I was surprised to find Math.NET to lack such a common distribution. I understand that MatrixNormal can be adapted to generate it, with a bit of computational and syntactic overhead. However, this is only possible for positive definite covariance matrices, since MatrixNormal is implemented via a Cholesky decomposition. Positive semi-definite covariance matrices are common and useful. Could a Multivariate Normal decomposition be added? Or else, could MatrixNormal be extended to handle positive semi-definite covariances?
I was surprised to find Math.NET to lack such a common distribution. I understand that
MatrixNormalcan be adapted to generate it, with a bit of computational and syntactic overhead. However, this is only possible for positive definite covariance matrices, sinceMatrixNormalis implemented via a Cholesky decomposition. Positive semi-definite covariance matrices are common and useful. Could a Multivariate Normal decomposition be added? Or else, couldMatrixNormalbe extended to handle positive semi-definite covariances?