How to properly split train/test with Random seed and np.random.rand()?

Question

Why using numpy.random.seed():

np.random.seed(40)

Does not garant me always the same train/test split with np.random.rand ?