[FEA] Groupby cumulative count

[beckernick]

Is your feature request related to a problem? Please describe.
As a cuDF user, I want to assign numbers to each observation of a group reflecting its order of occurrence in the group.

The equivalent in the pandas API doc is here.

Describe the solution you'd like
I'd like to be able to call df.groupby(col).cumcount() and return a column of the same size containing the numberings described above.

[harrism]

Implementation is basically the same as groupby.cumsum. Will move to 0.8 along with #1298

[daxiongshu]

Is this the same function as the following code?

def get_order_in_group(y, order):
    for i in range(cuda.threadIdx.x, len(y), cuda.blockDim.x):
        order[i] = i

got = df.groupby(['y']).apply_grouped(get_order_in_group,incols=['y'],
                                  outcols={'order': 'int32'},
                                  tpb=self.tpb)

The full notebook link, cell 15.

[karthikeyann]

This operation requires output to be in same order as input at column level. is that right?
because order only within group is simply sequence from 0 to size(group).

[kkraus14]

This operation requires output to be in same order as input at column level. is that right?
because order only within group is simply sequence from 0 to size(group).

It only requires the output order within groups to be stable. The ordering of the grouping keys is not required.

[harrism]

@karthikeyann is this implemented as part of 7387?

[karthikeyann]

libcudf part is implemented.
Cython, Python code is not done yet.

[kkraus14]

@karthikeyann unassigned you since I figured someone else would do the Cython / Python code, but if you're tackling it please reassign yourself.

[karthikeyann]

I am already working on it.