HiDimDA (version 0.2-4)

MldaInvE: Maximum uncertainty Linear Discriminant Analysis inverse matrix estimator.

Description

Builds a well-conditioned estimator for the inverse of a symmetric positive definite matrix, with a bad-conditioned or singular estimate, based on the “Maximum Uncertainty Linear Discriminant Analysis” approach of Thomaz, Kitani and Gillies (2006).

Usage

MldaInvE(M, check=TRUE, onlyMinv=TRUE, 
numtol=sqrt(.Machine$double.eps))

Value

If onlyMinv is set to true a matrix with the inverse estimate sought. Otherwise a list with components ME and MInvE, with a well-conditioned approximation to the matrix that M estimates and its inverse.

Arguments

M

Singular or bad-conditioned estimate of the matrix for which a well-conditioned inverse estimate is sought.

check

Boolean flag indicating if the symmetry of M and the sign of its eigenvalues should be check upfront.

onlyMinv

Boolean flag indicating if only an estimate of the matrix inverse is sought, or if a well-conditioned approximation to the matrix that M estimates should be returned as well.

numtol

Numerical tolerance.

Author

A. Pedro Duarte Silva

References

Thomaz, Kitani and Gillies (2006) “A maximum uncertainty LDA-based approach for limited sample size problems - with application to face recognition”, Journal of the Brazilian Computer Society, 12 (2), 7-18

See Also

Mlda