Indexing with Series.__getitem__ with DatetimeIndex differs based on monotonicity

[hsxyl]

Code Sample, a copy-pastable example if possible

import pandas as pd

# generate two Series randomly
t1=pd.Series([1.2,2.2,3.0,4,8],index=pd.date_range(start='2015-5-14 00:00:00',end='2015-5-14 00:04:00',freq='T'))
t2=pd.Series([6.5,7.2,8.1,9],index=pd.date_range(start='2015-5-15 00:00:00',end='2015-5-15 00:03:00',freq='T'))

# merge t2 ,t1  -> t3
t3=pd.concat([t2,t1])

key='2015-5-14 00:00:00'

print('type t1 is ',type(t1))
print('type t3 is ',type(t3))

print('(1):',t1[key])
print('(2):',t3[key])
print('(3):',t3[key:key])

Problem description

type t1 is  <class 'pandas.core.series.Series'>
type t3 is  <class 'pandas.core.series.Series'>
(1): 1.2
(2): 2015-05-14    1.2
dtype: float64
(3): 2015-05-14    1.2
dtype: float64

• firstly I concat t1 and t2 and got t3 ,than we can compare t1[key] , t3[key] and t3[key:key]
• I think we expect t3[key] just work like t1[key],but t3[key] work same as t3[key:key]
• it makes people confused extremely, in fact,I cost hours for that

Expected Output

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.6.final.0
python-bits: 64
OS: Darwin
OS-release: 17.5.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: None
LOCALE: None.None

pandas: 0.23.4
pytest: None
pip: 18.1
setuptools: 36.4.0
Cython: None
numpy: 1.12.1
scipy: 0.19.1
pyarrow: None
xarray: None
IPython: 7.1.1
sphinx: None
patsy: 0.5.1
dateutil: 2.7.5
pytz: 2017.2
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 3.0.1
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None

[mroeschke]

I agreed that this is pretty inconsistent. Here's a simpler example without the concat step, as overall I think this has to do with whether the index is monotonic:

In [16]: mon_s = pd.Series(range(3), index=pd.date_range(start='2015-5-13 23:59:00', freq='min', periods=3))

In [17]: non_mon_s = pd.Series(range(3), index=[mon_s.index[1], mon_s.index[0], mon_s.index[2]])

In [18]: non_mon_s
Out[18]:
2015-05-14 00:00:00    0
2015-05-13 23:59:00    1
2015-05-14 00:01:00    2
dtype: int64

In [19]: mon_s
Out[19]:
2015-05-13 23:59:00    0
2015-05-14 00:00:00    1
2015-05-14 00:01:00    2
Freq: T, dtype: int64

In [20]: mon_s[key]
Out[20]: 1

In [21]: non_mon_s[key]
Out[21]:
2015-05-14    0
dtype: int64

[jbrockmendel]

Using @mroeschke's example, it looks like the behavior diverges when we get to DatetimeIndex._partial_date_slice, which starts with a check:

        is_monotonic = self.is_monotonic
        if (
            is_monotonic
            and reso in ["day", "hour", "minute", "second"]
            and self._resolution >= Resolution.get_reso(reso)
        ):
            # These resolution/monotonicity validations came from GH3931,
            # GH3452 and GH2369.

            # See also GH14826
            raise KeyError

this raises for the monotonic index but not the non-monotonic one.