Pandas round to nearest hour inconsistently-CodePudding

If I ask pandas to round a datetime to the nearest hour, there is not consistent rounding to the nearest hour from the halfway point. Odd hours like 5 have their half-hour (5:30) rounded up, but even hours have their half hour rounded down. E.g. 5:30 and 6:30 are both rounded to 6.

Why? And how do I get it to do something consistent like round every half-hour up?

>df['nearest_hour']=df['datetime'].dt.round('H')
>df
    datetime    nearest_hour
141470  2019-10-26 04:45:00 00:00   2019-10-26 05:00:00 00:00
294282  2019-10-26 05:00:00 00:00   2019-10-26 05:00:00 00:00
178460  2019-10-26 05:15:00 00:00   2019-10-26 05:00:00 00:00
92696   2019-10-26 05:30:00 00:00   2019-10-26 06:00:00 00:00
227946  2019-10-26 05:45:00 00:00   2019-10-26 06:00:00 00:00
216924  2019-10-26 06:00:00 00:00   2019-10-26 06:00:00 00:00
167823  2019-10-26 06:15:00 00:00   2019-10-26 06:00:00 00:00
240039  2019-10-26 06:30:00 00:00   2019-10-26 06:00:00 00:00

CodePudding user response：

quoting from this issue on the pandas github:

This the expected behavior, albeit unintuitive the first time you encounter it, and is consistent with the rest of the Python ecosystem. The rule for rounding is that exact halfway cases are rounded to the nearest even result.

See also Python 3.x rounding behavior:

for i in [1.5, 2.5, 3.5, 4.5]:
    print(i, round(i))
    
1.5 2
2.5 2
3.5 4
4.5 4

If you want to achieve your expected rounding behavior nevertheless, you can conditionally floor and ceil to the hour:

df['nearest_hour'] = df["datetime"].dt.ceil("H").where(df["datetime"].dt.minute >= 30, df["datetime"].dt.floor("H"))

df
                   datetime              nearest_hour
0 2019-10-26 04:45:00 00:00 2019-10-26 05:00:00 00:00
1 2019-10-26 05:00:00 00:00 2019-10-26 05:00:00 00:00
2 2019-10-26 05:15:00 00:00 2019-10-26 05:00:00 00:00
3 2019-10-26 05:30:00 00:00 2019-10-26 06:00:00 00:00
4 2019-10-26 05:45:00 00:00 2019-10-26 06:00:00 00:00
5 2019-10-26 06:00:00 00:00 2019-10-26 06:00:00 00:00
6 2019-10-26 06:15:00 00:00 2019-10-26 06:00:00 00:00
7 2019-10-26 06:30:00 00:00 2019-10-26 07:00:00 00:00