special.softmax() should support pd.DataFrame as input

Is your feature request related to a problem? Please describe. when the input df is of type pandas.core.frame.DataFrame, softmax(df) doesn’t work:

gives this error

C:\ProgramData\Anaconda3\lib\site-packages\pandas\core\ops\__init__.py in _align_method_FRAME(left, right, axis)
    662             else:
    663                 raise ValueError(
--> 664                     "Unable to coerce to DataFrame, shape "
    665                     f"must be {left.shape}: given {right.shape}"
    666                 )

Describe the solution you’d like should be able to run softmax(df)

Describe alternatives you’ve considered currently have to either run:

softmax(df.values)

or

softmax(np.array(df))

Additional context (e.g. screenshots)

[josef-pkt]

FWIW, I agree that a pandas or xarray or … data structures shouldn’t be returned. Especially, scipy.special wouldn’t be the place to start to support it, IMO, because those are in general not typical datascience functions for end users.

pd.DataFrame(special.softmax(df), index=df.index, columns=df_columns) should work (after fixing asarray), (but maybe you want to change column names.)

[josef-pkt]

I don’t think that’s special casing pandas.

The function is missing asarray to support general “array_like”