Question

I’m new to NumPy and I’m trying to compute some simple statistics like the median or the stddev.

One of the “columns” I want to explore is a time difference (being its type the timedelta64 NumPy type), but I can’t apply those statistical ufuncs directly:

----> 1 age_request.std()

TypeError: ufunc 'divide' not supported for the input types, and the inputs could not be 
safely coerced to any supported types according to the casting rule 'safe'

Why is this happening?

I know i should look into Pandas, but first I would like to familiarize myself with NumPy.

Answer 1

Take a look at the documentation for datetime. It lists the operations you can perform on a timedelta object. Division is done like so:

t1 = t2 // i

Be aware that this computes the floor and throws away any remainder.

As far as I know, you can only do these operations on a timedelta object. Though perhaps I am wrong and you can use numpy operations.

I would suggest converting the timedelta object to microseconds and then doing your division, standard deviation, median, etc:

tdMicrosecs = td.microseconds + 1000.0 * td.seconds + 86400000000.0 * td.days
tdQuotient = tdMicrosecs / i

(I should add, though, that I’m not entirely sure what the maximum integer value is that python can store and if tdMicrosecs could exceed it.)