np.vectorize and nan - how can I make them play nice?

Question

Let's say I have

>>> import numpy as np
>>> nv = np.array([-1, np.nan, 1])

np.sin will work as expected

>>> np.sin(nv)
array([-0.84147098,         nan,  0.84147098])

However if I try that with vectorize on my own function it'll fail

>>> def noneg(n):
        if n < 0:
             return 0
        return n
>>> noneg(nv)
...
ValueError: cannot convert float NaN to integer

This is because the initize value returned by noneg is the integer zero and then we get nan which is a float.

The solution I've found so far is:

>>> @np.vectorize
    def noneg(n):
        if not np.isnan(n) and n < 0:
            return n.__class__(0)
        return n
>>> noneg(nv)
array([  0.,  nan,   1.])

However this looks ugly, is there a better way to ignore nan in vectorize?

Answer 1

Hmm, you already wrote the answer?

def noneg(n):
    if n < 0:
         return n.__class__(0)
    return n
noneg(nv)

The problem here is the variable 0 is not concerning your input type, I think.