How to group pandas dataframe by custom overlapping periods?

Question

Say I have a list of lists where each nested list has two values: a start date of a range and an end date of a range. So something like this:

ranges_list = [
    ['2020-03-12', '2020-06-12'],
    ['2020-03-13', '2020-06-13'],
    ['2020-03-14', '2020-06-14']
]

This represents 3 ranges:

March 12th 2020 - June 12th 2020
March 13th 2020 - June 13th 2020
March 14th 2020 - June 14th 2020

Say I also have some dataframe d which has multiple columns, one of which is a column called 'occurence_date' that contains datetimes in it.

Say the dataframe d looks like:

ID      LinkID   PC    occurence_date
10R46   R*1005   8017  2020-03-12
10R46   R*10335  5019  2020-04-15
100R91  R*1005   8017  2020-04-15
10R91   R*243    8870  2020-06-14

I want to group dataframe d using the occurence_date column on the ranges specified in ranges_list

So something like:

grouped = d.groupby('occurence_date', ranges=ranges_list)

Obviously, this groupby code is incorrect but helps serve what i want to do.

At the end of it all, the grouped object should have 3 distinct groups that look like:

group: ('2020-03-12', '2020-06-12')
ID      LinkID   PC    occurence_date
10R46   R*1005   8017  2020-03-12
10R46   R*10335  5019  2020-04-15
100R91  R*1005   8017  2020-04-15

group: ('2020-03-13', '2020-06-13')
ID      LinkID   PC    occurence_date
10R46   R*10335  5019  2020-04-15
100R91  R*1005   8017  2020-04-15

group: ('2020-03-14', '2020-06-14')
ID      LinkID   PC    occurence_date
10R46   R*10335  5019  2020-04-15
100R91  R*1005   8017  2020-04-15
10R91   R*243    8870  2020-06-14

How can I go about accomplishing this?

score 1 · Accepted Answer · answered Mar 14 '21 at 01:36

You can group by pd.IntervalIndex:

ranges_list = [
    (pd.Timestamp('2020-03-12'), pd.Timestamp('2020-06-12')),
    (pd.Timestamp('2020-03-13'), pd.Timestamp('2020-06-13')),
    (pd.Timestamp('2020-03-14'), pd.Timestamp('2020-06-14'))
]

idx = pd.IntervalIndex.from_tuples(ranges_list, closed='both')

def in_ranges(x, bins):
    rv = []
    for b in bins:
        if x in b:
            rv.append(b)
    return rv

df['groups'] = df['occurence_date'].apply(lambda x: in_ranges(x, idx))

for g in df.explode('groups').groupby('groups'):
    print(g[0])
    print('-' * 80)
    print(g[1][['ID', 'LinkID', 'PC', 'occurence_date']])
    print()

Prints:

[2020-03-12, 2020-06-12]
--------------------------------------------------------------------------------
       ID   LinkID    PC occurence_date
0   10R46   R*1005  8017     2020-03-12
1   10R46  R*10335  5019     2020-04-15
2  100R91   R*1005  8017     2020-04-15

[2020-03-13, 2020-06-13]
--------------------------------------------------------------------------------
       ID   LinkID    PC occurence_date
1   10R46  R*10335  5019     2020-04-15
2  100R91   R*1005  8017     2020-04-15

[2020-03-14, 2020-06-14]
--------------------------------------------------------------------------------
       ID   LinkID    PC occurence_date
1   10R46  R*10335  5019     2020-04-15
2  100R91   R*1005  8017     2020-04-15
3   10R91    R*243  8870     2020-06-14

score 1 · Answer 2 · answered Mar 14 '21 at 01:37

The following interactive session shows how to get to a point where you have the necessary data to group records together as you want it. There may be a more efficient way as this will iterate over len(d) * len(dranges) but if you do not have a lot of data, this is a straightforward solution.

>>> d
       ID   LinkID    PC occurence_date
0   10R46   R*1005  8017     2020-03-12
1   10R46  R*10335  5019     2020-04-15
2  100R91   R*1005  8017     2020-04-15
3   10R91    R*243  8870     2020-06-14

>>> dranges
            0           1
0  2020-03-12  2020-06-12
1  2020-03-13  2020-06-13
2  2020-03-14  2020-06-14

>>> d['overlaps'] = d.apply(lambda row: [f'{dr[0]} to {dr[1]}' 
                                         for _, dr in dranges.iterrows() 
                                         if row['occurence_date'] >= dr[0] 
                                         and row['occurence_date'] <= dr[1]]
                           , axis=1)

>>> d.explode('overlaps').sort_values('overlaps')
       ID   LinkID    PC occurence_date                  overlaps
0   10R46   R*1005  8017     2020-03-12  2020-03-12 to 2020-06-12
1   10R46  R*10335  5019     2020-04-15  2020-03-12 to 2020-06-12
2  100R91   R*1005  8017     2020-04-15  2020-03-12 to 2020-06-12
1   10R46  R*10335  5019     2020-04-15  2020-03-13 to 2020-06-13
2  100R91   R*1005  8017     2020-04-15  2020-03-13 to 2020-06-13
1   10R46  R*10335  5019     2020-04-15  2020-03-14 to 2020-06-14
2  100R91   R*1005  8017     2020-04-15  2020-03-14 to 2020-06-14
3   10R91    R*243  8870     2020-06-14  2020-03-14 to 2020-06-14

How to group pandas dataframe by custom overlapping periods?

2 Answers2