Question

import numpy
a = numpy.array([20090913, 20101020, 20110125])

Can you explain why numpy.datetime64(a.astype("S8").tolist()) converts correctly but not numpy.datetime64(a.astype("S8"))? Why would a numpy function not accept numpy strings and only regular python strings? I am using Numpy 1.6.1 with Python 2.7.2 Windows 7.

Answer 1

Reproducing your results:

>>> a = numpy.array([20090913, 20101020, 20110125])

>>> numpy.datetime64(a.astype("S8").tolist())
array([2009-09-13 00:00:00, 2010-10-20 00:00:00, 2011-01-25 00:00:00], dtype=datetime64[us])

>>> numpy.datetime64(a.astype("S8"))
array([1970-01-01 00:00:20.090913, 1970-01-01 00:00:20.101020,
       1970-01-01 00:00:20.110125], dtype=datetime64[us])

Here’s the key:

>>> a.astype("S8").tolist()
['20090913', '20101020', '20110125']
>>> a.astype("S8")
array(['20090913', '20101020', '20110125'],
      dtype='|S8')

In the first case, the string arguments get passed on to numpy.datetime64 and get parsed properly, exactly as you’ve described. In the second, it needs to perform an explicit coercion from |S8 as surmised. It turns out this is being considered, but currently explicitly isn’t supported:

This didn’t go in, because the datetime properties don’t exist on the
arrays after you convert them to datetime64, so there could be some
unintuitive consequences from that. When Martin implemented the
quaternion dtype, we discussed the possibility that dtypes could
expose properties that show up on the array object, and if this were
implemented I think the conversion and compatibility between python
datetime and datetime64 could be made quite natural.

The documentation has more examples of working coercions you may wish to consider, including from other numpy time formats. If you feel the need for explicit type coercion is in error, I’d recommend reporting it to the numpy team and, if possible, submitting your own patch.