How to get grouped cumulative duration in pandas?

I have the following data:

id	encounter_key	datetime
1	111	2019-04-14
1	111	2019-04-14
1	111	2019-07-18
1	122	2019-09-02
2	211	2019-10-03
2	211	2020-10-03

I want to find the cumulative duration, grouped by id and encounter_key to achieve the following:

id	encounter_key	datetime	cum_duration_days
1	111	2019-04-14	0
1	111	2019-04-14	0
1	111	2019-07-18	95
1	122	2019-09-02	0
2	211	2019-10-03	0
2	211	2020-10-03	366

I’ve tried df.groupby(['datetime']).apply( ... ) etc. but nothing seems to work. Thanks in advance.

Answer

I think this should work

df['cum_duration_days']=df.groupby(['id','encounter_key'])['datetime'].diff()/ np.timedelta64(1, 'D')
df['cum_duration_days'].fillna(0)

But @enke is right, the output desired seems to have an error on the ’31’ unless you are not showing all the rows for encounter_key=211 /id=2 …