Covariance

CovarianceMatrix::from_observations computes the unbiased sample covariance from finite row-major observations, requiring at least two equally wide rows. from_row_major accepts an explicitly symmetric finite matrix, and quadratic_form validates gradient dimension before computing gᵀΣg.

The covariance is not regularized, factorized, or required to be positive semidefinite. A symmetric supplied matrix can still be unsuitable for a probabilistic interpretation.