Concatenate and sort tensor

1

Hi,

I have 2 tensors like below:

hc1 = torch.randn(5,1, 1, 1)
hc2 = torch.randn(5,1, 1, 1)

I want to concatenate these 2 tensors as hc3 and then sort hc3 based on amounts of hc1.
How could I do it?

Thanks

2

Could you post a simple example with some values?
You could concatenate these tensor using out = torch.cat((hc1, hc2), dim=dim), however I’m not sure, how you would like to sort the tensors.

3

Dear ptrblck,

Thanks for your reply. Sure. I simplify my question with changing the dimensions of tensors to 2. Suppose we have these 2 tensors:

hc1 = tensor([[30],
              [20],
              [10],
              [50],
              [40]])

hc2=tensor([[1],
            [2],
            [3],
            [4],
            [5]])

1, 2, 3, 4, 5 are, subsequently, related to 30, 20, 10, 50, 40.
After sorting hc1 we would have 10, 20, 30, 40, 50 and 3, 2, 1, 5, 4. Because hc1 and hc2 are 1 to 1.
I thought that I should concatenate them as hc3 and sort it based on hc1.
I hope that I explained my question as well as possible.

Thanks

4

Hi,
You can get the permutation by calling perm = hc1.argsort(dim=0).squeeze().
In your example it returns tensor([2, 1, 0, 4, 3])

Then you can rearrange hc2 with this permutation : hc2_rearranged = hc2[perm].
In your example it returns

tensor([[3.],
        [2.],
        [1.],
        [5.],
        [4.]])

You can always concatenate those afterwards

5

Thanks for your reply.
I can not understand how to have sorted hc1 while you achieved sorted indices of hc1? Is there any way?

Thanks

6

Dear Phan_Phan,

Oh, how funny question I asked. It just needs to write:

print(hc1[perm])

Many thanks