convert from 'to_julian_date()' to 'np.datetime64'

Question

consider array below

dt = DatetimeIndex(['2016-01-01', '2016-01-01', '2016-01-21', '2016-01-21'], dtype='datetime64[ns]', name=u'date', freq=None)

i convert the above to_julian_date()dtype

j = dt.to_julian_date()
Float64Index([2457388.5, 2457388.5, 2457408.5, 2457408.5], dtype='float64')

how can I convert jback to dt

I tried

dt = pd.to_datetime(j, errors = 'coerce')

it converts j back to datetime object, but the values are not the same, here is the output

DatetimeIndex(['1970-01-01 00:00:00.002457388',
           '1970-01-01 00:00:00.002457388',
           '1970-01-01 00:00:00.002457408',
           '1970-01-01 00:00:00.002457408'],
          dtype='datetime64[ns]', freq=None)

Answer 1

Julian dates are number of days from a long time ago. A time prior to the pandas Timestamp epoch. We refer to the beginning of when a system of tracking time begins tracking time. For Timestamp that is '1970-01-01'. What I do is get the Timestamp epoch with pd.to_datetime(0, unit='s'). That is the first possible pandas.Timestamp. Then I assign the number of days of that Timestamp from the beginning of the Julian Date by running to_julian_date() on that Timestamp. Now that I have the number of days from the beginning of the Julian start date to the Timestamp epoch, I subtract that from each of the Julian dates in j and that becomes the number of days from the Timestamp epoch (or '1970-01-01'). I can then use pd.to_datetime(j - epoch, unit='D') to give me the Timestamps where each of my values represent the number of days from the Timestamp epoch.

I hope that is clear ;-)

first find the julian_date for the pd.Timestamp epoch

epoch = pd.to_datetime(0, unit='s').to_julian_date()

then conversion is done with pd.to_datetime using the parameter unit='D'

pd.to_datetime(j - epoch, unit='D')

DatetimeIndex(['2016-01-01', '2016-01-01', '2016-01-21', '2016-01-21'], dtype='datetime64[ns]', freq=None)