Truncate / floor a datetime

To truncate a datetime, use dt.floor()

# Generate datetimes precise to the millisecond
df = pd.DataFrame({
    'datetime': pd.to_datetime(
				np.random.randint(1673897218000, 1681673218000, 5), unit='ms'
		)
})
df

For example, to floor to the second, hour or day:

(
    df
    .assign(
        second=lambda x: x['datetime'].dt.floor('s'),
        hour=lambda x: x['datetime'].dt.floor('h'),
        date=lambda x: x['datetime'].dt.floor('d'),
    )
)

Another trick: if you’re converting a string to datetime, and don’t want to be more precise than the second, you can convert the string to datetime64[s] instead of datetime64 that has a precision up to the nanosecond:

# Convert datetimes to string
df['datetime'] = df['datetime'].astype('str')

# Convert to datetime only precise to the second
df['datetime'].astype('datetime64[s]')

0   2023-03-17 20:34:47
1   2023-04-04 17:47:19
2   2023-03-09 01:51:14
3   2023-02-27 16:20:03
4   2023-03-29 21:04:24
Name: datetime, dtype: datetime64[ns]

Change frequency of a date

To cast a date or datetime to a less granular frequency, use dt.to_period() with a specified offset, for example:

D: calendar day
W: week
M: month
Q: quarter
Y: year

# Truncate date to month
df['date_converted'].dt.to_period('M')

0    2020-01
1    2020-02
2    2020-03
3    2020-04
4    2020-05
Name: date_converted, dtype: period[M]

# Truncate date to quarter
df['date_converted'].dt.to_period('Q')

0    2020Q1
1    2020Q1
2    2020Q1
3    2020Q2
4    2020Q2
Name: date_converted, dtype: period[Q-DEC]

Floor datetimes and dates

Truncate / floor a datetime

Change frequency of a date