Symbol *, torch.max and torch.sum, item() methods

*The role can be referenced

Torch.max can refer to

a.size()
# Out[134]: torch.Size([6, 4, 3])
torch.max(a, 0)[1].size()
# Out[135]: torch.Size([4, 3])
torch.max(a, 1)[1].size()
# Out[136]: torch.Size([6, 3])
torch.max(a, 2)[1].size()
# Out[137]: torch.Size([6, 4])

Specific how to compare can see below

b

tensor([[[  0.,   1.,   2.,   3.],
         [  4.,   5.,   6.,   7.],
         [  8.,   9.,  10.,  11.]],
        [[ 12.,  13.,  14.,  15.],
         [ 16.,  17.,  18.,  19.],
         [ 20.,  21.,  22.,  23.]]])

torch.max(b,0)[0]
 
tensor([[ 12.,  13.,  14.,  15.],
        [ 16.,  17.,  18.,  19.],
        [ 20.,  21.,  22.,  23.]])

torch.max(b,1)[0]

tensor([[  8.,   9.,  10.,  11.],
        [ 20.,  21.,  22.,  23.]])

torch.max(b,2)[0]

tensor([[  3.,   7.,  11.],
        [ 15.,  19.,  23.]])

The corresponding subscript can be obtained

b

tensor([[[  0.,   1.,   2.,   3.],
         [  4.,   5.,   6.,   7.],
         [  8.,   9.,  10.,  11.]],
        [[ 12.,  13.,  14.,  15.],
         [ 16.,  17.,  18.,  19.],
         [ 20.,  21.,  22.,  23.]]])

torch.max(b,0)[1]

tensor([[ 1,  1,  1,  1],
        [ 1,  1,  1,  1],
        [ 1,  1,  1,  1]])

torch.max(b,1)[1]

tensor([[ 2,  2,  2,  2],
        [ 2,  2,  2,  2]])

torch.max(b,2)[1]

tensor([[ 3,  3,  3],
        [ 3,  3,  3]])

torch.sum:

torch.sum(input) → Tensor
torch.sum(input, dim, out=None) → Tensor
Parameters:

         Input (Tensor) – input tensor
         Dim (int) – reduced dimensions
         Out (Tensor, optional) – result tensor

The output of the function is a tensor

match
out:
tensor([[[ 0,  0,  2,  0],
         [ 0,  0,  0,  0],
         [ 0,  0,  0,  0]],
        [[ 0,  0,  0,  0],
         [ 0,  0,  0,  0],
         [ 0,  0,  0,  0]]], dtype=torch.uint8)

torch.sum(match)
Out: 
tensor(2)

torch.sum(match,0)
Out: 
tensor([[ 0,  0,  2,  0],
        [ 0,  0,  0,  0],
        [ 0,  0,  0,  0]])
torch.sum(match,1)
Out: 
tensor([[ 0,  0,  2,  0],
        [ 0,  0,  0,  0]])
torch.sum(match,2)
Out: 
tensor([[ 2,  0,  0],
        [ 0,  0,  0]])

One more thing to add is the use of the item method: if the tensor has only one element, then calling the item method will convert the tensor into scalar scalars; if the tensor is not a single element, then a ValueError will be raised, as below

b.item()
Traceback (most recent call last):
    b.item()
ValueError: only one element tensors can be converted to Python scalars

torch.sum(b)
Out: tensor(276.)
torch.sum(b).item()
Out: 276.0

So what is the item method in python? See

Intelligent Recommendation

Use the item symbol and number to organize text 1

Use the item symbol and number to organize text 1 Knowledge expansion Item Number: It is a point or other symbol before the text, which emphasizes the role. Reasonable use of the project number, you c...

[Pytorch function] Torch.sum

Torch.Sum () for a certain dimension of the input Tensor data, a total of two usage example: Output results: If you add Keepdim = True, you will keep the dimension of DIM is not SQUEEZE. Output result...

[Pytorch Basics] torch.sum ()

torch.sum()For a certain dimension of the input Tensor data, a total of two usage １．torch.sum(input, dtype=None) ２．torch.sum(input, list: dim, bool: keepdim=False, dtype=None) → Tensor input: Ent...

Torch.Sum, DIM parameter record

Xiaobai learned Pytorch, always confused the parameters of Dim, recorded Still small white, wrong place troubles Suppose already has a Tensor X, the dimension is [2, 4, 3], 2 is BATCH, 4 is the number...

Torch.Sum dimension parameter analysis

Personal understanding: For two-dimensional arrays, DIM = 0 is fixed line, add columns; DIM = 1, fixed column, add line. It is worth noting that DIM = -1 and DIM = 2 are the same. Personal estimation:...