BUG: DataFrame/Series.transform with list and non-list dict values

[rhshadrach]

df = pd.DataFrame({'a': [1, 2], 'b': [1, 4], 'c': [1, 4]})
print(df.transform({'b': ['sqrt', 'abs'], 'c': 'sqrt'}))
print(df['a'].transform({'b': ['sqrt', 'abs'], 'c': 'sqrt'}))

produces

     b        c
  sqrt abs    c
0  1.0   1  1.0
1  2.0   4  2.0
          b             c
       sqrt abs         a
0  1.000000   1  1.000000
1  1.414214   2  1.414214

In each case the last column label on the 2nd level should be "sqrt" and not "a" or "c".

[smohr003]

In which version are you getting this? I am running them on pandas==1.1.5 and the first case is producing the correct output.
The second case errors out as

`---------------------------------------------------------------------------
SpecificationError Traceback (most recent call last)
in
----> 1 print(df['a'].transform({'b': ['sqrt', 'abs'], 'c': 'sqrt'}))

/opt/conda/lib/python3.8/site-packages/pandas/core/series.py in transform(self, func, axis, *args, **kwargs)
4080 # Validate the axis parameter
4081 self._get_axis_number(axis)
-> 4082 return super().transform(func, *args, **kwargs)
4083
4084 def apply(self, func, convert_dtype=True, args=(), **kwds):

/opt/conda/lib/python3.8/site-packages/pandas/core/generic.py in transform(self, func, *args, **kwargs)
10674 2 1.414214 7.389056
10675 """

10676 result = self.agg(func, *args, **kwargs)
10677 if is_scalar(result) or len(result) != len(self):
10678 raise ValueError("transforms cannot produce aggregated results")

/opt/conda/lib/python3.8/site-packages/pandas/core/series.py in aggregate(self, func, axis, *args, **kwargs)
4047 func = dict(kwargs.items())
4048
-> 4049 result, how = self._aggregate(func, *args, **kwargs)
4050 if result is None:
4051

/opt/conda/lib/python3.8/site-packages/pandas/core/base.py in _aggregate(self, arg, *args, **kwargs)
338 raise SpecificationError("nested renamer is not supported")
339 elif isinstance(obj, ABCSeries):
--> 340 raise SpecificationError("nested renamer is not supported")
341 elif isinstance(obj, ABCDataFrame) and k not in obj.columns:
342 raise KeyError(f"Column '{k}' does not exist!")

SpecificationError: nested renamer is not supported
`

[rhshadrach]

Thanks @smohr003! I've verified your reported behavior on 1.1.x and that the code matches my reported behavior on 1.2.x (and master). The first output I would classify as a regression, but the 2nd behavior (using on a Series) is the correct one. That is, passing a dictionary to Series.transform should not raise.

[simonjayhawkins]

from #40090 (comment)

As identified in #40018, this is a regression from 1.1 -> 1.2 (almost certainly from one of my PRs, but I haven't checked which).

first bad commit: [ab5b38d] BUG/CLN: Decouple Series/DataFrame.transform (#35964)