BUG: dtype changes when row passed to a function via DataFrame.apply()

[buhtz]

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

Reproducible Example

#!/usr/bin/env python3
import pandas as pd

# 1.2.5 and 1.3.2
print(pd.__version__)

df = pd.DataFrame(data={'VAR': ['A', 'B'], 'X': range(2)})
df.VAR = df.VAR.astype('category')
df.X = df.X.astype('int8')

# Name: VAR, dtype: category
# Categories (2, object): ['A', 'B']
print(df.VAR)

# VAR    category
# X          int8
# dtype: object
print(df.dtypes)


def foo(row):
    # doing nothing
    return row

for arg in [None, 'reduce', 'expand', 'broadcast']:
    print(f'\n{arg}')
    print(df.apply(foo, axis=1, result_type='reduce').dtypes)

Issue Description

When I "iterate" on row level via DataFrame.apply(foo, axis=1) the dtype of the columns are changing.
Here in the MWE a category becomes object and an int8 becomes `int64'.

The argument result_type= seems to have no effect. The MWE checks each possible option for it.

This is the output of the MWE

1.3.2
0    A
1    B
Name: VAR, dtype: category
Categories (2, object): ['A', 'B']
VAR    category
X          int8
dtype: object

None
VAR    object
X       int64
dtype: object

reduce
VAR    object
X       int64
dtype: object

expand
VAR    object
X       int64
dtype: object

broadcast
VAR    object
X       int64
dtype: object

Expected Behavior

Just keep the dtype. Do not convert it without users permission. or minimally throw a warning if there is a good reason for it.

Possible related #42001

Installed Versions

INSTALLED VERSIONS

commit : 5f648bf
python : 3.9.4.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.18362
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 12, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : de_DE.cp1252

pandas : 1.3.2
numpy : 1.21.1
pytz : 2021.1
dateutil : 2.8.1
pip : 21.2.4
setuptools : 57.0.0
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : 1.1
pymysql : None
psycopg2 : None
jinja2 : 2.11.3
IPython : 7.23.1
pandas_datareader: None
bs4 : 4.9.3
bottleneck : None
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : 3.4.2
numexpr : None
odfpy : None
openpyxl : 3.0.7
pandas_gbq : None
pyarrow : 4.0.1
pyxlsb : None
s3fs : None
scipy : 1.7.0
sqlalchemy : 1.4.13
tables : None
tabulate : 0.8.9
xarray : None
xlrd : None
xlwt : None
numba : None

[WXZhao7]

I meet the similar problem when using datafram.apply(lambda x: f-string)

• At first, I check the different formatting methods.

import pandas as pd

data = {'A': [1, 2, 3], 'B': [1.0, 2.0, 3.0]}
df = pd.DataFrame(data)
temp = df.astype({'A': int, 'B': float})
print("-- Dataframe.apply change the data type --")
print(temp.apply(lambda x: f"{x['A']}-{x['B']}", axis='columns'))
print("-- data type --")
print(temp.apply(lambda x: f"{type(x['A'])}:{type(x['B'])}", axis='columns'))
print("-- Error")
try:
    d = temp.apply(lambda x: f"{x['A']:2d}:{x['B']:2f}", axis='columns')
except Exception as E:
    print(E)
else:
    print(d)
print("\n\n-- other test")
print("--- %-formatting is ok --")
print(temp.apply(lambda x: "%2d-%2f" % (x['A'], x['B']), axis='columns'))
print("--- str.format() wrong --")
print(temp.apply(lambda x: "{}-{}".format(type(x['A']), type(x['B'])), axis='columns'))

pd.show_versions()

The result shows that f-string and str.format() are different with %-formatting

-- Dataframe.apply change the data type --
0    1.0-1.0
1    2.0-2.0
2    3.0-3.0
dtype: object
-- data type --
0    <class 'numpy.float64'>:<class 'numpy.float64'>
1    <class 'numpy.float64'>:<class 'numpy.float64'>
2    <class 'numpy.float64'>:<class 'numpy.float64'>
dtype: object
-- Error
Unknown format code 'd' for object of type 'float'


-- other test
--- %-formatting is ok --
0     1-1.000000
1     2-2.000000
2     3-3.000000
dtype: object
--- str.format() wrong --
0    <class 'numpy.float64'>-<class 'numpy.float64'>
1    <class 'numpy.float64'>-<class 'numpy.float64'>
2    <class 'numpy.float64'>-<class 'numpy.float64'>
dtype: object

INSTALLED VERSIONS
------------------
commit           : 945c9ed766a61c7d2c0a7cbb251b6edebf9cb7d5
python           : 3.9.7.final.0
python-bits      : 64
OS               : Linux
OS-release       : 5.11.0-40-generic
Version          : #44~20.04.2-Ubuntu SMP Tue Oct 26 18:07:44 UTC 2021
machine          : x86_64
processor        : x86_64
byteorder        : little
LC_ALL           : None
LANG             : en_US.UTF-8
LOCALE           : en_US.UTF-8

pandas           : 1.3.4
numpy            : 1.21.4
pytz             : 2021.3
dateutil         : 2.8.2
pip              : 21.3.1
setuptools       : 58.5.3
Cython           : None
pytest           : None
hypothesis       : None
sphinx           : None
blosc            : None
feather          : None
xlsxwriter       : None
lxml.etree       : 4.6.4
html5lib         : None
pymysql          : None
psycopg2         : None
jinja2           : None
IPython          : 7.29.0
pandas_datareader: None
bs4              : None
bottleneck       : None
fsspec           : None
fastparquet      : None
gcsfs            : None
matplotlib       : 3.4.3
numexpr          : None
odfpy            : None
openpyxl         : None
pandas_gbq       : None
pyarrow          : None
pyxlsb           : None
s3fs             : None
scipy            : 1.7.2
sqlalchemy       : 1.4.26
tables           : None
tabulate         : None
xarray           : 0.20.1
xlrd             : None
xlwt             : None
numba            : None

Then I do some tests, I find this may be caused by the data type change during the Series obtained from Dataframe.

f = lambda x: f"{type(x['A'])}:{type(x['B'])}"
x = df.iloc[0,]
print("--- [check] lambda function")
print(f(x))
print("--- [check] get a series from dataframe")
print(">>> type(df.iloc[0]['A'])")
print(type(df.iloc[0]['A']))
print(">>> type(df.loc[0,'A'])")
print(type(df.loc[0, 'A']))

--- [check] lambda function
<class 'numpy.float64'>:<class 'numpy.float64'>
--- [check] get a series from dataframe
>>> type(df.iloc[0]['A'])
<class 'numpy.float64'>
>>> type(df.loc[0,'A'])
<class 'numpy.int64'>

Because df.loc[0,'A'] and df.iloc[0]['A'] show different data type.