README.rst

sklearn-sfa - An implementation of Slow Feature Analysis compatible with scikit-learn

sklearn-sfa or sksfa is an implementation of Slow Feature Analysis for scikit-learn.

It is meant as a standalone transformer for dimensionality reduction or as a building block for more complex representation learning pipelines utilizing scikit-learn's extensive collection of machine learning methods.

The package contains a solver for linear SFA and some auxiliary functions. The documentation provides an explanation of the algorithm, different use-cases, as well as pointers how to fully utilize SFA's potential, e.g., by employing non-linear basis functions or more sophisticated architectures.

For use with high-dimensional image data, sklearn-sfa now also includes an experimental implementation of Hierarchical SFA networks (HSFA) - please consult the introductory examples in the documentation.

Since sklearn-sfa is in its early stages, we also recommend taking a look at the Modular Toolkit for Data Processing MDP which provides stable SFA implementations that have stood the test of time.

Installation

The package can be installed via pip:

pip install --user sklearn-sfa

Basic usage

In Python 3.6+, the package can then be imported as

import sksfa

The package comes with an SFA transformer. Below you see an example of initializing a transformer that extracts 2-dimensional features:

sfa_transformer = sksfa.SFA(n_components=2)

The transformer implements sklearn's typical interface by providing fit, fit_transform, and transform methods.