REGR: DataFrame.mean(numeric_only=True) raises AttributeError on v1.0.3

[simonjayhawkins]

Code Sample, a copy-pastable example if possible

>>> import numpy as np
>>>
>>> import pandas as pd
>>>
>>> pd.__version__
'1.0.3'
>>>
>>> df_wide = pd.DataFrame(np.random.randint(1000, size=(1000, 100))).astype("Int64").copy()
>>>
>>> df_wide.mean(numeric_only=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\simon\pandas\pandas\core\generic.py", line 11215, in stat_func
    f, name, axis=axis, skipna=skipna, numeric_only=numeric_only
  File "C:\Users\simon\pandas\pandas\core\frame.py", line 7896, in _reduce
    res = df._data.reduce(op, axis=1, skipna=skipna, **kwds)
  File "C:\Users\simon\pandas\pandas\core\internals\managers.py", line 351, in reduce
    bres = func(blk.values, *args, **kwargs)
  File "C:\Users\simon\pandas\pandas\core\nanops.py", line 69, in _f
    return f(*args, **kwargs)
  File "C:\Users\simon\pandas\pandas\core\nanops.py", line 102, in f
    if values.size == 0 and kwds.get("min_count") is None:
AttributeError: 'IntegerArray' object has no attribute 'size'
>>>

Problem description

This is a regression from 0.25.3

0aa48f7 is the first bad commit
commit 0aa48f7
Author: jbrockmendel jbrockmendel@gmail.com
Date: Wed Jan 1 09:18:20 2020 -0800

PERF: perform reductions block-wise (#29847)

on master raises AttributeError: 'int' object has no attribute 'dtype'

>>> df_wide.mean(numeric_only=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\simon\pandas\pandas\core\generic.py", line 11114, in stat_func
    func, name=name, axis=axis, skipna=skipna, numeric_only=numeric_only
  File "C:\Users\simon\pandas\pandas\core\frame.py", line 7990, in _reduce
    res = df._data.reduce(blk_func)
  File "C:\Users\simon\pandas\pandas\core\internals\managers.py", line 362, in reduce
    bres = func(blk.values, *args, **kwargs)
  File "C:\Users\simon\pandas\pandas\core\frame.py", line 7985, in blk_func
    return op(values, axis=0, skipna=skipna, **kwds)
  File "C:\Users\simon\pandas\pandas\core\nanops.py", line 120, in f
    result = bn_func(values, axis=axis, **kwds)
  File "<__array_function__ internals>", line 6, in nanmean
  File "C:\Users\simon\Anaconda3\envs\pandas-dev\lib\site-packages\numpy\lib\nanfunctions.py", line 952, in nanmean
    avg = _divide_by_count(tot, cnt, out=out)
  File "C:\Users\simon\Anaconda3\envs\pandas-dev\lib\site-packages\numpy\lib\nanfunctions.py", line 219, in _divide_by_count
    return a.dtype.type(a / b)
AttributeError: 'int' object has no attribute 'dtype'

Expected Output

>>> import numpy as np
>>>
>>> import pandas as pd
>>>
>>> pd.__version__
'0.25.3'
>>>
>>> df_wide = pd.DataFrame(np.random.randint(1000, size=(1000, 100))).astype("Int64").copy()
>>>
>>> df_wide.mean(numeric_only=True)
0     520.057
1     507.735
2     501.618
3     506.590
4     501.500
       ...
95    507.594
96    483.273
97    506.330
98    497.068
99    508.118
Length: 100, dtype: float64
>>>

[jorisvandenbossche]

I noticed this as well while working on #32867. And normally this should also get solved in that PR.

[simonjayhawkins]

I was just looking at that PR, and couldn't find an issue. then realising it was a regression, felt that a separate issue should be raised. If we can get just the fix part of #32867 broken off, could backport if we do another patch release. (sorry, should have xrefed your work here originally)

[jorisvandenbossche]

No, problem, you're totally correct to open an issue about it.
I think it should be rather straightforward to separate a fix from that PR.

[jbrockmendel]

@simonjayhawkins can you @ me on these for issues that i caused

[simonjayhawkins]

changing the milestone, xref #33300 to track.