view general/numerical/matrix/logdet.m @ 61:eff6bddf82e3 tip

Finally implemented perceptual brightness thing.

function y = logdet(A)
% logdet - logdet(X)=log(det(X)) where A is positive-definite and Hermitian.
%
% logdet :: [[N,N]] -> nonneg.
%
% This is faster and more stable than using log(det(A)).
% Samer: Use LOGABSDET for general matrices.

%  From Tom Minka's lightspeed toolbox

[U,p] = chol(A);
if p>0, y=-inf;
else y = 2*sum(log(full(diag(U))));
end