BUG: DataFrameGroupBy.transform with axis=1 fails

[rhshadrach]

The following seems to happen on all transform groupby kernels:

df = DataFrame({"A": [1, 2, 3], "B": [4, 5, 6]})
df.groupby([1, 1], axis=1).transform("shift")

results in ValueError: Length mismatch: Expected axis has 2 elements, new values have 3 elements.

I think the issue is in pandas.core.groupby.generic._wrap_transformed_output; this method does not take into account self.axis when wrapping the output. All that needs to happen is result = result.T and to use the index labels rather than the column labels for columns there.

[ethanator]

Could I work on this issue?

[xxsacxx]

I tried the same example with :
df.groupby(['val']*df.shape[1],axis=1).transform('shift')
which failed with
Shape of passed values is (2, 3), indices imply (3, 3)

while
df.groupby(['val']*df.shape[1],axis=1).transform(lambda x:x.shift())

gives :
A B
0 NaN 1.0
1 NaN 2.0
2 NaN 3.0

what output are we expecting here ?

[rhshadrach]

@ethanator Absolutely! Simply make a comment here "take" and github will assign it to you. If you encounter any difficulties, feel free to reach out here.

@xxsacxx Your 2nd result looks correct to me; the values are shifted 1 to the right, and since there are no values to the left of the first column, you get NaN. The presence of NaN then coerces the dtype to being a float.

[ethanator]

take