[Numpy-discussion] array allocation using tuples gives views of same array

Timothy Hochberg tim.hochberg at ieee.org
Thu Nov 15 11:32:24 EST 2007

On Nov 15, 2007 9:11 AM, Hans Meine <meine at informatik.uni-hamburg.de> wrote:

> Am Donnerstag, 15. November 2007 16:29:12 schrieb Warren Focke:
> > On Thu, 15 Nov 2007, George Nurser wrote:
> > > It looks to me like
> > > a,b = (zeros((2,)),)*2
> > > is equivalent to
> > > x= zeros((2,))
> > > a,b=(x,)*2
> >
> > Correct.
> >
> > > If this is indeed a feature rather than a bug, is there an alternative
> > > compact way to allocate many arrays?
> >
> > a, b = [zeros((2,)) for x in range(2)]
> Let me add that this is a standard Python caveat, which also happens with
> lists -- many of us have once tried to initialize an array of empty lists
> with ([], ) * N, which results in N references to the same empty list.
> Warren pointed out the standard solution above.

And I'll just add that another option in this case is to avoid tuples and
lists altogether and just unpack a larger array:

a, b = zeros([2,2])

That's mostly inferior to the list comprehension solution in terms of
clarity, but may appeal to the fans of the pithy. In some cases it also
expresses intent better, but not often.


Using lists for shapes is clearer both typographically: compare zeros((2,))
and zeros([2]), and conceptually: shapes are closer to lists (variable
length, homogeneous) than tuples (fixed length, inhomogeneous).  Tuples are
used for the shape attribute since they need to be immutable, but there's no
reason to type them that way; it just makes things hard to read.


.  __
.   |-\
.  tim.hochberg at ieee.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20071115/7696e603/attachment.html>

More information about the NumPy-Discussion mailing list