Why is dill much faster and more disk-efficient than pickle for numpy arrays Why is dill much faster and more disk-efficient than pickle for numpy arrays numpy numpy