Pandas TimeDelta instantiation returns different object for same value

[Dgomzi]

pd.Timedelta(3,unit= 'M')
Out[26]: Timedelta('91 days 07:27:18')
pd.Timedelta('3M')
Out[27]: Timedelta('0 days 00:03:00')

Problem description

I tried instantiating Timedelta object using '3M' which returns Timedelta object with 3 minutes rather than 3Months

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.2.final.0
python-bits: 32
OS: Windows
OS-release: 10
machine: AMD64
processor: Intel64 Family 6 Model 78 Stepping 3, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None
LOCALE: English_United States.1252
pandas: 0.24.1
pytest: None
pip: 10.0.1
setuptools: 39.1.0
Cython: None
numpy: 1.16.1
scipy: 1.2.1
pyarrow: None
xarray: None
IPython: 7.3.0
sphinx: 1.8.4
patsy: 0.5.1
dateutil: 2.8.0
pytz: 2018.9
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 3.0.2
openpyxl: None
xlrd: None

[WillAyd]

Hmm I can see the confusion here. I think just a bug with the parser maybe not expecting months to be a valid identifier, though certainly should given the ISO 8601 spec.

Investigation and PRs would certainly be welcome. cc @mroeschke

[TomAugspurger]

It should raise right? Since 3 months can't be represented as a fixed span of time?

[WillAyd]

Also a good question. It's valid per the ISO 8601 spec:

https://en.wikipedia.org/wiki/ISO_8601#Durations

Though perhaps that would cause some issues with Python's time delta given it only stores up to days, so seems like it could be ambiguous the span of time we want to represent here. Not entirely sure myself...

[TomAugspurger]

This came up with JSON table schema too. It wasn't clear to me then that timedeltas were equivalent to durations. I don't think you can say that "P3M" is iso-8601-duration notation is equivalent "n seconds" where n is the .value of our timedelta.

[mroeschke]

We just deprecated the ability to pass M (month) and Y (year) as units because of the fixed span of time issue @TomAugspurger mentioned. #23264

The behavior of pd.Timedelta('3M') might have been changed recently in #23439, but since we are deprecating M, it might be a bug that we expect to go away with the deprecation.

[mroeschke]

Going to close this issue as it looks like a duplicate of #27285 with more examples