lowess 1.0.1

Lowess smoothed as defined for STATA 13.

The lowess Package

This package provides a function to perform a LOWESS on Pandas Series objects. LOWESS (locally weighted scatterplot smoothing) [1, 2] as defined by STATA [3]. The regressions utilises some of the methods in [4].

Description

Methods and Formula

Let x and y be the two variables each of length N, and assume that the data are ordered so that x_i = < x_i+1 for i = 1,...,N-1. For each y_i, a smoothed value y_i^s is calculated. The subset of points used in calculating y_i^s is i_- = max(1, i-k) through i₊ = min(i+k, N), where

k = Floor((N × bandwidth - 0.5) / 2).

The weights for each of the observations between j = i_-,...,i₊ are the tricube

w_j = [1 - (|x_j - x_i| / ∆)³]³,

where ∆ = 1.0001 max(x_i+-x_i, x_i-x_i-). The smoothed value y_i^s is then the weighted polynomial regression prediction at x_i.

NB: In this implemtation x and y should be Pandas Series objects. The series need not be sorted and x and y can be in different orders, so long as their indexes have the same elements.

Usage

Once the package has been installed it can be imported into a python script
import lowess
The package provides a single module lowess with a single function lowess.lowess. This function has the signiture:
lowess.lowess(x, y, bandwidth=0.2, polynomialDegree=1)
where the arguments are:

1. x (pandas.core.series.Series): a Pandas Series containing the x (independent/covariat) values.
2. y (pandas.core.series.Series): a Pandas Series containing the y (dependent) values. It must have the same index as x (although not necessarily in the same order.)
3. bandwidth (float, optional): the bandwidth for smoothing. It must be between 0 and 1. Default is 0.2
4. polynomialDegree (int, optional): The degree of polynomial to use in the regression. It must be >= 0. Default is 1.

It returns a Pandas Series containing the smoothed y values, with the same index as y.

If input is not valid or an error occurs, a LowessError exception is raised.

Examples

Some examples are given in the directory examples.

Instalation

The package can be installed with pip via the command: $ pip install lowess
Or download from GitHub and installed via

$ git clone [e-mail]:CCGE-Cambridge/lowess.git
$ cd lowess
$ pip install .

Documentaion

Documentaion of the API is provided via Sphinx. To make the cocumentaion

$ cd docs
$ make html
$ open build/html/index.html

This may require installation of the package sphinx.

Testing

Unit tests are implemented via unittest and are in the file tests/test_lowess.py. To run the tests first install the package and then run the command:
$ python tests/test_lowess.py

Requirements

numpy==1.18.2
pandas==1.0.3
python-dateutil==2.8.1
pytz==2019.3
six==1.14.0

License

Copyright (c) 2020 Andrew Lee

This software is provided as is without any warranty whatsoever. Permission to use, for non-commercial purposes is granted. Permission to modify for personal or internal use is granted, provided this copyright and disclaimer are included in all copies of the software. All other rights are reserved. In particular, redistribution of the code is not allowed.

References

1. Cleveland, W. S. 1979. Robust locally weighted regression and smoothing scatterplots. Journal of the American Statistical Association 74: 829–836. [https://www.jstor.org/stable/2286407]
2. Wikipedia: Local Regression - [https://en.wikipedia.org/wiki/Local_regression] (accessed 2020-04-20)
3. STATA: Lowess - [https://www.stata.com/manuals13/rlowess.pdf] (accessed 2020-04-20)
4. Cappellari et al. 2013 The ATLAS^3D project - XX. Mass-size and mass-σ distributions of early-type galaxies: bulge fraction drives kinematics, mass-to-light ratio, molecular gas fraction and stellar initial mass function Monthly Notices of the Royal Astronomical Society 432: 1862-1893 [https://doi.org/10.1093/mnras/stt644]