BUG: Inconsistent assignment of NAT values to datetime columns in MultiIndex

[galipremsagar]

• 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.

---

Note: Please read this guide detailing how to provide the necessary information for us to reproduce your bug.

Code Sample, a copy-pastable example

>>> import pandas as pd
>>> import numpy as np
>>> midx = pd.MultiIndex(
...         [
...             ["a", "b", "c"],
...             ["house", "store", "forest"],
...             ["clouds", "clear", "storm"],
...             ["fire", "smoke", "clear"],
...             [
...                 np.datetime64("2001-01-01", "ns"),
...                 np.datetime64("2002-01-01", "ns"),
...                 np.datetime64("2003-01-01", "ns"),
...             ],
...             [1.0, np.nan, 2],
...         ],
...         [
...             [0, 0, 0, 0, 1, 1, 2],
...             [1, 1, 1, 1, 0, 0, 2],
...             [0, 0, 2, 2, 2, 0, 1],
...             [0, 0, 0, 1, 2, 0, 1],
...             [1, 0, 1, 2, 0, 0, 1],
...             [1, 0, 1, 2, 0, 0, 1],
...         ],
...     )
>>> midx.names = ["alpha", "location", "weather", "sign", "timestamp", "float"]
>>> 
>>> pdf = pd.DataFrame({'a':[1] , 'b':[2] , 'c':[3] , 'd':[2] , 'e':[0] , 'f':[2], 'g':[10]})
>>> pdf.columns = midx
>>> pdf
alpha              a                                           b                     c
location       store                                       house                forest
weather       clouds                 storm                 storm     clouds      clear
sign            fire                  fire      smoke      clear       fire      smoke
timestamp 2002-01-01 2001-01-01 2002-01-01 2003-01-01 2001-01-01 2001-01-01 2002-01-01
float            NaN        1.0        NaN        2.0        1.0        1.0        NaN
0                  1          2          3          2          0          2         10
>>> pdf["new_col"] = [11]
>>> pdf
alpha              a                                           b                     c new_col
location       store                                       house                forest        
weather       clouds                 storm                 storm     clouds      clear        
sign            fire                  fire      smoke      clear       fire      smoke        
timestamp 2002-01-01 2001-01-01 2002-01-01 2003-01-01 2001-01-01 2001-01-01 2002-01-01     NaT
float            NaN        1.0        NaN        2.0        1.0        1.0        NaN        
0                  1          2          3          2          0          2         10      11
>>> pdf.columns
MultiIndex([(      'a',  'store', 'clouds',  'fire', '2002-01-01', nan),
            (      'a',  'store', 'clouds',  'fire', '2001-01-01', 1.0),
            (      'a',  'store',  'storm',  'fire', '2002-01-01', nan),
            (      'a',  'store',  'storm', 'smoke', '2003-01-01', 2.0),
            (      'b',  'house',  'storm', 'clear', '2001-01-01', 1.0),
            (      'b',  'house', 'clouds',  'fire', '2001-01-01', 1.0),
            (      'c', 'forest',  'clear', 'smoke', '2002-01-01', nan),
            ('new_col',       '',       '',      '',        'NaT',  '')],
           names=['alpha', 'location', 'weather', 'sign', 'timestamp', 'float'])

Problem description

It looks like just for datetime like types the NAT values are being inserted in multiIndex, and not for string or float types.

Previous behavior in 1.2.5 was consistent with all types, i.e., just insert empty string ''

[this should explain why the current behaviour is a problem and why the expected output is a better solution]

Expected Output

Consistency with all dtypes as in 1.2.5

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 5f648bf
python : 3.8.10.final.0
python-bits : 64
OS : Linux
OS-release : 4.15.0-76-generic
Version : #86-Ubuntu SMP Fri Jan 17 17:24:28 UTC 2020
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.3.2
numpy : 1.21.2
pytz : 2021.1
dateutil : 2.8.2
pip : 21.2.4
setuptools : 57.4.0
Cython : 0.29.24
pytest : 6.2.5
hypothesis : 6.17.4
sphinx : 4.1.2
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.0.1
IPython : 7.27.0
pandas_datareader: None
bs4 : 4.9.3
bottleneck : None
fsspec : 2021.08.1
fastparquet : None
gcsfs : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 5.0.0
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : 0.53.1

[simonjayhawkins]

Thanks @galipremsagar for the report.

Previous behavior in 1.2.5 was consistent with all types, i.e., just insert empty string ''

first bad commit: [412554b] BUG/API: DTI/TDI/PI.insert cast to object on failure (#39068)

cc @jbrockmendel

I'm not sure that the 1.2.5 behavior is correct, it returns a Series whereas partial indexing on the first level returns a DataFrame for the other labels, including c which is a single column DataFrame. I also would have expected the numeric level to have a np.nan rather than a empty string and the returned index on master to include all levels.

EDIT: the behavior I refer to above is the return value of pdf["new_col"] which is not in the code sample.

[jbrockmendel]

I also would have expected the numeric level to have a np.nan rather than a empty string

I think this boils down to the fact that pd.Timestamp("") returns NaT while float("") raises. Off the top of my head I can't think of a way to do this that isn't either a) pretty invasive or b) using casting logic in Index.insert different from everywhere else

[simonjayhawkins]

Thanks @jbrockmendel for looking into this. will move to 1.3.4

[simonjayhawkins]

changing milestone to 1.3.5

[jreback]

good to fix but not necessary for 1.3.x

[simonjayhawkins]

moving off 1.3.x milestone

[hagenw]

The same behavior happens for np.timedelta64