Numpy warning that the covariance is not PD
Currently numpy.random.multivariate_normal always throws a warning, that even though we clearly have covariances, they are not positive definite. This may be due to the svd algorithm used. In numpy 1.18 there will be a switch to use cholesky instead. Once 1.18 is available via conda, we should use this switch and see wether it makes the warning disappear.
The corresponding line is here.