mahon 0.1.0

Linear regression of data sets with correlated and uncorrelated uncertainties.

Mahon linear regression

The goal of this project is to provide a python package that allows to calculate linear regressions on data sets with uncertainties. Linear regressions can be performed for correlated and uncorrelated uncertainties. The calculations follow the methodology published by Mahon (1996). Typographical errors that were made in that work have been corrected. A method to allow calculating a linear regression through a fixed point, avoiding previously made errors, is also provided.

Installation

The package can be installed from pypi via:

pip install mahon

Usage

Below is an example on how to use the package.

>>> import numpy as np
>>> from mahon import LinReg

>>> # some data
>>> xdata = np.array([1, 2, 3.1, 4.9])
>>> ydata = np.array([1.1, 1.9, 3, 5.5])

>>> # some uncertainty and correlation
>>> xunc = 0.05 * xdata
>>> yunc = 0.073 * ydata
>>> rho = np.zeros_like(xdata) + 0.5

>>> # do regression
>>> my_reg = LinReg(xdata, xunc, ydata, yunc, rho)

>>> # print out the parameters and their uncertainties
>>> my_reg.slope
(0.9983617286559998, 0.0684389236571533)
>>> my_reg.intercept
(0.05545339135826666, 0.11811730191506546)
>>> my_reg.mswd
2.5105964767246842

A detailed example for correlated and uncorrelated uncertainties, and fixed point linear regressions can be found in this Jupyter notebook.

Development & Contributing

If you would like to contribute, clone the GitHub repo and then install the package locally from within the folder via:

pip install -e .[dev]

If you also want to install the full test environment, run:

pip install -e .[dev,test]

Code auto formatting is implemented using pre-commit hooks. Full testing of the package can be done with nox.

Please feel free to raise issues on GitHub and open pull requests if you have a feature to be added. Tests and adequate docstrings should be provided along with your new code.