问题
The following DF represents events received from users. Id of the user and the timestamp of the event:
id timestamp
0 1 2020-09-01 18:14:35
1 1 2020-09-01 18:14:39
2 1 2020-09-01 18:14:40
3 1 2020-09-01 02:09:22
4 1 2020-09-01 02:09:35
5 1 2020-09-01 02:09:53
6 1 2020-09-01 02:09:57
7 2 2020-09-01 18:14:35
8 2 2020-09-01 18:14:39
9 2 2020-09-01 18:14:40
10 2 2020-09-01 02:09:22
11 2 2020-09-01 02:09:35
12 2 2020-09-01 02:09:53
13 2 2020-09-01 02:09:57
I would like to get the average expanding session time. A session is defined as a sequence of events that is terminated by more than 5 minutes break.
I've grouped the sessions like so:
df.groupby(['id', pd.Grouper(key="timestamp", freq='5min', origin='start')])
And got the right groups:
id timestamp
3 1 2020-09-01 02:09:22
4 1 2020-09-01 02:09:35
5 1 2020-09-01 02:09:53
6 1 2020-09-01 02:09:57
id timestamp
0 1 2020-09-01 18:14:35
1 1 2020-09-01 18:14:39
2 1 2020-09-01 18:14:40
id timestamp
10 2 2020-09-01 02:09:22
11 2 2020-09-01 02:09:35
12 2 2020-09-01 02:09:53
13 2 2020-09-01 02:09:57
id timestamp
7 2 2020-09-01 18:14:35
8 2 2020-09-01 18:14:39
9 2 2020-09-01 18:14:40
Now I would like to calculate the average session time in seconds per user at any given row, so the output is:
id timestamp avg_session_time
0 1 2020-09-01 18:14:35 0 <-- first event
1 1 2020-09-01 18:14:39 4 <-- 2nd event after 4 seconds
2 1 2020-09-01 18:14:40 5 <-- 3rd event after 5 seconds
--- session end
3 1 2020-09-01 02:09:22 5 <-- first event of second session
4 1 2020-09-01 02:09:35 9 <-- 2nd event after 13 seconds (13 seconds in the 2nd session + 5 in first session divide by the number of sessions 2)
5 1 2020-09-01 02:09:53 18 <-- 3rd event after 31 seconds ((31 + 5) / 2 = 18)
6 1 2020-09-01 02:09:57 20 <-- 4th event after 35 seconds ((35 + 5) / 2 = 20)
---
7 2 2020-09-01 18:14:35 0
8 2 2020-09-01 18:14:39 4
9 2 2020-09-01 18:14:40 5
---
10 2 2020-09-01 02:09:22 5
11 2 2020-09-01 02:09:35 9
12 2 2020-09-01 02:09:53 18
13 2 2020-09-01 02:09:57 20
Any help would be awesome :)
回答1:
Use:
#converting to datetimes
df['timestamp'] = pd.to_datetime(df['timestamp'])
#grouping per 5Min and id
g = df.groupby(['id', pd.Grouper(key="timestamp", freq='5min', origin='start')])
#get first values per groups to new column
df['diff'] = g['timestamp'].transform('first')
#subtract by timestamp and convert timedeltas to seconds
df['diff'] = df['timestamp'].sub(df['diff']).dt.total_seconds()
#shifting per groups by id
df['new'] = df.groupby('id')['diff'].shift()
#get first value per groups, now shifted
df['new'] = g['new'].transform('first')
#replace 0 to misisng values and get average
df['last'] = df[['new','diff']].replace(0, np.nan).mean(axis=1).fillna(df['new'])
print (df)
id timestamp diff new last
0 1 2020-09-01 18:14:35 0.0 0.0 0.0
1 1 2020-09-01 18:14:39 4.0 0.0 4.0
2 1 2020-09-01 18:14:40 5.0 0.0 5.0
3 1 2020-09-01 02:09:22 0.0 5.0 5.0
4 1 2020-09-01 02:09:35 13.0 5.0 9.0
5 1 2020-09-01 02:09:53 31.0 5.0 18.0
6 1 2020-09-01 02:09:57 35.0 5.0 20.0
7 2 2020-09-01 18:14:35 0.0 0.0 0.0
8 2 2020-09-01 18:14:39 4.0 0.0 4.0
9 2 2020-09-01 18:14:40 5.0 0.0 5.0
10 2 2020-09-01 02:09:22 0.0 5.0 5.0
11 2 2020-09-01 02:09:35 13.0 5.0 9.0
12 2 2020-09-01 02:09:53 31.0 5.0 18.0
13 2 2020-09-01 02:09:57 35.0 5.0 20.0
来源：https://stackoverflow.com/questions/65561016/pandas-expanding-average-session-time