Printing time coordinates with long time intervals

[DPeterK]

If you print a coordinate with a long time unit interval (one of months or years), the time coordinate points are not rendered as datetimes. This is not the case for shorter time unit intervals. For example:

a = np.arange(0, 34, 3)

tc = iris.coords.DimCoord(a, standard_name='time',
                          units=cf_units.Unit('months since 1970-01-01', calendar='gregorian'))
print(tc)
DimCoord([ 0  3  6  9 12 15 18 21 24 27 30 33], standard_name='time', calendar='gregorian')

tc2 = iris.coords.DimCoord(a, standard_name='time',
                           units=cf_units.Unit('days since 1970-01-01', calendar='gregorian'))
print(tc2)
DimCoord([1970-01-01 00:00:00, 1970-01-04 00:00:00, 1970-01-07 00:00:00,
       1970-01-10 00:00:00, 1970-01-13 00:00:00, 1970-01-16 00:00:00,
       1970-01-19 00:00:00, 1970-01-22 00:00:00, 1970-01-25 00:00:00,
       1970-01-28 00:00:00, 1970-01-31 00:00:00, 1970-02-03 00:00:00], standard_name='time', calendar='gregorian')

Follows on from #2354. Thanks to @PAGWatson for originally bringing this to our attention there.

[DPeterK]

@PAGWatson is this the same as what you've seen?

[PAGWatson]

Thanks for following up on this. This is not quite the issue I had. I found that for a cube with a time coord with units in months, I got the following error message when I tried to print the coordinate: "ValueError: Time units with interval of "months", "years" (or singular of these) cannot be processed, got 'months'." This was in Iris 2.1.0.

[DPeterK]

@PAGWatson could you check the version of cf_units you're using? For reference, I'm using v1.2.0.

[PAGWatson]

Mine is 2.0.1

[DPeterK]

Interesting. Having upgraded cf_units I still can't reproduce your specific error. Would you mind printing the entire stacktrace here? That might shed some more light on what's going on.

[PAGWatson]

Sure, here it is:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-355-78dfe530799d> in <module>()
----> 1 print cube.coord('time')[0]

/network/home/aopp/watson/anaconda2/envs/main/lib/python2.7/site-packages/iris/coords.pyc in __str__(self)
    742             bounds = ''
    743             if self.has_bounds():
--> 744                 bounds = ', bounds=' + self._str_dates(self.bounds)
    745             result = fmt.format(self=self, cls=type(self).__name__,
    746                                 points=points, bounds=bounds,

/network/home/aopp/watson/anaconda2/envs/main/lib/python2.7/site-packages/iris/coords.pyc in _str_dates(self, dates_as_numbers)
    722 
    723     def _str_dates(self, dates_as_numbers):
--> 724         date_obj_array = self.units.num2date(dates_as_numbers)
    725         kwargs = {'separator': ', ', 'prefix': '      '}
    726         return np.core.arrayprint.array2string(date_obj_array,

/network/home/aopp/watson/anaconda2/envs/main/lib/python2.7/site-packages/cf_units/__init__.pyc in num2date(self, time_value)
   1986 
   1987         """
-> 1988         cdf_utime = self.utime()
   1989         return _num2date_to_nearest_second(time_value, cdf_utime)

/network/home/aopp/watson/anaconda2/envs/main/lib/python2.7/site-packages/cf_units/__init__.pyc in utime(self)
   1900             emsg = ('Time units with interval of "months", "years" '
   1901                     '(or singular of these) cannot be processed, got {!r}.')
-> 1902             raise ValueError(emsg.format(interval))
   1903 
   1904         #

ValueError: Time units with interval of "months", "years" (or singular of these) cannot be processed, got 'months'.

If I just type "cube.coord('time')[0]" rather than "print cube.coord('time')[0]" at the command prompt, I get the following, in case that's any help:

DimCoord(array([1800.5]), bounds=array([[1800., 1801.]]), standard_name=u'time', units=Unit('months since 1850-01-01 00:00', calendar='gregorian'), long_name=u'time', var_name='time', attributes={'realtopology': 'linear'})

[DPeterK]

@PAGWatson I think I've figured it out (with help from the stacktrace you provided - thanks!). The error is occurring when the time coordinate has bounds, and that gives us something to target a fix for.

Edit: really, we should have handled bounds equivalently to points way back in #2354. We didn't, so it must be time to correct that. PR incoming.

[DPeterK]

For what it's worth, the origin of this issue is in cftime itself, which doesn't handle months or years. This might be because the cf conventions recommend against using these units.

[PAGWatson]

Thanks for following up @dkillick. I guess if there's not an accepted way to handle converting times with units months or years to dates, then it might make sense to just leave the values as numbers, so at least an error isn't thrown? We can't always help the units used in files we get from other sources, of course.

[DPeterK]

it might make sense to just leave the values as numbers

@PAGWatson yup, that's my intention, at least as an interim solution 👍 There is probably a better, more robust solution that can be implemented long-term (perhaps even in cftime), but as a starting point we (a) don't get errors, and (b) get consistent behaviours between time points and bounds.