changed logic to use numpy instead, over 15x performance gain according to unprofessional testing

(left is old, right is new)