Wed, 28 Jul 2010 14:26:36 +0200, Mark Bakker wrote:
> I don't understand why ceil and floor return real values, while the doc
> string says:
>
> The ceil of the scalar `x` is the smallest integer `i`
>
> Wouldn't an integer make more sense?
Which integer? Only arbitrary-size integers (Python longs) are able to
span the whole floating-point range, but we don't necessarily want
np.ceil(some_array)
to start returning object arrays, because of efficiency reasons.
Returning machine integers, on the other hand, can result to overflow.
There, one should (i) raise exceptions on overflow, or, (ii) return bogus
results for overflowed values.
Numpy is here following the C tradition in defining the ceil and floor
functions as float -> float. This leaves overflow handling on casting to
the user.
Python can redefine its floor and ceil since arbitrary-size integers are
first-class citizens in the Python world. This sits less well with
Numpy: (i) Numpy tries to sit close to the hardware, and (ii) strictly
speaking, arbitrary-size integers cannot be a Numpy scalar type since
they by definition are not fixed-size in memory.
--
Pauli Virtanen
_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion
