Reshaping 2D numpy array to 4D-CodePudding

I have a 2D numpy array X (batch x channels х pool_size * pool_size),
where first row corresponds to the first channel in first batch, second row -
to the second channel in the first batch, third row - to the first channel
in the second batch and the last - to the second channel in the second batch:

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

Is it possible to convert it into 4D array Y (batch x pool_size x pool_size x channels)
without any loops using only numpy methods so that
Y[0, :, :, 0] is

[[1, 0],
[0, 0]]

Y[0, :, :, 1] is

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

Y[1, :, :, 0] is

[[0, 0], 
[3, 0]]

Y[1, :, :, 1] is

[[0, 0], 
[0, 4]]

UPD: For the input array

[[ 1, 2, 3, 4],
 [ 5, 6, 7, 8],
 [ 9, 10, 11, 12], 
 [ 13, 14, 15, 16]]

expected results are:
Y[0, :, :, 0] is

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

Y[0, :, :, 1] is

[[5, 6], 
[7, 8]]

Y[1, :, :, 0] is

[[9, 10], 
[11, 12]]

Y[1, :, :, 1] is

[[13, 14], 
[15, 16]]

CodePudding user response：

y = y.reshape(y.shape[0],28,28,1)

CodePudding user response：

You can use the reshape and then use the swapaxes.

A = np.array([[ 1, 0, 0, 0],
 [ 0, 2, 0, 0],
 [ 0, 0, 3, 0], 
 [ 0, 0, 0, 4]])
B = A.reshape(2,2,2, 2)

Here is the output:

array([[[[1, 0],
     [0, 0]],

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


   [[[0, 0],
     [3, 0]],

    [[0, 0],
     [0, 4]]]])

array([[[[ 1,  2],
     [ 3,  4]],

    [[ 5,  6],
     [ 7,  8]]],


   [[[ 9, 10],
     [11, 12]],

    [[13, 14],
     [15, 16]]]])