Timedelta returns wrong value

[ayang]

Code Sample

In [6]: pd.Timedelta(6, 'M')
Out[6]: Timedelta('182 days 14:54:36')

In [7]: pd.Timedelta('6M')
Out[7]: Timedelta('0 days 00:06:00')

In [8]: pd.Timedelta(6, 'W')
Out[8]: Timedelta('42 days 00:00:00')

In [9]: pd.Timedelta('6W')
Out[9]: Timedelta('42 days 00:00:00')

Problem description

Out[7] shoud return 182 days but 6 minutes

[jreback]

In [1]: pd.__version__                                                                                                                                                                                                                                   
Out[1]: '0.25.0.dev0+889.g903a09c6a'

In [2]: pd.Timedelta('6M')                                                                                                                                                                                                                               
Out[2]: Timedelta('0 days 00:06:00')

In [3]: pd.Timedelta('6min')                                                                                                                                                                                                                             
Out[3]: Timedelta('0 days 00:06:00')

In [4]: pd.Timedelta(6, 'min')                                                                                                                                                                                                                           
Out[4]: Timedelta('0 days 00:06:00')

In [5]: pd.Timedelta(6, 'M')                                                                                                                                                                                                                             
/nas/local/simscripts/jreback/.nest/bamboo/bin/ipython:1: FutureWarning: M and Y units are deprecated and will be removed in a future version.
  #!/usr/bin/python2.7
Out[5]: Timedelta('182 days 14:54:36')

M was deprecated, you should use min for minutes; in any event months are generally not really useful / supported with Timedelta as they don't represent a fixed period.

[jorisvandenbossche]

Given that 'M' is the unit for month, and also when specifying a freq in other situations it is interpreted like that, should we maybe not interpret '6M' as 6 minutes?

[jbrockmendel]

For strptime and strftime %M is minutes, e.g. %y-%m-%d %H:%M:%S

[jorisvandenbossche]

For strptime and strftime %M is minutes

Ah, that's true. But I would argue that units and offsets are "closer" to timedelta, so are more important to take into account when deciding what "M" should mean (but it is certainly unfortunate that formats strings and units are basically the other way around)

[jbrockmendel]

we now no longer allow passing unit="M" but we do allow pd.Timedelta("6M") to mean 6 minutes. Do we want to disallow it there too?

[theNicelander]

If I'm reading this correctly, the plan is to drop support for Monthly Timedeltas in the future. There are many cases where monthly Timedeltas are very useful, even if they don't represent a fixed amount of time as @jreback mentioned. Most businesses rely on monthly reporting, so having the option to do monthly Timedeltas is crucial.

[jreback]

@theNicelander you can already do monthly timeseltas with offsets which handle this type of. variable movement

Timedelta is a fixed unit and so doesn’t support non fixed values