Python pandas group by check if value changed then previous value-CodePudding

I've a problem with groupby function of pandas's library. I've the following dataframe.

d = {'id': [400001, 400001, 400001, 400002, 400003, 400003, 400004, 400004], 'result': ['P','N','N','N','N','N','N','P'], 'date':['27/10/2021','09/09/2021','03/07/2020','03/07/2020','30/06/2020','27/04/2022','27/04/2022','30/06/2020']}
df = pd.DataFrame(data=d)
df

id	result	date
400001	N	2020-07-03
400001	N	2021-09-09
400001	P	2021-10-27
400002	N	2020-07-03
400003	N	2020-06-30
400003	N	2022-04-27
400004	P	2020-06-30
400004	N	2022-04-27

I need to group by column 'id' and extract the value of column 'date' where the value of column 'result' change. If value in column 'result' doesn't change, keep the first value of column 'date'.

This an example:

id	date
400001	2021-10-27
400002	2020-07-03
400003	2020-06-30
400004	2022-04-27

I've tried this:

df['change'] = np.where(df.groupby('id').esito.apply(lambda x:x!=x.iloc[0]),'Y','N')

but the function doesn't works so well. The function verify the difference versus first element of column 'id' of groupby selection. I don't need this.

Can you help me? Thanks

CodePudding user response：

You can use drop duplicates twice to get what you want:

import pandas as pd

d = {'id': [400001, 400001, 400001, 400002, 400003, 400003, 400004, 400004],
     'result': ['N', 'N', 'P', 'N', 'N', 'N', 'P', 'N'],
     'date': ['27/10/2021', '09/09/2021', '03/07/2020', '03/07/2020', '30/06/2020', '27/04/2022', '27/04/2022',
              '30/06/2020']}
df = pd.DataFrame(data=d)

df.drop_duplicates(subset=['id', 'result'], keep='first', inplace=True)
df.drop_duplicates(subset=['id'], keep='last', inplace=True)

print(df)

Output:

       id result        date
2  400001      P  03/07/2020
3  400002      N  03/07/2020
4  400003      N  30/06/2020
7  400004      N  30/06/2020

Notice that the output in your question may be wrong based on your requirements.

CodePudding user response：

You can compute a cumsum of the booleans identifying the changes. Then get the max index:

idx = (df.groupby('id')['result']
         .apply(lambda s: s.ne(s.shift())
                .cumsum()
                .idxmax()
               )
       )
df.loc[idx]

Output:

       id result        date
1  400001      N  09/09/2021
3  400002      N  03/07/2020
4  400003      N  30/06/2020
7  400004      P  30/06/2020

NB. The input provided as DataFrame is different from the one as table. The output matching the DataFrame is shown here.

If needed, sort the dates first:

idx = (df.sort_values(by=['id', 'date'])
         .groupby('id')['result']
         .apply(lambda s: s.ne(s.shift())
                .cumsum()
                .idxmax()
               )
       )
df.loc[idx]

Output:

       id result        date
0  400001      P  27/10/2021
3  400002      N  03/07/2020
5  400003      N  27/04/2022
7  400004      P  30/06/2020