BUG: Inconsistent behavior for isin() over columns with different datatypes

[dmarx]

• 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

s_obj = pd.Series(['1', '2', '3'])
s_int = pd.Series([1, 2, 3])

s_obj.isin(s_int).sum()  # 0 
s_int.isin(s_obj).sum() # 3 
s_obj.astype(np.int).isin(s_int).sum() # 3

Problem description

It is not uncommon for numeric fields to be read into a dataframe column as text. The behavior of the isin function is inconsistent in these cases, working as expected in the s_int.isin(s_obj) case, but failing in the s_obj.isin(s_int) case unless the text field is coerced to numeric first by the user.

Expected Output

Behavior should be consistent in both directions. Either isin() should always fail when there is a type mismatch, or it should perform necessary coersions automatically. It would be helpful to throw a warning in either case to inform the user either that they should standardize their schemas or that a potentially costly coersion is going to happen automatically. At the very least, we could keep the current behavior and show the user a warning explaining that an empty result set could be due to a type mismatch.

Middle ground solution: add something like a coerce_if_necessary option to isin(). Defaulting to False would return the current behavior (which preserves backwards compatibility and ensures we don't surprise users with expensive coersions). We could augment this behavior with a new warning, alerting users to the type mismatch and recommending solutions. Users who aren't comfortable with datatypes (or are just feeling lazy) could then set coerce=True to have necessary coersions attempted for them behind the scenes.

Interested to hear other's thoughts regarding what the preferred behavior should be.

Output of pd.show_versions()

INSTALLED VERSIONS

commit : None
python : 3.8.3.final.0
python-bits : 64
OS : Windows
OS-release : 10
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 10, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : English_United States.1252

pandas : 1.0.3
numpy : 1.18.1
pytz : 2020.1
dateutil : 2.8.1
pip : 20.0.2
setuptools : 47.1.1.post20200604
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 : 7.13.0
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
pytest : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
xlsxwriter : None
numba : None

[dmarx]

Probably related to #32443, #34256, and #34125

[TomAugspurger]

Looks to be a duplicate of #34125. Closing in favor of that.