Pandas dataframe: find closest time before timestamp

January 5, 2024

I have two Pandas dataframes with "datetime" column.

For the df dataframe I would like to add a column with the seconds-difference with the nearest dflogdf["datetime"] BEFORE the df["datetime"]. This means I cannot use merge_asof direction="nearest".

For example
df["datetime"]:

2023-11-15T18:00:00
2023-11-20T19:00:00
2023-11-20T20:00:00
2023-11-20T21:00:00

dflogs["datetime"]:

2023-11-17T18:00:00
2023-11-20T20:00:00

Expected output:

2023-11-15T18:00:00  None   (Nothing before)
2023-11-20T19:00:00  262800 (2023-11-17T18:00:00)
2023-11-20T20:00:00  0      (2023-11-20T20:00:00)
2023-11-20T21:00:00  3600   (2023-11-20T20:00:00)

I was thinking around a function like this (not working correctly):

def check_time_diff(item):
    item["timediff"] = (item["datetime"] - dflogs['datetime']).min() / pd.Timedelta(seconds=1)
    return item
df = df.apply(check_time_diff, axis=1)

>Solution :

This is a case for merge_asof with direction backward

out = pd.merge_asof(df[['datetime']], 
              dflogs[['datetime']].assign(logtime=dflogs['datetime']),
              on='datetime', direction='backward')
out['diff'] = out['datetime'].sub(out['logtime']).dt.total_seconds()

Output:

             datetime             logtime      diff
0 2023-11-15 18:00:00                 NaT       NaN
1 2023-11-20 19:00:00 2023-11-17 18:00:00  262800.0
2 2023-11-20 20:00:00 2023-11-20 20:00:00       0.0
3 2023-11-20 21:00:00 2023-11-20 20:00:00    3600.0