torch. roll

x = torch.arange(1,10).view(3,3)
print(x,"\n")
print(torch.roll(x,shifts=1,dims=0)) #shift  Indicates the number of offsets  ,dim Indicates which dimension to follow   here dim=0 Said line 
print(torch.roll(x,shifts=1,dims=0)==torch.roll(x,shifts=4,dims=0)) # so shift It can be counted circularly , With 3 For cycles 

# The parameter can be a combination , It means to go along the ground successively dim 0 ,1 Sequential migration 1 That's ok ,1 Column  , Be careful ：shift Is the overall cycle offset , Does not replace the value of the specified position 
torch.roll(x,shifts=(1,1),dims=(0,1))

Output

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

tensor([[7, 8, 9],
        [1, 2, 3],
        [4, 5, 6]])
        
tensor([[True, True, True],
        [True, True, True],
        [True, True, True]])
tensor([[9, 7, 8],
        [3, 1, 2],
        [6, 4, 5]])

x = torch.arange(1,10).view(3,3)
print(x,"\n")
print(torch.roll(x,shifts=1,dims=0)) #shift  Positive and negative indicate positive offset   And reverse offset  
print(torch.roll(x,shifts=-1,dims=0))

Output

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

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