Concatenate Tensors without Loop

1

I have a tensor t that I would like to transform. Therefore I made up a for loop that yields the desired result:

t = torch.tensor([[ 1,  2,  3], [ 4,  5,  6], [ 7,  8,  9], [10, 11, 12]])
N, d = t.shape
new = torch.zeros((N, N, 2*d))
for i in range(N):
    for j in range(N):
        new[i, j, :d] = t[i]
        new[i, j, d:] = t[j]
new

>>>
tensor([[[ 1.,  2.,  3.,  1.,  2.,  3.],
         [ 1.,  2.,  3.,  4.,  5.,  6.],
         [ 1.,  2.,  3.,  7.,  8.,  9.],
         [ 1.,  2.,  3., 10., 11., 12.]],

        [[ 4.,  5.,  6.,  1.,  2.,  3.],
         [ 4.,  5.,  6.,  4.,  5.,  6.],
         [ 4.,  5.,  6.,  7.,  8.,  9.],
         [ 4.,  5.,  6., 10., 11., 12.]],

        [[ 7.,  8.,  9.,  1.,  2.,  3.],
         [ 7.,  8.,  9.,  4.,  5.,  6.],
         [ 7.,  8.,  9.,  7.,  8.,  9.],
         [ 7.,  8.,  9., 10., 11., 12.]],

        [[10., 11., 12.,  1.,  2.,  3.],
         [10., 11., 12.,  4.,  5.,  6.],
         [10., 11., 12.,  7.,  8.,  9.],
         [10., 11., 12., 10., 11., 12.]]])

Do the for loops cause any errors with the gradient calculations? Is there any other way that concatenates the tensors in the same way without any for loops?

2

You could repeat the tensor t and use torch.cat to create the new tensor:

t_new = torch.cat((t.repeat(4, 1, 1).transpose(0, 1), t.repeat(4, 1, 1)), 2)
1 Like
3

Hi Ptrblck

I get my result but i need to concatenated predicted and target for 10-folds I use torch.stack but can not concatenate 3 tensors in a loop. I need concatenate them for each loop vertically to use them.Indeed at the end ROCTotal should be (xx,2), which ROCTotal[:,0] should be all predicted value and ROCTotal[:,1] all targeted value. It works for two tensors but not for concatenating in a loop. the size of Predictedvalue and Target is (106,1) which 106 can be different for each loop

ROCTotal=tensot.Torch()

for ii in range(1,11,1)

ROCTotalFolds=torch.stack([ROCTotalFolds,torch.from_numpy(AveragePredicted),torch.from_numpy(Target1)], dim=0)
ROCTotalFolds=torch.cat([ROCTotalFolds,torch.from_numpy(AveragePredicted),torch.from_numpy(Target1)], 1)

4

Could you post an example code using random tensors with the right shapes, so that we can have a look at it, as Iā€™m currently unsure how all tensors are defined.

5

Dear ptrblck

I solved the problem