topk(num,dim)_topk=[1, 3]

topk(num,dim=1)

>>> output=torch.randn(3,4)
>>> output
tensor([[-1.9291,  1.4127, -2.2464,  0.8932],
        [-0.4483, -0.3458,  0.8384,  1.9580],
        [-0.5633, -2.2806,  0.6278,  1.3552]])
在行上取一个最大值
>>> topkv,topki=output.topk(1,1)
>>> topkv
tensor([[1.4127],
        [1.9580],
        [1.3552]])
>>> topki
tensor([[1],
        [3],
        [3]])
在行上取前两个最大值
>>> topkv,topki=output.topk(2,1)
>>> topkv
tensor([[1.4127, 0.8932],
        [1.9580, 0.8384],
        [1.3552, 0.6278]])
>>> topki
tensor([[1, 3],
        [3, 2],
        [3, 2]])


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

topk(num,dim=0)

>>> output=torch.randn(3,4)
>>> output
tensor([[-1.9291,  1.4127, -2.2464,  0.8932],
        [-0.4483, -0.3458,  0.8384,  1.9580],
        [-0.5633, -2.2806,  0.6278,  1.3552]])
在列上取一个最大值        
>>> topkv,topki=output.topk(1,0)
>>> topkv
tensor([[-0.4483,  1.4127,  0.8384,  1.9580]])
>>> topki
tensor([[1, 0, 1, 1]])
在列上取两个最大值
>>> topkv,topki=output.topk(2,0)
>>> topkv
tensor([[-0.4483,  1.4127,  0.8384,  1.9580],
        [-0.5633, -0.3458,  0.6278,  1.3552]])
>>> topki
tensor([[1, 0, 1, 1],
        [2, 1, 2, 2]])
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19