.0

    If classes members are completely split across different clusters,
    the assignment is totally in-complete, hence the NMI is null::

      >>> normalized_mutual_info_score([0, 0, 0, 0], [0, 1, 2, 3])
      ... # doctest: +SKIP
      0.0
    r