Closed gettodaze closed 4 months ago
Interestingly, out-performance degrades with scale. No material difference seen due to the size of the tuple. Might be able un-roll the loop, or special case a contiguous input array.
There was a flaw in the preparation of the fixture arrays used in the performance tests (row / column sizes were swapped). Fixing this error, we now see excellent out-performance maintained with scale; perhaps more intuitively, we see out-performance diminish with tuple size.
I have created an alternate iterator interface, array2d_tuple_iter
, that iterates lazily over tuples rather than creating a 1D array. This too shows excellent out-performance to native Python alternatives, both in creating a list and in just calling next()
.
Yay! Thanks @flexatone!
From https://github.com/static-frame/arraykit/issues/22