Feature Selection Package Documentation

Feature Selection Package - Algorithms - Spectral Feature Selection

Description

Spectral feature selection is a feature selection strategy that is largely based on the work of spectral graph theory. This is the unification of supervised and unsupervised feature selection.

Usage

Method Signature:

[ wFeat, SF ] = fsSpectrum( W, X, style, spec )

Output:
wFeat: The weight of the features.
SF: The statistical relevance of the features.
Input:
W: The similarity matrix or a kernel matrix for the features.
X: The input data, each row is an instance.
style: Must be one of the following values: -1: use all, 0: use all except the 1st, k: use first k except 1st.
spec: The spectral function to modify the eigen values.

Code Example

% Using the wine.dat data set, which can be found at
% [fspackage_location]/classifiers/knn/wine.mat
W = X*X';
fsSpectrum(W,X,-1);

Keyword in Evaluator Framework

spectrum

Paper

BibTex entry for:

Spectral Feature Selection for Supervised and Unsupervised Learning by Zheng Zhao and Huan Liu.

@article{LiuZhao,
  title = {Spectral Feature Selection for Supervised and Unsupervised Learning},
   author = {Liu, Huan and Zhao, Zheng},
   year = {2007},
   journal = {Proceedings of the 24th International Conference on Machine Learning}
}