I want to create a multivariate normal distribution that spans features across multiple dimensions. For example, the mean tensor is of shape
(10, 2) and I want all entries to be possibly correlated, which means I provide a tensor with 20^2 entries that represent the covariance. Is this achievable using the
MultivariateNormal object of torch distribution? I would expect a distribution of event shape
(10, 2) then.