How to generate cumulative unique count at a group level in python?

Question

I have some hospital visit healthcare data in a dataframe of the form:

record_id	client_id	date_of_encounter	hospital_id
1	MK456	2014-01-01	01J
2	JJ103	2016-04-01	02J
3	MK456	2014-02-26	01J
4	JJ103	2016-05-01	02H
5	MK456	2014-03-01	02H
6	JJ103	2016-06-06	02J

I want to create a column hospital_count which is a cumulative count of the UNIQUE hospitals visits by each client at the date_of_counter. I have already sorted is by the client_id and
date_of_counter. The result transformation would be

record_id	client_id	date_of_encounter	hospital_id	hospital_count
1	MK456	2014-01-01	01J	1
3	MK456	2014-02-26	01J	1
5	MK456	2014-03-01	02H	2
2	JJ103	2016-04-01	02J	1
4	JJ103	2016-05-01	02H	2
6	JJ103	2016-06-06	02J	2

Some suggest using a combination of a groupby and cumsum() but I am not too sure how?

Answer 1

Using GoupBy.cumcount

Cumulative count of the number of distinct hospitals visited by each client

import pandas as pd

df = pd.DataFrame({
  'record_id': list(range(1,7)),
  'client_id':['MK', 'JJ', 'MK', 'JJ', 'MK', 'JJ'],
  'date': [20140101, 20160401,20140226,20160501,20140301,20160606],
  'hospital': ['1j', '2j', '1j', '2h', '2h', '2j']
})

df.sort_values(by=['client_id', 'date'], inplace=True)

df['hospital_count'] = df.drop_duplicates(subset=['client_id', 'hospital']
  ).groupby('client_id').cumcount() + 1

df.fillna(method='ffill', inplace=True)

print(df)
#    record_id client_id      date hospital  hospital_count
# 1          2        JJ  20160401       2j             1.0
# 3          4        JJ  20160501       2h             2.0
# 5          6        JJ  20160606       2j             2.0
# 0          1        MK  20140101       1j             1.0
# 2          3        MK  20140226       1j             1.0
# 4          5        MK  20140301       2h             2.0

Explanation: We drop successive visits of the same client to the same hospital using drop_duplicates; then we can simply count the visits of each client using groupby and cumcount. However, this leaves NaN values in the lines that were dropped; we fill those values using fillna.

Cumulative count of the number of visits of each client to each hospital

import pandas as pd

df = pd.DataFrame({
  'record_id': list(range(1,7)),
  'client_id':['MK', 'JJ', 'MK', 'JJ', 'MK', 'JJ'],
  'date': [20140101, 20160401,20140226,20160501,20140301,20160606],
  'hospital': ['1j', '2j', '1j', '2h', '2h', '2j']
})

df['hospital_count'] = df.sort_values(by=['client_id', 'hospital', 'date']
  ).groupby(['client_id', 'hospital']
  ).cumcount() + 1

print(df)
#    record_id client_id      date hospital  hospital_count
# 0          1        MK  20140101       1j               1
# 1          2        JJ  20160401       2j               1
# 2          3        MK  20140226       1j               2
# 3          4        JJ  20160501       2h               1
# 4          5        MK  20140301       2h               1
# 5          6        JJ  20160606       2j               2