to_timedelta does not recognize 'h' and 'm' units

[vfilimonov]

'h' and 'm' units are not correctly recognized in to_timedelta:

print pd.to_timedelta(1,unit='D')
print pd.to_timedelta(1,unit='h')
print pd.to_timedelta(1,unit='m')
print pd.to_timedelta(1,unit='s')
print pd.to_timedelta(1,unit='ms')
print pd.to_timedelta(1,unit='us')
print pd.to_timedelta(1,unit='ns')

results in

86400000000000 nanoseconds
1 nanoseconds
1 nanoseconds
1000000000 nanoseconds
1000000 nanoseconds
1000 nanoseconds
1 nanoseconds

version: 0.14.0

Most likely related to Issue #6423

[jreback]

these are not valid numpy units (eg have to use H and M)
that said I thought this did work

[vfilimonov]

It's the same issue for 'H' and 'M...

I picked 'h' and 'm' from to_timedelta's docstring:

unit : unit of the arg (D,h,m,s,ms,us,ns) denote the unit, which is an integer/float number

[jreback]

actually this is completely tested
what is your numpy version?

[vfilimonov]

Sorry - forgot to tell: numpy 1.8.1

[vfilimonov]

Great, @jreback, thanks a lot!

[jreback]

@vfilimonov thanks for reporting! was a bit buggy

[shoyer]

@jreback Is there still a good reason for not handling h and m in pd.to_timedelta? It looks like pd.tslib.cast_from_unit can handle them now, and everything appears to work if I add them to regex in pandas.tseries.timedeltas.

[jreback]

#7616

fixed that - it was buggy before

[shoyer]

@jreback Can we re-open this issue then? I'd like to submit a patch...

[jreback]

can u post what is not working?

[shoyer]

>>> pd.to_timedelta('1h')
ValueError                                Traceback (most recent call last)
<ipython-input-80-e5635a653a40> in <module>()
----> 1 pd.to_timedelta('1h')

/Users/shoyer/miniconda/envs/xray-dev/lib/python2.7/site-packages/pandas/tseries/timedeltas.pyc in to_timedelta(arg, box, unit)
     67 
     68     # ...so it must be a scalar value. Return scalar.
---> 69     return _coerce_scalar_to_timedelta_type(arg, unit=unit)
     70 
     71 _unit_map = {

/Users/shoyer/miniconda/envs/xray-dev/lib/python2.7/site-packages/pandas/tseries/timedeltas.pyc in _coerce_scalar_to_timedelta_type(r, unit)
    117 
    118         # we are already converting to nanoseconds
--> 119         converter = _get_string_converter(r, unit=unit)
    120         r = converter()
    121         unit='ns'

/Users/shoyer/miniconda/envs/xray-dev/lib/python2.7/site-packages/pandas/tseries/timedeltas.pyc in _get_string_converter(r, unit)
    181 
    182     # no converter
--> 183     raise ValueError("cannot create timedelta string converter for [{0}]".format(r))
    184 
    185 def _possibly_cast_to_timedelta(value, coerce=True, dtype=None):

ValueError: cannot create timedelta string converter for [1h]

While we're at it, it would also be nice to include support all relevant long names (e.g., "hour", "min", "minute" etc). It is weird that only "day" and "days" work.

[jreback]

oh you are taking about the actual parser

ok (this issue covers things like
to_timedelta(1,unit='h') )

pls create a new issue for updates parsing. of strings then
the current impl was a bit quick and dirty so didn't cover all possibilities (although not hard I think)

prob should include some vbench / maybe cythonize at some point

it's a pretty limited regular format so shouldn't be that hard to make fast