[ARROW-3899] [Python] Table.to_pandas converts Arrow date32[day] to pandas datetime64[ns]

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.12.0
• Component/s: Python
• External issue URL: https://github.com/apache/arrow/issues/20514
• Assignee: Wes McKinney
• Reporter: Wes McKinney
• Created: 29/Nov/18 03:09
• Updated: 11/Jan/23 07:30
• Resolved: 08/Feb/19 04:31

Description

This issue was raised here:

https://github.com/wesm/feather/issues/359

I explored this minimally against Arrow master:

https://gist.github.com/wesm/2ebe0ca2461d1ecfba6185777238ad1f

While it's pretty memory-wasteful, it might be better to preserve the intent of the data type when converting to pandas data structures. It also allows the data to round trip successfully

Issue Links

• is related to: ARROW-3911 [Python] Deduplicate datetime.date objects in Table.to_pandas internals
• relates to: ARROW-3910 [Python] Set date_as_object to True in *.to_pandas as default after deduplicating logic implemented