[Numpy-discussion] ufunc oddities

Robert Kern robert.kern@gmail....
Sun May 25 03:02:00 CDT 2008


On Sat, May 24, 2008 at 11:14 PM, Charles R Harris
<charlesr.harris@gmail.com> wrote:

> It used to be stay in type and has been changed, and I don't disagree with
> that, it was discussed on the list. Nevertheless, booleans are different,
> both their own kind and integers. But my problem is not convenience, my
> problem is the very inconvenient one of writing comprehensive tests, and for
> that the desired behavior has to be specified; it can't simply be taken as
> whatever currently happens.

Fine. I have verified that the current behaviors you have mentioned
are intended. If you want a more concise specification of the special
cases, it is this: add.reduce() (and consequently sum()) is the
special case. In that context, bool_s are treated as integers, and the
integer dtypes plus bool_ use the default integer dtype for the
accumulator to forestall overflow in the most common usage. Everything
else should follow the generic rules.

When bool_s are operated with bool_s, +*- take on Boolean algebraic
meanings. When a bool_ is cast to another dtype, True->1 and False->0.
bool_ is not part of the integer "kind" so the generic cross-kind
rules apply when operations combine bool_s with integers.

Does that help?

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless
enigma that is made terrible by our own mad attempt to interpret it as
though it had an underlying truth."
 -- Umberto Eco


More information about the Numpy-discussion mailing list