stjordanis / featureforge

A set of tools for creating and testing machine learning features, with a scikit-learn compatible API

Feature Forge

This library provides a set of tools that can be useful in many machine learning applications (classification, clustering, regression, etc.), and particularly helpful if you use scikit-learn (although this can work if you have a different algorithm).

Most machine learning problems involve an step of feature definition and preprocessing. Feature Forge helps you with:

• Defining and documenting features
• Testing your features against specified cases and against randomly generated cases (stress-testing). This helps you making your application more robust against invalid/misformatted input data. This also helps you checking that low-relevance results when doing feature analysis is actually because the feature is bad, and not because there's a slight bug in your feature code.
• Evaluating your features on a data set, producing a feature evaluation matrix. The evaluator has a robust mode that allows you some tolerance both for invalid data and buggy features.
• Experimentation: running, registering, classifying and reproducing experiments for determining best settings for your problems.

Installation

Just pip install featureforge.

Documentation

Documentation is available at http://feature-forge.readthedocs.org/en/latest/

Contact information

Feature Forge is copyright 2014 Machinalis (http://www.machinalis.com/). Its primary authors are:

• Javier Mansilla <[email protected]> (jmansilla at github)
• Daniel Moisset <[email protected]> (dmoisset at github)
• Rafael Carrascosa <[email protected]> (rafacarrascosa at github)

Any contributions or suggestions are welcome, the official channel for this is submitting github pull requests or issues.

Changelog

0.1.7:

• StatsManager api change (order of arguments swapped)
• For experimentation, enabled a way of booking experiments forever.

0.1.6:

• Bug fixes related to sparse matrices.
• Small documentation improvements.
• Reduced default logging verbosity.

0.1.5:

• Using sparse numpy matrices by default.

0.1.4:

• Discarded the need of using forked version of Schema library.

0.1.3:

• Added support for running and generating stats for experiments

0.1.2:

• Fixing installer dependencies

0.1.1:

• Added support for python 3
• Added support for bag-of-words features

0.1:

• Initial release