Covariance reducing models: An alternative to spectral modelling of covariance matrices

Abstract

We introduce a software package running under Matlab that implements several recently proposed likelihood-based methods for sufficient dimension reduction. Current capabilities include estimation of reduced subspaces with a fixed dimension d, as well as estimation of d by use of likelihood-ratio testing, permutation testing and information criteria. The methods are suitable for preprocessing data for both regression and classification. Implementations of related estimators are also available. Although the software is more oriented to command-line operation, a graphical user interface is also provided for prototype computations.


Published: Biometrika 95 (2008), no. 4, 799--812