4D array from 2D arrays

Question

I have certain number of Numpy arrays with the shape

print(x.shape)
>>>(256,256)

How can I stack them so that the shape is

print(y.shape)
>>>(certainnumber,256,256,1)

I've been trying with np.stack and np.concatenate but I only get out of axis errors or stuff like

print(y.shape)
>>>(anothernumber,256)

U need to imagine how the array looks like before coding...then you won't get such errors — bigbounty
– bigbounty, Commented Dec 21, 2017 at 16:27

Divakar · Accepted Answer · 2017-12-21 16:41:16Z

3

Method #1

Here's one with np.stack -

np.stack(list_of_arrays)[...,None]

Method #2

You can prepend a new axis with None/np.newaxis for each of those arrays and concatenate along the first axis for (certainnumber,256,256) shape, like so -

np.concatenate([i[None] for i in list_of_arrays],axis=0)

Then, add new axis as the trailing one for the final (certainnumber,256,256,1) shape, like so -

np.concatenate([i[None] for i in list_of_arrays],axis=0)[...,None]

Sample runs

In [32]: a = np.random.rand(3,4)

In [33]: b = np.random.rand(3,4)

In [34]: list_of_arrays = [a,b]

In [42]: np.stack(list_of_arrays)[...,None].shape
Out[42]: (2, 3, 4, 1)

In [35]: np.concatenate([i[None] for i in list_of_arrays],axis=0)[...,None].shape
Out[35]: (2, 3, 4, 1)

edited Dec 21, 2017 at 16:41

answered Dec 21, 2017 at 16:30

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

2 Comments

juanpa.arrivillaga Over a year ago

I like your answer, it gives a good window into how your mind thinks about these sorts of problems.

Divakar Over a year ago

@juanpa.arrivillaga Not sure if it shows, but my mind works better with shapes, rather than the actual values. This problem fits right into that criteria :)

juanpa.arrivillaga · Accepted Answer · 2017-12-21 16:38:41Z

2

Assuming you have your arrays in some sort of container (you can always put them in a container):

>>> ax = [np.random.randint(0, 10, (3,3)) for _ in range(4)]
>>> ax
[array([[0, 3, 1],
       [4, 2, 4],
       [2, 2, 8]]), array([[8, 4, 6],
       [7, 1, 4],
       [8, 9, 8]]), array([[6, 3, 8],
       [4, 6, 8],
       [2, 2, 9]]), array([[1, 8, 1],
       [0, 9, 2],
       [9, 2, 3]])]

So, you can use np.concatenate but you have to reshape as well:

>>> final = np.concatenate([arr.reshape(1, 3,3,1) for arr in ax], axis=0)

with a result:

>>> final.shape
(4, 3, 3, 1)
>>> final
array([[[[0],
         [3],
         [1]],

        [[4],
         [2],
         [4]],

        [[2],
         [2],
         [8]]],


       [[[8],
         [4],
         [6]],

        [[7],
         [1],
         [4]],

        [[8],
         [9],
         [8]]],


       [[[6],
         [3],
         [8]],

        [[4],
         [6],
         [8]],

        [[2],
         [2],
         [9]]],


       [[[1],
         [8],
         [1]],

        [[0],
         [9],
         [2]],

        [[9],
         [2],
         [3]]]])
>>>

Edit

Inspired by @Divakar to be more generic:

np.concatenate([arr[None,..., None] for arr in ax], axis=0)

edited Dec 21, 2017 at 16:38

answered Dec 21, 2017 at 16:31

juanpa.arrivillaga

97.6k14 gold badges141 silver badges190 bronze badges

4 Comments

Divakar Over a year ago

A generic one (using shape params) would be nice to see.

Divakar Over a year ago

Nice one that latest addition!

Boris Over a year ago

and btw how do I add another array of 256x256 to this afterwards?

Boris Over a year ago

It would be out = np.append(out,ar[None,...], axis=0) being ar another array

jakevdp · Accepted Answer · 2017-12-21 16:45:48Z

1

You can add an axis argument to np.stack to specify which axis you want to stack along:

arrs = [np.random.rand(256, 256) for i in range(11)]
out = np.stack(arrs, axis=0)
out.shape
# (11, 256, 256)

(Note that axis defaults to zero).

If you need to add a one at the end of the shape, then use a newaxis

out[..., np.newaxis].shape
(11, 256, 256, 1)

edited Dec 21, 2017 at 16:45

answered Dec 21, 2017 at 16:43

jakevdp

88.6k15 gold badges140 silver badges177 bronze badges

2 Comments

Divakar Over a year ago

By default, it works along axis=0 :) So, we can skip that axis param, but good for explanation I guess.

Boris Over a year ago

easiest to understand

Collectives™ on Stack Overflow

4D array from 2D arrays

3 Answers 3

2 Comments

Edit

4 Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

2 Comments

Edit

4 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related