str.split on np.nan gives np.nan in one column but None in another column

[JeroenDelcour]

import pandas as pd
import numpy as np

s = pd.Series(['19HT|C2', np.nan, '20ZT|C1'])
print(s)

0    19HT|C2
1        NaN
2    20ZT|C1
dtype: object

s_split = s.str.split('|', expand=True)
print(s_split)

      0     1
0  19HT    C2
1   NaN  None
2  20ZT    C1

print(s_split.dtypes)

0    object
1    object
dtype: object

print(type(s_split.loc[1,0]))

float

print(type(s_split.loc[1,1]))

NoneType

Problem description

When np.nan gets split, it becomes np.nan (of type float) in the first column but None (of type NoneType) in the second column. I'd consider this unexpected behavior. How come splitting a value of one type results in two values of different types?

Expected Output

      0     1
0  19HT    C2
1   NaN   NaN
2  20ZT    C1

Either np.nan or None in both columns, but not a mix of both. I'd say np.nan makes most sense, since that's the original value of the row.

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.0.final.0
python-bits: 64
OS: Linux
OS-release: 4.10.0-40-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.21.0
pytest: 3.0.5
pip: 9.0.1
setuptools: 27.2.0
Cython: 0.25.2
numpy: 1.13.1
scipy: 0.19.1
pyarrow: 0.7.1
xarray: None
IPython: 5.1.0
sphinx: 1.5.1
patsy: 0.4.1
dateutil: 2.6.0
pytz: 2016.10
blosc: None
bottleneck: 1.2.1
tables: 3.4.2
numexpr: 2.6.2
feather: 0.4.0
matplotlib: 2.0.2
openpyxl: 2.4.1
xlrd: 1.0.0
xlwt: 1.2.0
xlsxwriter: 0.9.6
lxml: 4.1.1
bs4: 4.5.3
html5lib: 0.9999999
sqlalchemy: 1.1.5
pymysql: None
psycopg2: None
jinja2: 2.9.4
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None

[WillAyd]

Just to be clear the issue here is how expand=True handles NaN values. You'd still mix str and None instances using expansion if the split character isn't in one of the records. For example,

s = pd.Series(['19HT|C2', np.nan, '20ZT|C1', 'foo'])
s.str.split("|", expand=True)

yields

      0     1
0  19HT    C2
1   NaN  None
2  20ZT    C1
3   foo  None

The str_split docstring is a little ambiguous because it says it should propagate NaN values, but I think that references the value returned irrespective of the expansion mechanism. Will leave it to others here to comment as to whether or not we think this is a bug with how expansion works, or if the docstring should be modified.

pandas/pandas/core/strings.py

Line 1005 in 97fea48

pattern, propagating NA values. Equivalent to :meth:`str.split`.

[jreback]

this is a bug; we want to use np.nan as the missing value indicator

[louis-red]

FYI this behavior (pd.Series.str.split with expand=True expands with Nones) is still present in my version of pandas : 0.23.1

[markemus]

+1 for me, 0.23.4