reshape numpy 3D array to 2D

Question

I have a very big array with the shape = (32, 3, 1e6) I need to reshape it to this shape = (3, 32e6)

On a snippet, how to go from this::

>>> m3_3_5
array([[[8, 4, 1, 0, 0],
        [6, 8, 5, 5, 2],
        [1, 1, 1, 1, 1]],

       [[8, 7, 1, 0, 3],
        [2, 8, 5, 5, 2],
        [1, 1, 1, 1, 1]],

       [[2, 4, 0, 2, 3],
        [2, 5, 5, 3, 2],
        [1, 1, 1, 1, 1]]])

to this::

>>> res3_15
array([[8, 4, 1, 0, 0, 8, 7, 1, 0, 3, 2, 4, 0, 2, 3],
       [6, 8, 5, 5, 2, 2, 8, 5, 5, 2, 2, 5, 5, 3, 2],
       [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])

I did try various combinations with reshape with no success::

>>> dd.T.reshape(3, 15)
array([[8, 8, 2, 6, 2, 2, 1, 1, 1, 4, 7, 4, 8, 8, 5],
       [1, 1, 1, 1, 1, 0, 5, 5, 5, 1, 1, 1, 0, 0, 2],
       [5, 5, 3, 1, 1, 1, 0, 3, 3, 2, 2, 2, 1, 1, 1]])

>>> dd.reshape(15, 3).T.reshape(3, 15)
array([[8, 0, 8, 2, 1, 8, 0, 8, 2, 1, 2, 2, 5, 2, 1],
       [4, 0, 5, 1, 1, 7, 3, 5, 1, 1, 4, 3, 5, 1, 1],
       [1, 6, 5, 1, 1, 1, 2, 5, 1, 1, 0, 2, 3, 1, 1]])

You probably need a transpose in addition to a reshape. I'm sure you'll get answers but you'll probably learn more by experimenting yourself. — hpaulj
– hpaulj, Commented Feb 18, 2016 at 0:41
Indeed. And for what it's worth, I just found the appropriate transpose by trying a few things until one worked. :-) — DSM
– DSM, Commented Feb 18, 2016 at 0:42
Try transpose with an argument, something like transpose([2,0,1]) (just a guess). — hpaulj
– hpaulj, Commented Feb 18, 2016 at 1:02

Akavall · Accepted Answer · 2016-02-18 01:26:57Z

3

a.transpose([1,0,2]).reshape(3,15) will do what you want. (I am basically following comments by @hpaulj).

In [14]: a = np.array([[[8, 4, 1, 0, 0],
        [6, 8, 5, 5, 2],
        [1, 1, 1, 1, 1]],

       [[8, 7, 1, 0, 3],
        [2, 8, 5, 5, 2],
        [1, 1, 1, 1, 1]],

       [[2, 4, 0, 2, 3],
        [2, 5, 5, 3, 2],
        [1, 1, 1, 1, 1]]])

In [15]: a.transpose([1,0,2]).reshape(3,15)
Out[15]: 
array([[8, 4, 1, 0, 0, 8, 7, 1, 0, 3, 2, 4, 0, 2, 3],
       [6, 8, 5, 5, 2, 2, 8, 5, 5, 2, 2, 5, 5, 3, 2],
       [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])

answered Feb 18, 2016 at 1:26

Akavall

86.8k58 gold badges214 silver badges260 bronze badges

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

1 Comment

user3313834 Over a year ago

ok it clear now, I was missing the transpose *axes argument for nD array n>2

Garrett R · Accepted Answer · 2016-02-18 02:05:37Z

3

You can get the desired behavior with np.hstack

# g is your (3,3,5) array from above
reshaped = np.hstack(g[i,:,:] for i in range(3))  #uses a generator exp
reshaped_simpler = np.hstack(g) # this produces equivalent output to the above statmement
print reshaped # (3,30)

Output

array([[8, 4, 1, 0, 0, 8, 7, 1, 0, 3, 2, 4, 0, 2, 3],
       [6, 8, 5, 5, 2, 2, 8, 5, 5, 2, 2, 5, 5, 3, 2],
       [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])

edited Feb 18, 2016 at 2:05

answered Feb 18, 2016 at 0:40

Garrett R

2,66213 silver badges15 bronze badges

2 Comments

user3313834 Over a year ago

yes, the trivial m3_3_5.reshape(3, 15) does not provide the expected result

hpaulj Over a year ago

You can simplify this to np.hstack(g) or np.concatenate(g, axis=-1). It iterates of the input, effectively splitting it on the 1st axis, and then reassembles it on the last axis. This reorders the values as desired (in this case).

Collectives™ on Stack Overflow

reshape numpy 3D array to 2D

2 Answers 2

1 Comment

Output

2 Comments

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

Output

2 Comments

Linked

Related