Tools for Information-Based Feature Selection

A toolbox of fast, native and parallel implementations of various information-based importance criteria estimators and feature selection filters based on them, inspired by the overview by Brown, Pocock, Zhao and Lujan (2012) < http://www.jmlr.org/papers/v13/brown12a.html>. Contains, among other, minimum redundancy maximal relevancy ('mRMR') method by Peng, Long and Ding (2005) ; joint mutual information ('JMI') method by Yang and Moody (1999) < http://papers.nips.cc/paper/1779-data-visualization-and-feature-selection-new-algorithms-for-nongaussian-data>; double input symmetrical relevance ('DISR') method by Meyer and Bontempi (2006) as well as joint mutual information maximisation ('JMIM') method by Bennasar, Hicks and Setchi (2015) .

News

Changes in 5.0:

JIM algorithm as an alternative version of JMI which is using Gini impurity instead of information gain.
Gini impurity feature scorer.
Fixed instability in breaking ties between equivalent features in multi-threaded case. Now, feature of a lowest index is always returned.

Changes in 4.1:

Fixed code used to auto-cut numerical values into bins; maximal value used to be wrapped to the minimal bin, which was substantially altering results.

Changes in 4.0:

MI, CMI, JMI and NJMI feature scorers.
Parallelism is now controlled by threads argument.
Documentation improvements.
Fixed MIM handing of many features with zero score; thanks to smilesun for spotting this.

Changes in 3.0: