indexing image with a list of indices cannot, in general, return a consistent affine

[Bago]

In [66]: img[..., np.array((0,1))]

     95 # allow slicing of form [...,1]
     96 if Ellipsis in index:
---> 97 if sum([i == Ellipsis for i in index]) > 1:
     98 raise ValueError("only one Ellipsis (...) allowed in slice")
     99 # convert ellipsis to series of slice(None) objects. For

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

I believe the i == Ellipsis should be i is Ellipsis

[Matthew Brett (matthew-brett)]

Hi Bago,

On Thu, Jan 28, 2010 at 10:45 PM, Bago <email address hidden> wrote:
> Public bug reported:
>
> In [66]:  img[..., np.array((0,1))]
>
>
>     95         # allow slicing of form [...,1]
>     96         if Ellipsis in index:
> ---> 97             if sum([i == Ellipsis for i in index]) > 1:
>     98                 raise ValueError("only one Ellipsis (...) allowed in slice")
>     99             # convert ellipsis to series of slice(None) objects.  For
>
> ValueError: The truth value of an array with more than one element is
> ambiguous. Use a.any() or a.all()
>
> I believe the i == Ellipsis should be i is Ellipsis

Ah - yes - the problem is, that implementing fancy indexing (indexing
with an array) would be rather hard. Do you have a particular
use-case in mind? Can it be covered by getting the array:

data = np.asarray(img)
sampled = data[index_array]

and then putting in your own affine? The complications arise trying
to deal with all the possible crazy things you can do with fancy
indexing, when keeping track of the affine...

Thanks a lot,

Matthew

[Bago (mrbago)]

Ah, i see what you're saying. I didn't consider that because my use case is pretty simple. b0 = diff_img[..., b_values == 0] (b_values == 0 can have multiple true entries). I can use the fix you suggested, but this brings up a new issue. img[[0,1,10], ...] does return an image but img[np.array([0,1,10]), ...] does not. Should the first case raise an error, should the second case raise a more informative error?

Bago

[Matthew Brett (matthew-brett)]

Yo,

On Thu, Jan 28, 2010 at 11:38 PM, Bago <email address hidden> wrote:
> Ah, i see what you're saying. I didn't consider that because my use case
> is pretty simple. b0 = diff_img[..., b_values == 0] (b_values == 0 can
> have multiple true entries). I can use the fix you suggested, but this
> brings up a new issue. img[[0,1,10], ...] does return an image but
> img[np.array([0,1,10]), ...] does not. Should the first case raise an
> error, should the second case raise a more informative error?

I think the second case should raise a more informative error, like

'Sorry, we don't support fancy (numpy array) indexing on image objects'

Does that sound reasonable?

See you,

Matthew

[Bago (mrbago)]

I think the first case should raise the same error

Bago

[Matthew Brett (matthew-brett)]

I agree - thanks for keeping up with this.

[Matthew Brett (matthew-brett)]

Fixed in trunk I believe - could you test? Thanks a lot.