Cannot replace all occurences of infs and nans to None with a single df.replace

[dougvj]

Example code

>>> df = pd.DataFrame(
...         {'foo':[
...             np.nan,
...             np.inf,
...             1.1,
...             2.2
...             ],
...         'bar':[
...             1,
...             -np.inf,
...             2.1,
...             3.2
...             ]
...         })
>>> df
   foo  bar
0  NaN  1.0
1  inf -inf
2  1.1  2.1
3  2.2  3.2
>>> df.replace({np.nan: None, -np.inf:None, np.inf:None})
   foo  bar
0  NaN  1.0
1  NaN  NaN
2  1.1  2.1
3  2.2  3.2
>>> 

Problem description

When trying to replace all occurrences of NaNs and Infs to Nones the resulting dataframe simply contains 'NaNs' for every occurrence, retaining the dtype of float64 rather than the expected object. A work around is to simply issue the replace again with only np.nan to None.

This behavior was discovered when trying to convert the contents of a dataframe containing Infs, and NaNs to Nones in preparation for JSON compliant output via a dictionary for compatibility with third party tools rather than directly through panda's to_json() method (which admittedly would make this process unnecessary as it produces the compliant output already)

I strongly suspect this problem is related to inconsistencies outlined here: #29024

Expected Output

I would expect a column type of object with Nones in place as I see when replacing only NaNs.

Output of pd.show_versions()

[paste the output of pd.show_versions() here below this line]
INSTALLED VERSIONS

commit : None
python : 3.8.1.final.0
python-bits : 64
OS : Linux
OS-release : 5.4.13-arch1-1
machine : x86_64
processor :
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.0.0rc0+233.gec0996c675
numpy : 1.18.1
pytz : 2019.3
dateutil : 2.8.1
pip : 19.3
setuptools : 44.0.0
Cython : 0.29.14
pytest : None
hypothesis : None
sphinx : 2.2.1
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.4.2
html5lib : 1.0.1
pymysql : None
psycopg2 : None
jinja2 : 2.10.3
IPython : 7.11.1
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : 4.4.2
matplotlib : 3.1.2
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
pytest : None
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

[s-scherrer]

I can replicate this here:

>>> df = pd.DataFrame({
...         'foo':[np.nan, np.inf, 1.1, 2.2 ],
...         'bar':[ 1, -np.inf, 2.1, 3.2 ]
... })
>>> dfnew = df.replace({np.nan: None, -np.inf: None, np.inf: None})
>>> dfnew
   foo  bar
0  NaN  1.0
1  NaN  NaN
2  1.1  2.1
3  2.2  3.2

As @dougvj mentioned, running replace a second time works:

>>> dfnew.replace({np.nan: None, -np.inf: None, np.inf: None})
    foo   bar
0  None     1
1  None  None
2   1.1   2.1
3   2.2   3.2

However, if the replacement dictionary contains a replacement pair that will actually replace something, it doesn't work anymore:

>>> dfnew.replace({np.nan: None, 1: 2})
   foo  bar
0  NaN  2.0
1  NaN  NaN
2  1.1  2.1
>>>     

Replacing something else will also turn Nones back to Nans:

>>> dfnew.replace({np.nan: None, -np.inf: None, np.inf: None}).replace({1:2})
   foo  bar
0  NaN  2.0
1  NaN  NaN
2  1.1  2.1
3  2.2  3.2

So the issue here is probably that replace converts None to NaN.