Inconsistent behavior when using GroupBy and pandas.Series.mode

[jointfull]

Code Sample

import pandas

# Works great
df1 = pandas.DataFrame([[20,'A'],[20,'B'],[10,'C']])
gb1 = df1.groupby(0).agg(pandas.Series.mode)
display(gb1)

# Exception: Must produce aggregated value
df2 = pandas.DataFrame([[20,'A'],[20,'B'],[30,'C']])
gb2 = df2.groupby(0).agg(pandas.Series.mode)
display(gb2)

Problem Description

As it seems, the above code works great for df1, returning the following result:

         1
0         
10       C
20  [A, B]

(where C is a str, and [A, B] is a numpy.ndarray)

However, it doesn't work for df2, throwing the following exception:

...
C:\ProgramData\Anaconda3\envs\py36\lib\site-packages\pandas\core\groupby\generic.py in _aggregate_named(self, func, *args, **kwargs)
    907             output = func(group, *args, **kwargs)
    908             if isinstance(output, (Series, Index, np.ndarray)):
--> 909                 raise Exception('Must produce aggregated value')
    910             result[name] = self._try_cast(output, group)
    911 

Exception: Must produce aggregated value

It looks like the order of the processing of the agg function affects pandas error checking; if the first result (GroupBy row) returns an numpy.ndarray - the above exception is thrown, but if the first result return a str/scalar - the processing continues and suppresses further such exceptions.

In my opinion, item may be related to #2656 or one of its root causes ("fast apply vs. old pathway" as they put it), and to #24016 (which fails to accept numpy.ndarray as a return value). However, in our case - where pandas.Series.mode sometimes returns a scalar and sometimes a numpy.ndarray, it is more illusive and more confusing and therefore inconsistent (I spent ~2 hours debugging this trying to understand why so many of my agg function calls work but only one doesn't).

Expected Output

         1
0         
20  [A, B]
30       C

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.4.final.0
python-bits: 64
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 78 Stepping 3, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None
LOCALE: None.None

pandas: 0.24.1
pytest: None
pip: 19.0.1
setuptools: 40.4.3
Cython: None
numpy: 1.15.2
scipy: 1.0.0
pyarrow: None
xarray: None
IPython: 6.2.1
sphinx: None
patsy: 0.5.0
dateutil: 2.6.1
pytz: 2017.3
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 3.0.0
openpyxl: 2.5.12
xlrd: 1.2.0
xlwt: None
xlsxwriter: None
lxml.etree: None
bs4: None
html5lib: 1.0.1
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None
gcsfs: None

[WillAyd]

Yea I see where this is confusing but generally we don't support agg reducing to anything but a scalar, so I would think the first example returning something is pure coincidence and if anything the second example would be what we should be doing here.

Others may disagree though so let's see. FWIW apply is a more general function so you'd be safer to do something like:

df2.groupby(0)[1].apply(lambda x: list(x.mode()))

[j-musial]

As explained in the #19254 df2.groupby(0)[1].apply(lambda x: list(x.mode())) is really slow so it would be beneficial to add a separate groupby.mode() function implemented in the Cython. Such functionality is very often used for the categorical data so I am supprised that it still has not been implemented in Pandas.

[mroeschke]

This looks to work on master now (albeit agreed that agg should just be a reducer). Could use a test

In [9]: df2 = pandas.DataFrame([[20,'A'],[20,'B'],[30,'C']])
   ...: gb2 = df2.groupby(0).agg(pandas.Series.mode)

In [10]: gb2
Out[10]:
         1
0
20  [A, B]
30       C

[jointfull]

..so did someone fix this, or is it just another 'accidental' example of something that works, but given a different example it may fail again?

[jacobdadams]

I am still seeing this behavior in 1.41.1.

If the first groupby result returns a single mode, a subsequent groupby result that contains multiple modes will insert a list of the multiple modes. However, if the first groupby result would return multiple modes, it returns a ValueError: Must produce aggregated value similar to the original exception.

In [2]: test_df = pd.DataFrame({
           'value': [1, 1, 2, 3],
           'group': ['a', 'a', 'b', 'b'],
       })
In [4]: gb = test_df.groupby('group')
In [5]: mode_series = gb['value'].agg(pd.Series.mode)
In [6]: mode_series
Out[6]:
group
a         1
b    [2, 3]
Name: value, dtype: object

In [7]: test2_df = pd.DataFrame({
           'value': [1, 2, 3, 3,],
           'group': ['a', 'a', 'b', 'b'],
       })
In [8]: gb2 = test2_df.groupby('group')
In [9]: mode_series2 = gb2['value'].agg(pd.Series.mode)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Input In [9], in <cell line: 1>()
----> 1 mode_series2 = gb2['value'].agg(pd.Series.mode)
# <snip>
File ~\Miniconda3\envs\pandas\lib\site-packages\pandas\core\groupby\ops.py:1010, in BaseGrouper._aggregate_series_pure_python(self, obj, func)
   1006 res = libreduction.extract_result(res)
   1008 if not initialized:
   1009     # We only do this validation on the first iteration
-> 1010     libreduction.check_result_array(res, group.dtype)
   1011     initialized = True
   1013 counts[i] = group.shape[0]

File ~\Miniconda3\envs\pandas\lib\site-packages\pandas\_libs\reduction.pyx:13, in pandas._libs.reduction.check_result_array()

File ~\Miniconda3\envs\pandas\lib\site-packages\pandas\_libs\reduction.pyx:21, in pandas._libs.reduction.check_result_array()

ValueError: Must produce aggregated value

pd.show_versions()

INSTALLED VERSIONS

commit : 06d2301
python : 3.10.0.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.19044
machine : AMD64
processor : Intel64 Family 6 Model 158 Stepping 10, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : English_United States.1252

pandas : 1.4.1
numpy : 1.21.5
pytz : 2021.3
dateutil : 2.8.2
pip : 21.2.4
setuptools : 58.0.4
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : 8.1.1
pandas_datareader: None
bs4 : None
bottleneck : 1.3.2
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : None
numba : None
numexpr : 2.8.1
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None

[jreback]

this is an open issue

community patches are how pandas is updated

[nbarone1]

take

[nbarone1]

What is the structure supposed to be for the result?