[Numpy-discussion] Bug in dstack?
Robert Kern
robert.kern@gmail....
Thu Apr 2 20:30:30 CDT 2009
On Thu, Apr 2, 2009 at 19:18, Charles R Harris
<charlesr.harris@gmail.com> wrote:
>
>
> On Thu, Apr 2, 2009 at 5:50 PM, <josef.pktd@gmail.com> wrote:
>>
>> On Thu, Apr 2, 2009 at 6:46 PM, Charles R Harris
>> <charlesr.harris@gmail.com> wrote:
>> > Note:
>> >
>> > In [133]: l = [[1,0,0],[1,1,0],[1,1,1]]
>> >
>> > In [134]: dstack(l)
>> > Out[134]:
>> > array([[[1, 1, 1],
>> > [0, 1, 1],
>> > [0, 0, 1]]])
>> >
>> > In [135]: dstack(l).shape
>> > Out[135]: (1, 3, 3)
>> >
>> >
>> > Shouldn't the shape be (3,3)? Also, for generalized ufuncs and
>> > broadcasting
>> > I think a function that stacked along the first axis instead of the last
>> > would be useful. Maybe gstack or astack?
>> >
>> > Chuck
>> >
>>
>> I think dstack, vstack and hstack work exactly as advertised in the
>> docs, at least that's how I interpret and use them
>>
>> Josef
>>
>> >>> l = [[1,0,0],[1,1,0],[1,1,1]]
>> >>> np.dstack(l).shape
>> (1, 3, 3)
>> >>> np.vstack(l).shape
>> (3, 3)
>> >>> np.hstack(l).shape
>> (9,)
>> >>> np.vstack(l)
>> array([[1, 0, 0],
>> [1, 1, 0],
>> [1, 1, 1]])
>> >>> np.dstack(l)[:,:,0]
>> array([[1, 0, 0]])
>> >>> np.dstack(l)[:,:,1]
>> array([[1, 1, 0]])
>> >>> np.dstack(l)[:,:,2]
>> array([[1, 1, 1]])
>> ________
>
> But stacking 2D arrays gives the same number of dimensions as stack 1D
> vectors:
>
> In [1]: dstack([eye(3)]*3)
> Out[1]:
> array([[[ 1., 1., 1.],
> [ 0., 0., 0.],
> [ 0., 0., 0.]],
>
> [[ 0., 0., 0.],
> [ 1., 1., 1.],
> [ 0., 0., 0.]],
>
> [[ 0., 0., 0.],
> [ 0., 0., 0.],
> [ 1., 1., 1.]]])
>
> In [2]: dstack([eye(3)]*3).shape
> Out[2]: (3, 3, 3)
>
> So dstack is turning 1D vectors iinto 1xn arrays so that
>
> In [8]: dstack([[1,0]]*3).shape
> Out[8]: (1, 2, 3)
>
> In [9]: dstack([[[1,0]]]*3).shape
> Out[9]: (1, 2, 3)
>
> Note the different bracket counts. This messes up general applications so
> that dstacking 1D vectors has to be treated differently than dstacking
> arrays. You can't sum along the last axis of the stack and get a 1D vector
> back, whereas for dstacked 2D arrays summing along the last axis gives back
> a 2D array.
Depth stack presumes at-least-3D arrays (hence "along the third axis"
rather than "along the last axis") and "upcasts" input arrays with
atleast_3d(). That's just what it does and has always done.
--
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