Why pd.BooleanDtype() is casted to Float64 by groupby/last?

[ghuname]

Code Sample, a copy-pastable example if possible

>>> import pandas as pd
>>>
>>> df = pd.DataFrame({'a': ['x', 'x', 'y', 'y'], 'b': ['x', 'x', 'y', 'y'], 'c': [False, False, True, False]})
>>> df['d'] = df.c.astype(pd.BooleanDtype())
>>>
>>> df.dtypes
a     object
b     object
c       bool
d    boolean
dtype: object
>>>
>>> df.groupby(['a', 'b']).c.last()
a  b
x  x    False
y  y    False
Name: c, dtype: bool
>>>
>>> df.groupby(['a', 'b']).d.last()
a  b
x  x    0.0
y  y    0.0
Name: d, dtype: float64
>

Problem description

df.groupby(['a', 'b']).c.last() returns False, but df.groupby(['a', 'b']).d.last() returns Float64.
Why the difference?

Expected Output

I expect that both values should be False

Output of pd.show_versions()

python : 3.7.4.final.0
pandas : 1.0.3

[dsaxton]

Thanks. Looks like the bug exists at least for min and max as well:

In [1]: import pandas as pd                                                                                                   

In [2]: df = pd.DataFrame({"a": [1, 2], "b": pd.array([True, False])})                                                        

In [3]: df.dtypes                                                                                                             
Out[3]: 
a      int64
b    boolean
dtype: object

In [4]: df.groupby("a")["b"].min()                                                                                            
Out[4]: 
a
1    1.0
2    0.0
Name: b, dtype: float64

In [5]: df.groupby("a")["b"].max()                                                                                            
Out[5]: 
a
1    1.0
2    0.0
Name: b, dtype: float64

[simonjayhawkins]

also occurs for IntDtype see #32194

>>> import pandas as pd
>>> pd.__version__
'1.1.0.dev0+999.gc47e9ca8b'
>>>
>>> df = pd.DataFrame(
...     {"a": ["x", "x", "y", "y"], "b": ["x", "x", "y", "y"], "c": [0, 1, 2, 3]}
... )
>>> df["d"] = df.c.astype(pd.Int64Dtype())
>>>
>>> df.dtypes
a    object
b    object
c     int64
d     Int64
dtype: object
>>>
>>>
>>> df.groupby(["a", "b"]).c.last()
a  b
x  x    1
y  y    3
Name: c, dtype: int64
>>>
>>>
>>> df.groupby(["a", "b"]).d.last()
a  b
x  x    1.0
y  y    3.0
Name: d, dtype: float64
>>>

[ghuname]

Thanks. Looks like the bug exists at least for min and max as well:

You are welcome. I am glad that I can participate (at least by testing) to the development of such a marvel as pandas is.

Best regards.

[ghuname]

In dataframe bellow, lag is Int64 dtype, that I had to cast to int in order make it work:

df_window_lag_sum.assign(lag=lambda x: x.lag.astype(int)).groupby(['cgrp', 'topic']).lag.quantile(quantily_cut_off_value)

Regards.

[dsaxton]

@ghasemnaddaf Can you provide a small reproducible example for this problem?

[ghuname]

@ghasemnaddaf Can you provide a small reproducible example for this problem?

Here you go:

Python 3.7.4 (default, Aug  9 2019, 18:34:13) [MSC v.1915 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import pandas as pd
>>> from pandas import Timestamp
>>> df_dict = {'dtime': {0: Timestamp('2020-03-28 14:15:00'),
...   1: Timestamp('2020-03-28 14:15:00'),
...   2: Timestamp('2020-03-28 14:15:00')},
...  'cgrp': {0: 'grp1',
...   1: 'grp2',
...   2: 'grp3'},
...  'topic': {0: 'top1',
...   1: 'top1',
...   2: 'top1'},
...  'lag': {0: 18, 1: 1, 2: 83}}
>>> df = pd.DataFrame(df_dict)
>>> df.cgrp = df.cgrp.astype(pd.StringDtype())
>>> df.topic = df.topic.astype(pd.StringDtype())
>>> df.lag = df.lag.astype(pd.Int64Dtype())
>>> df
                dtime  cgrp topic  lag
0 2020-03-28 14:15:00  grp1  top1   18
1 2020-03-28 14:15:00  grp2  top1    1
2 2020-03-28 14:15:00  grp3  top1   83
>>> df.dtypes
dtime    datetime64[ns]
cgrp             string
topic            string
lag               Int64
dtype: object
>>> df.groupby(['cgrp', 'topic']).lag.quantile(0.5)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\User\Miniconda3\lib\site-packages\pandas\core\groupby\groupby.py", line 1913, in quantile
    interpolation=interpolation,
  File "C:\Users\User\Miniconda3\lib\site-packages\pandas\core\groupby\groupby.py", line 2291, in _get_cythonized_result
    func(**kwargs)  # Call func to modify indexer values in place
  File "pandas\_libs\groupby.pyx", line 720, in pandas._libs.groupby.__pyx_fused_cpdef
TypeError: No matching signature found
>>> df.assign(lag=lambda x: x.lag.astype(int)).groupby(['cgrp', 'topic']).lag.quantile(0.5)
cgrp  topic
grp1  top1     18.0
grp2  top1      1.0
grp3  top1     83.0
Name: lag, dtype: float64
>>>    

[dsaxton]

Thanks @ghuname , I've opened up a separate issue / PR for this bug

[ghuname]

You are welcome.