Turns out np.roll is much more expensive than initially envisioned. By unrolling this (no pun intended) we can get the same result in around 5-6x less time. This results in substantial speedups for longer runsm pairwise comparisons, and more complex derivations such as 3mr. Adding also uint indices as apparently there is no overflow checking in this case.
Turns out
np.rollis much more expensive than initially envisioned. By unrolling this (no pun intended) we can get the same result in around 5-6x less time. This results in substantial speedups for longer runsm pairwise comparisons, and more complex derivations such as 3mr. Adding alsouintindices as apparently there is no overflow checking in this case.