Regarding 2., did you have a particular approach in mind? This new lookup table method is already O(n) scaling (similar to a counting sort), so I cannot fathom a method that, as you suggest, would get significantly better performance for integer arrays. The sorting here is "free" in some sense since you aren't spending additional cycles, the table is initialized pre-sorted. I could see in the future that one could create a similar approach for arbitrary datatypes, and you would use a hash table rather than a fixed-size array. In this case you would similarly get O(n) performance, and would produce an **unsorted** output. But in any case a hash table would probably be much less efficient for integers than an array as implemented here - the latter which also gives you a sorted output as a side effect. Personally I have used np.unique for integer data far more than any other type, and I would _guess_ it is similar in the broader community (though I have no data or insight to support that)–so I think having a big speedup like this could be quite nice for users. Thanks, Miles