[Numpy-discussion] Binary ufuncs: minimum

Charles R Harris charlesr.harris@gmail....
Tue May 27 16:54:37 CDT 2008


On Tue, May 27, 2008 at 3:28 PM, Charles R Harris <charlesr.harris@gmail.com>
wrote:

>
>
> On Tue, May 27, 2008 at 2:40 PM, Travis E. Oliphant <
> oliphant@enthought.com> wrote:
>
>> Charles R Harris wrote:
>> >
>> >
>> > On Tue, May 27, 2008 at 2:10 PM, Travis E. Oliphant
>> > <oliphant@enthought.com <mailto:oliphant@enthought.com>> wrote:
>> >
>> >     Stéfan van der Walt wrote:
>> >     > Did this change recently?
>> >     >
>> >     > In [33]: np.__version__
>> >     > Out[33]: '1.1.0.dev5211'
>> >     >
>> >     > In [34]: np.minimum(np.uint8(164), np.uint64(12807)).dtype
>> >     > Out[34]: dtype('uint64')
>> >     >
>> >     > But yes, that looks like it should return a uint8.
>> >     >
>> >     This discussion is really moot unless a proposal for how to handle
>> >     different casting rules for different ufuncs is proposed.   Right
>> now,
>> >     the type-promotion rules are generic and do not depend on the
>> >     ufunc only
>> >     on coercion rules for the mixed types.
>> >
>> >     One problem with a casting-rules-per-ufunc approach is that it
>> >     makes it
>> >     harder to add new types and have them fit in to the casting
>> structure
>> >     (which is currently possible now).   Some mechanism for allowing the
>> >     types to plug-in to the per-ufunc rules would be needed.
>> >
>> >     These are not impossible things, just a bit of work and not on my
>> >     personal priority list.
>> >
>> >
>> > Not everything follows those rules, however. So I have put these
>> > things up for review so that we can agree on what the are. Of
>> > particular concern in my mind are the bitwise and shift operators
>> > which currently work in a counter intuitive manner.
>>
>> Right now, the only complete description of the rules is the code that
>> implements them.  So, from that perspective, yes everything follows
>> those rules ;-)
>>
>
> So the segfaults are defined behavior? ;) It's like pulling teeth without
> anesthesia to get these things defined and everyone is going to think I'm an
> a-hole. It's a dirty job, but someone has got to do it.
>
> What about abs(-128) returning a negative number for int8? Might it not be
> better to return the corresponding unsigned type? Can't check that at the
> moment, as I'm trying to get 64 bit Ubuntu installed on a software raid1
> partition for further checks, and Ubuntu overwrote my MBR without asking.
> It's a traditional Ubuntu thing.
>

Yep, abs fails:

In [1]: abs(array([-128,-128], dtype=int8))
Out[1]: array([-128, -128], dtype=int8)

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/numpy-discussion/attachments/20080527/7f512a68/attachment.html 


More information about the Numpy-discussion mailing list