nth function/method for groupby

[samukweku]

An nth function, similar to pandas groupby nth, where the selected row/rows stay as part of the new aggregations in the presence of by:

from datatable import dt, f, by

data = {'x': [1, 1, 1, 2, 2, 3],
        'y': [1, 2, 3, 1, 2, 1],
        'n': [3, 2, 1, 1, 2, 1]}

DT = dt.Frame(data)

DT

   |     x      y      n
   | int32  int32  int32
-- + -----  -----  -----
 0 |     1      1      3
 1 |     1      2      2
 2 |     1      3      1
 3 |     2      1      1
 4 |     2      2      2
 5 |     3      1      1
[6 rows x 3 columns]

The nth function/method should return the row or rows along with other aggregations:

DT[:, {'sum': f.y.sum(), 'nth' : f.y.nth(1)}, 'x']

    |   x    sum      nth
    | int32  int64  int32
 -- + -----  -----  -----
  0 |     1      6      2
  1 |     2      3      2
  2 |     3      1     NA
 [3 rows x 3 columns]

One way to currently implement this would be to do a cbind:

dt.cbind(DT[:, {"sum": f.y.sum()}, "x"], DT[1, 'n', by('x', add_columns=False)], force=True)

which might not properly align; a much safer option would be a left join.

The implementation in #2176 selects a single row or slice; and since within datatable i is executed before j, the results wont come out right. Also, at the moment, sequences are not accepted within the i; it would be nice to be able to select lists with the nth function

[st-pasha]

So, this is supposed to be similar to first() or last(), but instead it selects the "n-th" row?

[samukweku]

@st-pasha , yes. It can be any row, and if the row is missing, then NA is returned

[samukweku]

WIP