numpy.unique needs the type to be comparable by < and >, not just ==. that seems like overkill.
a Schelling point for those who seek one
