Stack several 2D arrays to produce a 3D array

Question

I have 4 numpy arrays, each of shape (5,5). I would like to stack them such that I get a new array of shape (5,5,4). I tried using:

N = np.stack((a, b, c, d))

but, as I am new to using numpy, I cannot understand why that is giving a shape of (4, 5, 5) instead of (5, 5, 4). Is there another method I should be using? dstack works but changes my arrays, I think it transposes them.

For example, 4 arrays

[[1,2]
 [3,4]]

[[1,2]
 [3,4]]

[[1,2]
 [3,4]]

[[1,2]
 [3,4]]

when stacked I am expecting:

[[[1,2]
 [3,4]]

[[1,2]
 [3,4]]

[[1,2]
 [3,4]]

[[1,2]
 [3,4]]]

This is working as expected with stack but would give a shape of (4,2,2) instead of (2,2,4). From my understanding, shape is (rows, columns, depth) Am I wrong in this?

Answer 1

I believe you could concatenate the arrays, and reshape into a 3D array as:

l = [a,b,c,d]
np.concatenate(l).reshape(len(l), *a.shape)

Or if you want to avoid creating that list and know the amount of arrays beforehand:

np.concatenate((a,b,c,d)).reshape(4, *a.shape)

Checking on the shared example:

a = [[1, 2], [3, 4]]
d = c = b = a

np.concatenate((a,b,c,d)).reshape(4, *np.array(a).shape)
array([[[1, 2],
        [3, 4]],

       [[1, 2],
        [3, 4]],

       [[1, 2],
        [3, 4]],

       [[1, 2],
        [3, 4]]])

Answer 2

In [10]: arr = np.arange(1,5).reshape(2,2)
In [11]: np.stack((arr,arr,arr))
Out[11]: 
array([[[1, 2],
        [3, 4]],

       [[1, 2],
        [3, 4]],

       [[1, 2],
        [3, 4]]])
In [12]: _.shape
Out[12]: (3, 2, 2)

Default stack joins the arrays on a new first axis, the same as np.array((arr,arr,arr)).shape

If given an axis parameter it can join them as:

In [13]: np.stack((arr,arr,arr), axis=2)
Out[13]: 
array([[[1, 1, 1],
        [2, 2, 2]],

       [[3, 3, 3],
        [4, 4, 4]]])
In [14]: _.shape
Out[14]: (2, 2, 3)

np.dstack does the same thing, where d stands for 'depth`'.

The last dimension (here 3) is displayed as the innermost columns.

Selecting one 'channel' produces a 2d array:

In [17]: np.stack((arr,arr,arr), axis=2)[:,:,0]
Out[17]: 
array([[1, 2],
       [3, 4]])

For 3 dimensions, the first dimension is blocks or planes, and the middle rows. Those names are conveniences, helping us visualize the action, but don't have inherent means in numpy . For images the last dimension often is called colors or channels, and has size 3 or 4. A 4d array of images could described as

(batches, height, width, color)

But the actual meanings depend on how you are processing the array.