Closed carlthome closed 1 year ago
The only quibble I would have is that we could be a little more careful with the precision here. For example, np.int64 is overkill for most uses, and in some cases we'd actually prefer uint types (eg if it's do be used as an array index).
This mostly doesn't matter, but we may as well be efficient if possible.
Yeah! Good point! 🎉
Reflected on this as well when refactoring but was unsure what could be made smaller. Figured the current behaviour was longs and doubles, if I've understood the removed np.int
, np.float
aliases correctly, and that not changing behaviour would be safest.
Would love to take the opportunity to set more appropriate types for each metric here.
Backwards-compatibility on the first pass seems reasonable to me - will merge. Thanks both.
As per https://numpy.org/devdocs/release/1.24.0-notes.html#expired-deprecations these aliases are deprecated and now cause errors.