Pandas date-time utilities

Pandas date-time utilities#

Actigraphy rows are timestamps. Pandas can parse them once and then extract calendar fields without string hacking.

Parse once#

df['timestamp'] = pd.to_datetime(df['timestamp'])

After that, use the .dt accessor:

df['hour'] = df['timestamp'].dt.hour
df['weekday'] = df['timestamp'].dt.day_name()
df['date'] = df['timestamp'].dt.date

Two ways to aggregate in time#

groupby on extracted fields#

Good when you want mean activity by person and hour, or by person and weekday:

by_hour = (
    df.groupby(['person_id', 'group', 'hour'], as_index=False)['activity']
      .mean()
)

resample on a DatetimeIndex#

Good when one series is already indexed by time and you want regular bins (15 min, 1 H, 1 D):

one = df.loc[df['person_id'] == 'control_1'].set_index('timestamp').sort_index()
hourly = one['activity'].resample('1h').mean()

resample needs a datetime index. Multi-person tables usually groupby person first, then resample inside each group (or use groupby + Grouper).

Unit-of-analysis warning#

Hourly means still leave many rows per person. A t-test on all hourly rows treats hours as independent subjects and inflates n. Collapse to one summary per person (or a pre-registered person-day model) before the group comparison that answers the module question.

Contents