[Numpy-discussion] Should we fix Ticket #709?

Stéfan van der Walt stefan@sun.ac...
Fri Jul 4 01:48:05 CDT 2008

2008/7/3 Zbyszek Szmek <zbyszek@in.waw.pl>:
> On Wed, Jul 02, 2008 at 10:55:12PM +0200, Stéfan van der Walt wrote:
>> 2008/7/2 Zbyszek Szmek <zbyszek@in.waw.pl>:
>> >> That's Ticket #709 <http://projects.scipy.org/scipy/numpy/ticket/709>:
>> >>
>> >> > I'm faily sure that:
>> >> > numpy.isnan(datetime.datetime.now()
>> >> > ...should just return False and not raise an exception.
>> > IMHO numpy.isnan() makes no sense for non-numerical types.
>> I agree with Chuck's rationale [if someone asks me whether a peanut
>> butter sandwhich is a Koala bear, then I'd say no, without trying to
>> cast the sandwhich to a mammal], and it seems that other ufuncs try to
>> be general this way, e.g. np.add handles arbitrary classes with
>> __add__ methods.
> Exactly, you can add things only when they are numerical or when you have
> defined __add__, otherwise it's a TypeError.
> If it was important and general enough, we could add an attribute __isnan__,
> and extend isnan-iness to other things, but it probably isn't.
> The operator 'is' makes sense for both mammals and sandwiches, so the statement
> 'a butter sandwich is a Koala' makes sense.
> I would compare isnan-iness to asking if
> 'a butter sandwich is even-toed'? It is defined for some mammals, and doesn't
> have a definite value for other things.

We may be talking about two different things here:

- NaN: the IEEE floating point number
- Not a number: anything that is not a number

The `isnan`-function specifically aims to identify the first case.

> Defining isnan for arbitrary objects seems a workaround for passing
> objects to a wrong place function and pollution of code with a kludge for
> a special case.

Sure, this can be done.  Then the function will simply become:

Return False unless the input object is a NumPy floating-point scalar
(or a Python float) that matches the IEEE NaN bit-pattern.

Long live even-toed peanut-butter eating Koalas!


More information about the Numpy-discussion mailing list