Using two indexing arrays on a numpy array

Question

I have a 3-dimensional numpy array, and I'm trying to index it along its second and third dimensions using two single-dimension numpy arrays. To wit:

np.random.seed(0)
dims = (1,5,5)
test_array = np.arange(np.prod(dims)).reshape(*dims)
test_array

produces:

array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19],
        [20, 21, 22, 23, 24]]])

if I then create the two arrays to index it:

idx_dim1 = np.array([0,2,4])
idx_dim2 = np.array([1,3])

I find that I can't apply these both at the same time:

test_array[:,idx_dim1,idx_dim2]
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-193-95b23ed3210c> in <module>()
----> 1 test_array[:,idx_dim1,idx_dim2]

IndexError: shape mismatch: indexing arrays could not be broadcast together with shapes (3,) (2,)

I can split them out, by, say

test_array[:, idx_dim1,:][:,:,idx_dim2]

which gives

array([[[ 1,  3],
        [11, 13],
        [21, 23]]])

But this only works in a read-only sense - I can't assign values to test_array like this, since I'm assigning to a slice rather than the original. I'm also not quite sure why using the two indexers together doesn't work. Is there some conceptual nuance of numpy that I'm not understanding here? And is there a good solution that allows me to assign values?

Thanks!

Are you coming from a MATLAB background? There selecting a block is easy, selecting a scatted set of items awkward. In numpy the block indexing is a bit more awkward, at least on the surface. But notice that the error talks about broadcasting. How would you add a shape (3,) array to a shape (2,)? The same broadcasting rules apply. — hpaulj
– hpaulj, Commented Nov 30, 2018 at 5:38

cs95 · Accepted Answer · 2018-11-30 04:57:26Z

5

Try

test_array[:, idx_dim1[:,None], idx_dim2] = some_value

Where,

>>> test_array[:, idx_dim1[:,None], idx_dim2]
array([[[ 1,  3],
        [11, 13],
        [21, 23]]])

edited Nov 30, 2018 at 4:57

answered Nov 30, 2018 at 4:54

cs95

406k106 gold badges744 silver badges797 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Chris J Harris Over a year ago

Thanks, that's right on the money. But why is that necessary?

cs95 Over a year ago

@Chrisper It's a little complicated. Thankfully, the docs should be able to explain it much better than I can. I would recommend starting here.

denis · Accepted Answer · 2021-10-18 11:09:07Z

1

To index cross products (rectangular subsets) of arrays, use np.ix_ —

A = np.zeros(( 1, 5, 5 ))
A[np.ix_( [0], [0,2,4], [1,3] )] = 1
# print dot0( A )
[[[. 1 . 1 .]
  [. . . . .]
  [. 1 . 1 .]
  [. . . . .]
  [. 1 . 1 .]]]

You can also set cross products with ix_ —

B = np.arange( 3 * 2 ).reshape( 3, 2 )
A[np.ix_( [0], [0,2,4], [1,3] )] = B

edited Oct 18, 2021 at 11:09

answered Oct 18, 2021 at 9:27

denis

22k12 gold badges68 silver badges92 bronze badges

Collectives™ on Stack Overflow

Using two indexing arrays on a numpy array

2 Answers 2

2 Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related