BUG: Groupby: Int-Datatype (of group-by-columns) always casted to int64

[Mueller2-Patrick]

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of pandas.
• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample, a copy-pastable example

import pandas as pd
import numpy as np

df = pd.DataFrame({'col1':[1,2,3],'col2':[10,11,12]})
df['col1'] = df['col1'].astype(np.int8)

print(df.dtypes)

print(df.groupby('col1', as_index=False)['col2'].count().dtypes)

Problem description

Before the groupby: The dtype of col1 is np.int8:
Output of

print(df.dtypes):

col1     int8
col2    int64
dtype: object

After the groupby the dtype of col1 is np.int64
Output of

print(df.groupby('col1', as_index=False)['col2'].count().dtypes):

col1    int64
col2    int64
dtype: object

Expected Output

print(df.groupby('col1', as_index=False)['col2'].count().dtypes) gives output:

col1    int8
col2    int64
dtype: object

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.7.1.final.0
python-bits : 64
OS : Windows
OS-release : 10
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 9, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None
pandas : 1.0.4
numpy : 1.17.4
pytz : 2019.3
dateutil : 2.8.1
pip : 10.0.1
setuptools : 45.2.0
Cython : None
pytest : 5.4.1
hypothesis : None
sphinx : 2.3.0
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.10.3
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : None
matplotlib : 3.0.3
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 0.12.0
pytables : None
pytest : 5.4.1
pyxlsb : None
s3fs : None
scipy : 1.4.1
sqlalchemy : 1.3.12
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
xlsxwriter : None
numba : None

[rhshadrach]

Thanks for reporting this - it seems to me this is expected, datatypes may need to change to hold the correct result. For example, taking the mean of a Boolean/integer should result in float. Taking the count of a column with say 10k rows with dtype int8, changing the datatype to int64 so that the correct result can be stored seems preferable.

[rhshadrach]

Ahh, I missed that it was the grouping column that is changing. That changes things.

[rhshadrach]

Even when using as_index=False, internally the unique values for the groupby are stored as an Index. I think that makes this a duplicate of #16404. Perhaps a conversion could be done after the operation is computed in this case. But then, with 16404 open, we would have two different results when using as_index=True/False.

[TomAugspurger]

Closing as a duplicate of #16404. To support this we'll need hashtables / algos that operate on all these integer dtypes.