This repository contains a MATLAB implementation of the group-sparse sensor selection method for GEVD problems described in [1].
This package requires CVX.
The software was developed with CVX version 2.2 using the MOSEK optimizer.
The software performs sensor selection in the context of GEVD problems were two classes X1 and X2 are discriminated.
In its most basic usage and given the covariance matrices R1 = cov(X1) and R2 = cov(X2), the software can be used the optimal subset of nbGroupToSel sensors:
gsl1infSensorSelection(R1, R2, nbGroupToSel);
The package allows to specify the desired number of eigenvector (filter) outputs though the argument K:
gsl1infSensorSelection(R1, R2, nbGroupToSel, K);
The package allows for grouped sensor selection by defining groupSelector. The variable is a nbVariables x nbGroups binary matrix, indicating per group (column) which variables of the covariance matrices belong to that group with ones at the corresponding positions.
gsl1infSensorSelection(R1, R2, nbGroupToSel, K, groupSelector );
The search options for the problem are set with the params argument.
--
[1] J. Dan, S. Geirnaert, A. Bertrand, "Grouped Variable Selection for Generalized Eigenvalue Problems," 2021. arXiv:2105.13667 (https://arxiv.org/abs/2105.13667)