np.sum(TimedeltaIndex) should return a Timedelta, not np.timedelta64

[jbrockmendel]

ditto TimedeltaArray. xref #25329.

I think the relevant method is __array_ufunc__, @jorisvandenbossche any idea about this?

[TomAugspurger]

I think the relevant method is __array_ufunc__, @jorisvandenbossche any idea about this?

The datetimelike indexes and arrays don't implement __array_ufunc__ yet, do they? Or are you saying that the fix is to implement it?

The old (pre-numpy 1.13) way of doing this would be with __array_ufunc__, which is defined on our indexes.

[jorisvandenbossche]

The old (pre-numpy 1.13) way of doing this would be with __array_ufunc__, which is defined on our indexes.

I suppose you mean __array_wrap__ ?

[jorisvandenbossche]

See also #22798 and #23293 (that last one is a PR that started implementing this, I hope to find some time soon to pick it up again).

[TomAugspurger]

Yes, sorry.

…

On Fri, Feb 15, 2019 at 10:22 AM Joris Van den Bossche < ***@***.***> wrote: The old (pre-numpy 1.13) way of doing this would be with __array_ufunc__, which is defined on our indexes. I suppose you mean __array_wrap__ ? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#25335 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHIgyck90AcKRJeb708fzL_WyJSEXOks5vNt65gaJpZM4a98EU> .