Python Reshape 3d array into 2d

Question

I want to reshape the numpy array as it is depicted, from 3D to 2D. Unfortunately, the order is not correct.

A assume to have a numpy array (1024, 64, 100) and want to convert it to (1024*100, 64).

Does anybody has an idea how to maintain the order?

I have a sample data

data[0,0,0]=1
data[0,1,0]=2
data[0,2,0]=3
data[0,3,0]=4
data[1,0,0]=5
data[1,1,0]=6
data[1,2,0]=7
data[1,3,0]=8
data[2,0,0]=9
data[2,1,0]=10
data[2,2,0]=11
data[2,3,0]=12
data[0,0,1]=20
data[0,1,1]=21
data[0,2,1]=22
data[0,3,1]=23
data[1,0,1]=24
data[1,1,1]=25
data[1,2,1]=26
data[1,3,1]=27
data[2,0,1]=28
data[2,1,1]=29
data[2,2,1]=30
data[2,3,1]=31

and I would like to have an outcome like this:

array([[  1.,   2.,   3.,   4.],
       [  5.,   6.,   7.,   8.],
       [  9.,  10.,  11.,  12.],
       [ 20.,  21.,  22.,  23.],
       [ 24.,  25.,  26.,  27.],
       [ 28.,  29.,  30.,  31.]])

---

Moreover, I would also like to have the reshaping in the other way, i.e. from:

array([[  1.,   2.,   3.,   4.],
       [  5.,   6.,   7.,   8.],
       [  9.,  10.,  11.,  12.],
       [ 20.,  21.,  22.,  23.],
       [ 24.,  25.,  26.,  27.],
       [ 28.,  29.,  30.,  31.]])

to the desired output:

 [[[  1.  20.]
  [  2.  21.]
  [  3.  22.]
  [  4.  23.]]

 [[  5.  24.]
  [  6.  25.]
  [  7.  26.]
  [  8.  27.]]

 [[  9.  28.]
  [ 10.  29.]
  [ 11.  30.]
  [ 12.  31.]]]

Answer 1

Using einops:

# start with (1024, 64, 100) to (1024*100, 64):
einops.rearrange('h w i -> (i h) w')

# or we could concatenate along horizontal axis to get (1024, 64 * 100):
einops.rearrange('h w i -> h (i w)')

See docs for more examples

Answer 2

It looks like you can use numpy.transpose and then reshape, like so -

data.transpose(2,0,1).reshape(-1,data.shape[1])

Sample run -

In [63]: data
Out[63]: 
array([[[  1.,  20.],
        [  2.,  21.],
        [  3.,  22.],
        [  4.,  23.]],

       [[  5.,  24.],
        [  6.,  25.],
        [  7.,  26.],
        [  8.,  27.]],

       [[  9.,  28.],
        [ 10.,  29.],
        [ 11.,  30.],
        [ 12.,  31.]]])

In [64]: data.shape
Out[64]: (3, 4, 2)

In [65]: data.transpose(2,0,1).reshape(-1,data.shape[1])
Out[65]: 
array([[  1.,   2.,   3.,   4.],
       [  5.,   6.,   7.,   8.],
       [  9.,  10.,  11.,  12.],
       [ 20.,  21.,  22.,  23.],
       [ 24.,  25.,  26.,  27.],
       [ 28.,  29.,  30.,  31.]])

In [66]: data.transpose(2,0,1).reshape(-1,data.shape[1]).shape
Out[66]: (6, 4)

To get back original 3D array, use reshape and then numpy.transpose, like so -

In [70]: data2D.reshape(np.roll(data.shape,1)).transpose(1,2,0)
Out[70]: 
array([[[  1.,  20.],
        [  2.,  21.],
        [  3.,  22.],
        [  4.,  23.]],

       [[  5.,  24.],
        [  6.,  25.],
        [  7.,  26.],
        [  8.,  27.]],

       [[  9.,  28.],
        [ 10.,  29.],
        [ 11.,  30.],
        [ 12.,  31.]]])