Reshaping Tensor: Basic

1

Hi,
If we have a linear tensor of 24 elements from 0 to 23.
X = [0 1 2 3 …23]
How can we reshape it to something like that
Y = [[0,1 2 3 12 13 14 15],
[4 5 6 7 16 17 18 19],
[8 9 10 11 20 21 22 23]]

Thanks

2 
x = torch.range(1,24)
print (x)
x = x.view(3,-1)
print (x)


Output:
tensor([[ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.],
        [ 9., 10., 11., 12., 13., 14., 15., 16.],
        [17., 18., 19., 20., 21., 22., 23., 24.]])
3

Well The issue is we don’t want it [1 2 3 4 5 6 7 8]
we want it to be [0 1 2 3 12 13 14 15]
and so on… there is gap. See expected output given in the question

4

Well you can first split your original tensor in half, reshape both halves and concatenate them again:

# initilizing list 
X = list(range(0, 24))

# converting list to array 
X = np.array(X) 

# converting array to tensor
X = torch.from_numpy(X)

# split tensor in half
X_split = torch.split(X, int(X.shape[0]/2), dim=0)

# reshape the two tensors
X_left = X_split[0].view(3,-1)
X_right = X_split[1].view(3,-1)

# concatenate the reshaped tensors again
X = torch.cat((X_left, X_right), dim=1)

That wil give you:

tensor([[ 0,  1,  2,  3, 12, 13, 14, 15],
        [ 4,  5,  6,  7, 16, 17, 18, 19],
        [ 8,  9, 10, 11, 20, 21, 22, 23]])
2 Likes
5

Ohh I didn’t notice. Sorry about that. Chris (@vdw) has just posted a solution. I was about to write the same. See if that works for you.