CODEWITHSUNDEEP
pythonpandastimedelta
paranormaldist
paranormaldist

Reputation: 508

Python transform('sum') on timedelta data

I would like to do a transform on data that includes timedelta format.

My data looks like this with Time column as timedelta type.

    user                in               out location  overlap    Time
0    ron  12/21/2021 10:11  12/21/2016 17:50     home     0  4:19:03
1    ron  12/21/2016 13:26  12/21/2016 13:52   office     2  0:25:28
2  april   12/21/2016 8:12  12/21/2016 17:27   office     0  8:15:03
3  april  12/21/2016 18:54  12/21/2016 22:56   office     0  4:02:36
4   andy   12/21/2016 8:57  12/21/2016 12:15     home     0  2:59:40

Based on user and overlap, I would like to do a transform on 'Time'. I've done this:

groups = sample.groupby('user')['Time']
flag = sample.groupby('user')['overlap'].transform('max')
sample.loc[:,'time_new'] = np.select([flag.eq(0), flag.isin([1,2])], [groups.transform('sum'), groups.transform('max')])

But I get the following error: 

TypeError: Cannot cast scalar from dtype('<m8[ns]') to dtype('<m8') according to the rule 'same_kind'

How can I do the transform properly?

Upvotes: 1

Views: 62

Answers (1)

ALollz
ALollz

Reputation: 59579

Convert the timedelta to a float that counts the number of seconds, and do the math. Then if you want, convert back to a timedelta

groups = sample['Time'].dt.total_seconds().groupby(sample['user'])

flag = sample.groupby('user')['overlap'].transform('max')
sample.loc[:,'time_new'] = np.select([flag.eq(0), flag.isin([1,2])], 
                                     [groups.transform('sum'), groups.transform('max')]) 

sample['time_new'] = pd.to_timedelta(sample['time_new'], unit='s')

    user                in               out location  overlap     Time time_new
0    ron  12/21/2021 10:11  12/21/2016 17:50     home        0 04:19:03 04:19:03
1    ron  12/21/2016 13:26  12/21/2016 13:52   office        2 00:25:28 04:19:03
2  april   12/21/2016 8:12  12/21/2016 17:27   office        0 08:15:03 12:17:39
3  april  12/21/2016 18:54  12/21/2016 22:56   office        0 04:02:36 12:17:39
4   andy   12/21/2016 8:57  12/21/2016 12:15     home        0 02:59:40 02:59:40

Upvotes: 1

Related Questions