BUG/QST: Series.transform with a dictionary

[rhshadrach]

What is the expected output of passing a dictionary to Series.transform? For example:

s = pd.Series([1, 2, 3])
result1 = s.transform({'a': lambda x: x + 1})
result2 = s.transform({'a': lambda x: x + 1, 'b': lambda x: x + 2})

The docs say that dict of axis labels -> functions is acceptable, but I can't find any example in the docs where the output is described/shown. Under the hood, Series.transform is just calling Series.aggregate which produces the following outputs for result1 and result2.

# result1
a  0    2
   1    3
   2    4
dtype: int64

# result2
a  0    2
   1    3
   2    4
b  0    3
   1    4
   2    5
dtype: int64

result1 is deemed acceptable (the length of the result equals the length of the input) and is returned, but result2 raises; it is not a transformation.

I am wondering if a better return would be a DataFrame where the keys are the column names ('a' and 'b' in this example).

[onshek]

I find this is a bug actually, the docs for pd.Series.transform is the same with pd.DataFrame.transform as

    @doc(
        NDFrame.transform,
        klass=_shared_doc_kwargs["klass"],
        axis=_shared_doc_kwargs["axis"],
    )

And I get this:

In [3]: s = pd.Series([1, 2, 3]) 
   ...: result1 = s.transform({'a': lambda x: x + 1}) 
   ...: result2 = s.transform({'a': lambda x: x + 1, 'b': lambda x: x + 2})        
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-3-32dce75aad7a> in <module>
      1 s = pd.Series([1, 2, 3])
      2 result1 = s.transform({'a': lambda x: x + 1})
----> 3 result2 = s.transform({'a': lambda x: x + 1, 'b': lambda x: x + 2})

~/Applications/anaconda3/lib/python3.7/site-packages/pandas/core/series.py in transform(self, func, axis, *args, **kwargs)
   3910         # Validate the axis parameter
   3911         self._get_axis_number(axis)
-> 3912         return super().transform(func, *args, **kwargs)
   3913 
   3914     def apply(self, func, convert_dtype=True, args=(), **kwds):

~/Applications/anaconda3/lib/python3.7/site-packages/pandas/core/generic.py in transform(self, func, *args, **kwargs)
  10812         result = self.agg(func, *args, **kwargs)
  10813         if is_scalar(result) or len(result) != len(self):
> 10814             raise ValueError("transforms cannot produce " "aggregated results")
  10815 
  10816         return result

ValueError: transforms cannot produce aggregated results

[rhshadrach]

@onshek - yes, agreed. The question then is, what should be the output? A dataframe with columns being the keys of the dictionary seems reasonable to me, but I can't find any documentation saying/showing this. I think that fits in well with DataFrame.transform where the keys subset the columns (and, thus, are the column names in the results)

[onshek]

@rhshadrach I think the main point is axis:

In [11]: s = pd.Series(range(3))

In [13]: s.transform([np.sqrt, np.exp], axis=0)
Out[13]:
       sqrt       exp
0  0.000000  1.000000
1  1.000000  2.718282
2  1.414214  7.389056

And s.transform([np.sqrt, np.exp], axis=1) can be designed to (but now it fails)

        0  0.000000  
sqrt    1  1.000000  
        2  1.414214  
        0  1.000000  
exp     1  2.718282
        2  7.389056