numpy.random.multivariate_normal() result is not reproducible even if numpy.random.seed() is set.
np.random.seed(9) class1 = np.random.multivariate_normal(mean, cov, n_samples) class2 = np.random.multivariate_normal(mean, cov, n_samples)
class1 won't be the same as class2. When I generate mean and cov with np.random.rand() it is reproducible. cov is a positive semi-definite matrix I generated using np.dot(A,A.T) where A is generated with random.rand()

This was on Python 2.7.12 with Numpy 1.11.1 on Ubuntu 16.04.1