Question

I have a numpy.ndarray. The first two columns are dates of type datetime.datetime. I would like to find the difference in months (as a float) between the two where I define this as (difference in days)*(12/365). So I would like to get access to the .days property of the timedelta objects returned by subtracting two datetimes.

I can’t do this: (MyArray[1] - MyArray[0]).days because Python doesn’t know to expect timedelta objects. In C# I might try casting like this:

((timedelta)(MyArray[1] - MyArray[0])).days

What’s the way to do this in Python (numpy)? Do I need to loop?

Answer 1

Unfortunately, it’s not doable without iterating on your array of timedelta

deltas = MyArray[1]-MyArray[0]
diff_days = np.fromiter((d.days for d in deltas), dtype=float, count=len(deltas))

Here, we’re using np.fromiter with the count argument to be more efficient (it preallocates the size of the array).

Note that the new datetime64 dtype will not at this date (2012/09) allow you to get properties like days.