Changed the group index method to the same method as used by Python's dict
This will increase the speed for aggregations over larger multi-column aggregations. For small multi-column aggregations it's still slightly faster.
There are two future enhancements still:
Make creating the group index carray based instead of numpy (which will greatly improve memory usage)
Cache multi-column unique combinations, which will greatly enhance the speed
Changed the group index method to the same method as used by Python's dict
This will increase the speed for aggregations over larger multi-column aggregations. For small multi-column aggregations it's still slightly faster. There are two future enhancements still: