BUG: DataFrame.mad(axis=1) not working for nullable integer dtype

[jorisvandenbossche]

In [32]: df = pd.DataFrame(np.random.randn(100000, 4).astype(int)).astype("Int64")  

In [33]: df.mad()  
Out[33]: 
0    0.363361
1    0.365456
2    0.369014
3    0.368195
dtype: float64

In [34]: df.mad(axis=1) 
---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
<ipython-input-34-ebf68ac71360> in <module>
----> 1 df.mad(axis=1)

~/scipy/pandas/pandas/core/generic.py in mad(self, axis, skipna, level)
  10042                 demeaned = data - data.mean(axis=0)
  10043             else:
> 10044                 demeaned = data.sub(data.mean(axis=1), axis=0)
  10045             return np.abs(demeaned).mean(axis=axis, skipna=skipna)
  10046 

~/scipy/pandas/pandas/core/ops/__init__.py in f(self, other, axis, level, fill_value)
    751             axis = self._get_axis_number(axis) if axis is not None else 1
    752             return _combine_series_frame(
--> 753                 self, other, pass_op, axis=axis, str_rep=str_rep
    754             )
    755         else:

~/scipy/pandas/pandas/core/ops/__init__.py in _combine_series_frame(left, right, func, axis, str_rep)
    551 
    552             array_op = get_array_op(func, str_rep=str_rep)
--> 553             bm = left._data.apply(array_op, right=values.T)
    554             return type(left)(bm)
    555 

~/scipy/pandas/pandas/core/internals/managers.py in apply(self, f, filter, align_keys, **kwargs)
    431 
    432             if callable(f):
--> 433                 applied = b.apply(f, **kwargs)
    434             else:
    435                 applied = getattr(b, f)(**kwargs)

~/scipy/pandas/pandas/core/internals/blocks.py in apply(self, func, **kwargs)
    365         """
    366         with np.errstate(all="ignore"):
--> 367             result = func(self.values, **kwargs)
    368 
    369         return self._split_op_result(result)

~/scipy/pandas/pandas/core/ops/array_ops.py in arithmetic_op(left, right, op, str_rep)
    203     if should_extension_dispatch(lvalues, rvalues) or isinstance(rvalues, Timedelta):
    204         # Timedelta is included because numexpr will fail on it, see GH#31457
--> 205         res_values = op(lvalues, rvalues)
    206 
    207     else:

~/scipy/pandas/pandas/core/ops/common.py in new_method(self, other)
     61         other = item_from_zerodim(other)
     62 
---> 63         return method(self, other)
     64 
     65     return new_method

~/scipy/pandas/pandas/core/arrays/integer.py in integer_arithmetic_method(self, other)
    618 
    619             if getattr(other, "ndim", 0) > 1:
--> 620                 raise NotImplementedError("can only perform ops with 1-d structures")
    621 
    622             if isinstance(other, IntegerArray):

NotImplementedError: can only perform ops with 1-d structures

[jreback]

thought we deprecated .mad

[jorisvandenbossche]

You mentioned that recently here as well: #31742 (comment), but I am not aware of any issues/discussion about that

[jreback]

#11787

[jbrockmendel]

This is now working on master, though #33600 is breaking it for the axis=0 case.

[notwopr]

Hi. I've used Pandas 1.1.3 thru 1.1.5, but all versions have produced the same problem...pandas.DataFrame.mad(axis=1) producing NaNs on perfectly fine data. The dataframe I have has several columns of floats. When I run my script to create a new column using any other command like .std(axis=1), .max(axis=1), .min(axis=1), .mean(axis=1), and .median(axis=1), they all produce correct numbers.
But when it calculates for .mad(axis=1), it produces a blank column. No error messages are produced. It's just blank.

[jbrockmendel]

This is working on master, could use tests.