Python numpy 2D array sum over certain indices

Question

There is a 2-d array like this:

img = [
  [[1, 2, 3], [4, 5, 6], [7, 8, 9]],
  [[2, 2, 2], [3, 2, 3], [6, 7, 6]],
  [[9, 8, 1], [9, 8, 3], [9, 8, 5]]
]

And i just want to get the sum of certain indices which are like this:

indices = [[0, 0], [0, 1]] # which means img[0][0] and img[0][1]
# means here is represents

There was a similar ask about 1-d array in stackoverflow in this link, but it got a error when I tried to use print(img[indices]). Because I want to make it clear that the element of img are those which indicates by indices, and then get the ~~mean~~ sum of it.

Expected output

[5, 7, 9]

Just to be cleared, when you sum [0,0] and [0,1], the result would be [5, 7, 9], or total 5 +7 + 9 = 21? — Tấn Nguyên, Jan 18 '21 at 03:00
Can you provide an example that uses actual arrays, not lists, shows the expected output, and indices that aren't transposeable? — Mad Physicist, Jan 18 '21 at 03:18

Pablo C · Accepted Answer · 2021-01-18T03:17:29.723

3

Use NumPy:

import numpy as np

img = np.array(img)
img[tuple(indices)].sum(axis = 0)
#array([5, 7, 9])

edited Jan 18 '21 at 03:17

answered Jan 18 '21 at 03:00

Pablo C

4,661
2
8
24

@MadPhysicist why? – Pablo C Jan 18 '21 at 03:10
I am sorry to say that i did not clear that `img[0][0] + img[0][1]` should be `[5, 7, 9]`, so in this answer i got the right answer when i changed the `axis = 1` to `axis = 0`. – luneice Jan 18 '21 at 03:15
@luneice edited. by the way, you want the sum or the mean? – Pablo C Jan 18 '21 at 03:17
1

He want the sum basically. The comment is quite confuse "# which means image[0][0] and image[0][1]", `means` here is `represents` – Tấn Nguyên Jan 18 '21 at 03:21
Also, you want to transpose the indices most likely – Mad Physicist Jan 18 '21 at 03:21
i want to sum of it. – luneice Jan 18 '21 at 03:28
yes, In comment `# which means image[0][0] and image[0][1]`, the word `means` here is `represents` – luneice Jan 18 '21 at 03:33
@TấnNguyên I was talking about the text above that – Pablo C Jan 18 '21 at 03:35
@MadPhysicist I still don't get why you say this is not quite right – Pablo C Jan 18 '21 at 03:38
I'm pretty sure the index is transposed, but the question is actually pretty ambiguous, so you might be right – Mad Physicist Jan 18 '21 at 04:14

Mad Physicist · Answer 2 · 2021-01-18T03:22:49.630

When you supply a fancy index, each element of the index tuple represents a different axis. The shape of the index arrays broadcasts to the shape of the output you get.

In your case, the rows of indices.T are the indices in each axis. You can convert them into an index tuple and append slice(None), which is the programmatic equivalent of :. You can take the mean of the resulting 2D array directly:

img[tuple(indices.T) + (slice(None),)].sum(0)

Another way is to use the splat operator:

img[(*indices.T, slice(None))].sum(0)

score 1 · Answer 3 · answered Jan 18 '21 at 03:16

1

If the result would be [5, 7, 9] which is sum over the column of the list. Then easy:

img = np.asarray(img)
indices = [[0, 0], [0, 1]]
img[(indices)].sum(axis = 0)

Result:

array([5, 7, 9])

answered Jan 18 '21 at 03:16

Tấn Nguyên

1,607
4
15
25

Python numpy 2D array sum over certain indices

3 Answers3

Linked