In pytorch, torch.unique is returning repititions

user_13486093 0
Viewed 42

I have this 2-D tensor:

tmp = torch.tensor([[ 0,  0,  0,  0,  1,  1,  1,  2,  2,  2,  3,  3,  3,  4,  4,  4,  5,  5,
          5,  6,  6,  6,  7,  7,  7,  8,  8,  8,  9,  9,  9, 10, 10, 10, 11, 11,
          11, 12, 12, 12, 13, 13, 13, 14, 14, 14, 15, 15, 15, 15, 16, 16, 16, 17,
          17, 17, 18, 18, 18, 19, 19, 19, 20, 20, 20, 21, 21, 21, 22, 22, 22, 23,
          23, 23, 24, 24, 24, 25, 25, 25, 26, 26, 26, 27, 27, 27, 28, 28, 28, 29,
          29, 29, 30, 30, 30, 31, 31, 31, 31],
        [ 0,  0,  0,  0,  1,  1,  1,  2,  2,  2,  3,  3,  3,  4,  4,  4,  5,  5,
          5,  6,  6,  6,  7,  7,  7,  8,  8,  8,  9,  9,  9, 10, 10, 10, 11, 11,
          11, 12, 12, 12, 13, 13, 13, 14, 14, 14, 15, 15, 15, 15,  0, 16, 16, 17,
          17, 17, 18, 18, 18, 19, 19, 19, 20, 20, 20, 21, 21, 21, 22, 22, 22, 23,
          23, 23, 24, 24, 24, 25, 25, 25, 26, 26, 26, 27, 27, 27, 28, 28, 28, 29,
          29, 29, 30, 30, 30, 31, 31, 31, 31]])

So there is 0 in the 50th column of row 2. When I apply torch.unique along dim=1, I get:

a,c = torch.unique(tmp,dim=1,return_counts=True)
a
tensor([[ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 16,
         17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31],
        [ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,  0, 16,
         17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]])

It can be seen that the second row of the output has two 0s and the first row has two 16s. Am I doing something wrong here or this is suspicious?

pytorch
1 Answers

It is because you specified dim=1. PyTorch is thus checking for unique pairs (which it correctly does). Like (0, 0), (1, 1), (16, 0): these are the unique pairs that it generated. In general the pair (temp[0,i], temp[1,i]) is unique for all i.

If you want all the elements to be unique, just throw away the dim: torch.unique(tmp).

If you need to maintain the two list structure, the output cannot be stored as a single tensor because their sizes might not match. You can do something like output1 = torch.unique(tmp[0]) and output2 = torch.unique(tmp[1]).

user_32374380
Related