Adding rows per group in pandas / ipython if per g

Question

I have a dataframe that contains for each group the number of observations during a certain period. Some groups don't contain all periods, and for these groups I want to append x rows with the missing periods in it. So that each group has a row for all 6 periods

My current df looks something like this:

> ID      PERIOD       VAlUE
  1       1            10
  1       2            8
  1       3            8  
  1       4            15
  1       5            6
  1       6            44
  2       1            NONE
  3       2            4
  3       5            25

I want a dataframe looking like this.

> ID      PERIOD       VAlUE
  1       1            10
  1       2            8
  1       3            8  
  1       4            15
  1       5            6
  1       6            44
  2       1            NONE
  2       2            NONE
  2       3            NONE
  2       4            NONE
  2       5            NONE
  2       6            4
  3       1            NONE
  3       2            4
  3       3            NONE
  3       4            NONE
  3       5            25
  3       6            NONE

So what happenend:

• For ID == 1, nothing happened, because it contained all 6 periods
• For ID == 2, it appended 5, rows for each period that it didn't have in the first df.
• For ID == 2, it appended 4, rows for each period that it didn't have in the first df. So it added rows for the periods 1,3,4 & 6.

I really don't have a clue how to do it, so help would really be appreciated.

Answer 1

You can set the index to 'ID' and 'PERIOD' and then construct a new index by generating the product of both columns and pass this as the new index to reindex, this has an optional fill_value param which you can set to the str NONE:

In [158]:
iterables = [df['ID'].unique(),df['PERIOD'].unique()]
df = df.set_index(['ID','PERIOD'])
df = df.reindex(index=pd.MultiIndex.from_product(iterables, names=['ID', 'PERIOD']), fill_value='NONE').reset_index()
df

Out[158]:
    ID  PERIOD VAlUE
0    1       1    10
1    1       2     8
2    1       3     8
3    1       4    15
4    1       5     6
5    1       6    44
6    2       1  NONE
7    2       2  NONE
8    2       3  NONE
9    2       4  NONE
10   2       5  NONE
11   2       6  NONE
12   3       1  NONE
13   3       2     4
14   3       3  NONE
15   3       4  NONE
16   3       5    25
17   3       6  NONE

So breaking the above down:

In [160]:
# create a list of the iterable index values we want to generate all product combinations from
iterables = [df['ID'].unique(),df['PERIOD'].unique()]
iterables

Out[160]:
[array([1, 2, 3], dtype=int64), array([1, 2, 3, 4, 5, 6], dtype=int64)]

In [163]:
# set the index to ID and PERIOD
df = df.set_index(['ID','PERIOD'])
df

Out[163]:
          VAlUE
ID PERIOD      
1  1         10
   2          8
   3          8
   4         15
   5          6
   6         44
2  1       NONE
3  2          4
   5         25

In [164]:
# reindex and pass the product from iterables as the new index
df.reindex(index=pd.MultiIndex.from_product(iterables, names=['ID', 'PERIOD']), fill_value='NONE').reset_index()
Out[164]:
    ID  PERIOD VAlUE
0    1       1    10
1    1       2     8
2    1       3     8
3    1       4    15
4    1       5     6
5    1       6    44
6    2       1  NONE
7    2       2  NONE
8    2       3  NONE
9    2       4  NONE
10   2       5  NONE
11   2       6  NONE
12   3       1  NONE
13   3       2     4
14   3       3  NONE
15   3       4  NONE
16   3       5    25
17   3       6  NONE

Answer 2

You can unstack the results on PERIOD and then stack them back with the dropna option set to False.

>>> df.groupby(['ID', 'PERIOD']).VAlUE.sum().unstack('PERIOD').stack('PERIOD', dropna=False)
ID  PERIOD
1   1          10
    2           8
    3           8
    4          15
    5           6
    6          44
2   1         NaN
    2         NaN
    3         NaN
    4         NaN
    5         NaN
    6         NaN
3   1         NaN
    2           4
    3         NaN
    4         NaN
    5          25
    6         NaN
dtype: object