How can I reshape a numpy array in a certain way without using a loop?-CodePudding

I have been trying to reshape a numpy array by using not more than one loop or even better no loop at all. But after several trials with np.reshape() and np.transpose(), I still do not have a clue how to transform an array like this:

array([[1,2,6,3,2,5],
      [9,4,7,3,2,6],
      [8,4,6,3,2,6],
      [9,5,0,4,5,2],
      [0,5,2,6,4,2],
      [8,6,2,5,7,4]])

into an array like this:

[[1,2,9,4,8,4,0,5,8,6],
[6,3,7,3,6,3,0,4,2,6,2,5],
[2,5,2,6,2,6,5,2,4,2,7,4]]

I see the pattern is,

# original
[[x11, x12, x13, x14, x15, x16],          
[x21, x22, x23, x24, x25, x26],                 
[x31, x32, x33, x34, x35, x36],       
[x41, x42, x43, x44, x45, x46],          
[x51, x52, x53, x54, x55, x56],          
[x61, x62, x63, x64, x65, x66]]

# result
[[x11, x12, x21, x22, x31, x32, x41, x42, x51, x52, x61, x62],
[x13, x14, x23, x23, x33, x34, x43, x44, x53, x54, x63, x64],
[x15, x16, x25, x26, x35, x36, x45, x46, x55, x56, x65, x66]]

but my biggest problem is to connect to rows like shown above, especially by using vectorization instead of loops. But I was not even able to do it with one Loop. I needed two.

As I said, I tried many things. For example, I reshaped the original array to

[[1,2]
[6,3],
[2,5],
[9,4],
[7,3],
[2,6]
[8,4],
[6,3],
[2,6]
[9,5],
[0,4],
[5,2],
[0,5],
[2,6],
[4,2],
[8,6],
[2,5],
[7,4]]

From there I had no clue how to work with the rows to get the desired array.

I also tried to instantly reshape the array to match the desired shape, but then I also struggled to make the values match my desired result.

CodePudding user response：

Note that reshape cannot change the order of the elements in an array, only the strides. But it's quite straightforward to do with sliding_window_view. Of course we just want every other pair of columns, hence the ::window[1], and finalyl you want the pairs of columns as one row, hence the reshape.

window = (a.shape[0], 2)
c = np.lib.stride_tricks.sliding_window_view(a, window)[:, ::window[1], ...].reshape(-1, window[0]*window[1])

CodePudding user response：

We can use a simple approach if we know the dimensions of the array:

arr = np.array([[1,2,6,3,2,5],
[9,4,7,3,2,6],
[8,4,6,3,2,6],
[9,5,0,4,5,2],
[0,5,2,6,4,2],
[8,6,2,5,7,4]])

d1, d2= arr.shape
np.reshape(np.split(arr, d1//2,axis=1),(d1//2,d2*2))

CodePudding user response：

There are plenty of ways to do this, but directly reshaping is not one of it. In your case I would do as

# let a be your input array
a = np.array( 
    [
    [1,2,6,3,2,5],
    [9,4,7,3,2,6],
    [8,4,6,3,2,6],
    [9,5,0,4,5,2],
    [0,5,2,6,4,2],
    [8,6,2,5,7,4]]
)

np.concatenate([a[:, :2] , a[:, 2:4], a[:, 4:]]).reshape(-1,12)

array([[1, 2, 9, 4, 8, 4, 9, 5, 0, 5, 8, 6],
       [6, 3, 7, 3, 6, 3, 0, 4, 2, 6, 2, 5],
       [2, 5, 2, 6, 2, 6, 5, 2, 4, 2, 7, 4]])