the residual sum of squares function are found. This is in contrast to numerically approximate routines such as grid search or non-linear optimization.
Hudson's algorithm is exact, but can involve a fair amount of computation for two or more breakpoints. The implementation here strives to be fast.
This is a pre-release. Further small changes will be made before the code is considered available for general use.
pip install segreg
We strongly recommended using a virtual environment (venv), or a conda environment, to avoid possible conflicts with other packages or other issues.
segreg is licensed under a BSD-3-Clause License. See LICENSE.
For a technical overview of Segmented Regression and algorithms used in segreg,
see Technical Paper.
Code documentation and tutorials are available here: https://stevelill.github.io/segreg/
To build core modules, run this:
python setup.py build_ext --inplace
This builds C code which has already been generated using cython. As such,
there is no direct cython dependency.