BUG: to_datetime ignores "AM" and "PM" unless format is explicitly set

[juli4nb4dillo]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd
>>> df = pd.DataFrame(['4/21/2023 11:06:35 AM','4/21/2023 12:06:35 PM', '4/21/2023 1:00:48 PM', '4/21/2023 2:35:37 PM'])
>>> df
                       0
0  4/21/2023 11:06:35 AM
1  4/21/2023 12:06:35 PM
2   4/21/2023 1:00:48 PM
3   4/21/2023 2:35:37 PM
>>> pd.to_datetime(df[0])
0   2023-04-21 11:06:35
1   2023-04-21 12:06:35
2   2023-04-21 01:00:48  # PM is ignored, parsing as 01:00 instead of 13:00
3   2023-04-21 02:35:37

Issue Description

to_datetime ignores the AM and PM particles, hence parsing both as 24H format.

Expected Behavior

>>> df = pd.DataFrame(['4/21/2023 11:06:35 AM','4/21/2023 12:06:35 PM', '4/21/2023 1:00:48 PM', '4/21/2023 2:35:37 PM'])
>>> df
                       0
0  4/21/2023 11:06:35 AM
1  4/21/2023 12:06:35 PM
2   4/21/2023 1:00:48 PM
3   4/21/2023 2:35:37 PM
>>> pd.to_datetime(df[0])
0   2023-04-21 11:06:35
1   2023-04-21 12:06:35
2   2023-04-21 13:00:48
3   2023-04-21 14:35:37

Installed Versions

INSTALLED VERSIONS

commit : 37ea63d
python : 3.10.4.final.0
python-bits : 64
OS : Darwin
OS-release : 21.6.0
Version : Darwin Kernel Version 21.6.0: Thu Mar 9 20:12:21 PST 2023; root:xnu-8020.240.18.700.8~1/RELEASE_ARM64_T6000
machine : arm64
processor : arm
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 2.0.1
numpy : 1.24.2
pytz : 2023.3
dateutil : 2.8.2
setuptools : 67.4.0
pip : 23.0.1
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : 3.0.9
lxml.etree : 4.9.2
html5lib : None
pymysql : None
psycopg2 : 2.9.5
jinja2 : None
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : None
numba : None
numexpr : None
odfpy : None
openpyxl : 3.1.2
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.10.1
snappy : None
sqlalchemy : None
tables : None
tabulate : 0.9.0
xarray : None
xlrd : 2.0.1
zstandard : None
tzdata : 2023.3
qtpy : None
pyqt5 : None

[davep33l]

Hi @juli4nb4dillo, I am new here so apologies if I misunderstand the bug you have raised.

Using the format parameter would get the desired result

I recreated your code with the format parameter and returned the following:

import pandas as pd
df = pd.DataFrame(['4/21/2023 11:06:35 AM','4/21/2023 12:06:35 PM', '4/21/2023 1:00:48 PM', '4/21/2023 2:35:37 PM'])
print(df)
df[0] = pd.to_datetime(df[0],format='%m/%d/%Y %I:%M:%S %p')
print(df)

0 4/21/2023 11:06:35 AM
1 4/21/2023 12:06:35 PM
2 4/21/2023 1:00:48 PM
3 4/21/2023 2:35:37 PM

0 2023-04-21 11:06:35
1 2023-04-21 12:06:35
2 2023-04-21 13:00:48
3 2023-04-21 14:35:37

[juli4nb4dillo]

@Rainman33 That's correct, it's not a big deal if I manually set the format. However it used to work (without format parameter) with pandas 1.5.3, hence it triggered a few hidden bugs in our software once we upgraded to pandas 2.x.
I suggest raising a warning in case of no format?

[lithomas1]

Hi @juli4nb4dillo,
Not at all an expert with our datetime code, but the code changes you're seeing might be a consequence of PDEP-4.
You can read about it here https://pandas.pydata.org/pdeps/0004-consistent-to-datetime-parsing.html

If I pass in format="mixed", I get

>>> pd.to_datetime(df[0], format="mixed")
0   2023-04-21 11:06:35
1   2023-04-21 12:06:35
2   2023-04-21 13:00:48
3   2023-04-21 14:35:37
Name: 0, dtype: datetime64[ns]

which I think is what you expected.

It's strange that the parsing didn't throw an error, though. cc @MarcoGorelli who probably knows whether this is a bug or not.

[MarcoGorelli]

thanks @juli4nb4dillo for the report (and @lithomas1 for the ping!)

this looks like a bug: we have

In [2]: guess_datetime_format('4/21/2023 11:06:35 AM')
Out[2]: '%m/%d/%Y %H:%M:%S %p'

and it looks like this should instead have been

'%m/%d/%Y %I:%M:%S %p'

?

[MarcoGorelli]

This is the part of the code which would need changing

pandas/_libs/tslibs/parsing.pyx

Not sure whether to suggest special-casing '%H' to become '%I' if there's %p, or to just stop guessing '%p' for safety. Tempted to go with the latter

[MarcoGorelli]

cc @davesque @mroeschke

[mroeschke]

If AM/PM is in the guessing string, then switching to %I where relevant makes sense. Otherwise I don't think it's safe to ever guess %p if it's not

[MarcoGorelli]

so something like

• if %I is in the guessed format and %H is there too, replace %H with %I
• if %I is in the guessed format, and neither %H nor %I are, then return None

Anyone fancy making a PR?