Projection Pursuit

Documentation, How it works.

This repository is home to a couple scikit-learn-compatible estimators based on Jerome Friedman's generalizations[1] of his and Werner Stuetzle's Projection Pursuit Regression algorithm[2][3]. A regressor capable of multivariate estimation and dimensionality reduction and a univariate classifier based on regression to a one-hot multivariate representation are included.

This repository is also meant to serve as a fairly pared-down example of how to use Github Actions, Coveralls, Sphinx, PyTest, how to deploy to PyPI and Github Pages, and how to create a Scikit-Learn Estimator that passes the sklearn checks and follows the PEP 8 style standard.

Installation and Usage

The package by itself comes with a single module containing the estimators. Before installing the module you will need numpy, scipy, scikit-learn, and matplotlib. To install the module execute:

pip install projection-pursuit

or

$ python setup.py install

If the installation is successful, you should be able to execute the following in Python:

>>> from skpp import ProjectionPursuitRegressor
>>> estimator = ProjectionPursuitRegressor()
>>> estimator.fit(np.arange(10).reshape(10, 1), np.arange(10))

Sphinx is run via continuous integration to generate the API.

For a few usage examples, see the examples and benchmarks directories. For an intuition of what the learner is doing, try running viz_training_process.py. For comparisons to other learners and an intuition of why you might want to try PPR, try the benchmarks. For a deep dive in to the math and an explanation of exactly how and why this works, see math.pdf.

References

1. Friedman, Jerome. (1985). "Classification and Multiple Regression Through Projection Pursuit." http://www.slac.stanford.edu/pubs/slacpubs/3750/slac-pub-3824.pdf
2. Hastie, Tibshirani, & Friedman. (2016). The Elements of Statistical Learning 2nd Ed., section 11.2.
3. (2017) Projection pursuit regression https://en.wikipedia.org/wiki/Projection_pursuit_regression