ENH: Add dtype_backend parameter to to_datetime

[wirable23]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

>>> a = pd.Series([4194304], dtype="int64[pyarrow]")
>>> pd.to_datetime(a).dtype
dtype('<M8[ns]')
>>>

In the above example, passing in a pyarrow backed series to to_datetime returns a numpy backed series, in to_numeric, there's a dtype_backend parameter that allows preserving of pyarrow dtypes, it would be nice to have the same option in to_datetime.

Feature Description

Same way as in to_numeric, adding a dtype_backend parameter.

Alternative Solutions

N/A

Additional Context

No response

[phofl]

Yep this sounds good. Same as for to_numeric basically. Pinging @MarcoGorelli

[MarcoGorelli]

sounds good - interested in submitting a pull request @wirable23 ?

[wirable23]

@MarcoGorelli yes! The issue I was facing was that I didn't see arrow compute functions for some functionality of to_datetime, for example, pyarrow does not seem to have a compute function for creating datetimes from day month year components. Also, pyarrow does not support casting floating points to datetimes but there might be some other compute function that can be used.

[ruimamaral]

@lithomas1 @wirable23 I see there haven't been any updates to this issue in a while.
If you're both okay with it, I would like to implement this functionality.
Thanks!

[lithomas1]

Go for it.

[ruimamaral]

take

[ruimamaral]

I initially assigned myself as I thought I had a good idea of how this should be implemented, however after looking more deeply into to_datetime and pyarrow, I realize that I might've bit off more than I could chew...

My main issue is trying to figure out the best places to convert the dates between their numpy and pyarrow counterparts.
Since I don't feel certain about the full implementation details, I think it would be best to leave this one out to someone more familiar with pyarrow and this part of the codebase.

Therefore, I will be unassigning myself from this issue.

I apologize for any inconvenience caused!