torch Sparse COO tensors; COO format; coordinate format

https://pytorch.org/docs/stable/sparse.html#sparse-coo-docs

PyTorch implements the so-called Coordinate format, or COO format, as one of the storage formats for implementing sparse tensors.

In COO format, the specified elements are stored as tuples of element indices and the corresponding values. In particular,

the indices of specified elements are collected in indices tensor of size (ndim, nse) and with element type torch.int64,

the corresponding values are collected in values tensor of size (nse,) and with an arbitrary integer or floating point number element type,

where ndim is the dimensionality of the tensor and nse is the number of specified elements.

举例：

Suppose we want to define a sparse tensor with the entry 3 at location (0, 2), entry 4 at location (1, 0), and entry 5 at location (1, 2). Unspecified elements are assumed to have the same value, fill value, which is zero by default. We would then write:

>>> i = [[0, 1, 1],
         [2, 0, 2]]
>>> v =  [3, 4, 5]
>>> s = torch.sparse_coo_tensor(i, v, (2, 3))
>>> s
tensor(indices=tensor([[0, 1, 1],
                       [2, 0, 2]]),
       values=tensor([3, 4, 5]),
       size=(2, 3), nnz=3, layout=torch.sparse_coo)
>>> s.to_dense()
tensor([[0, 0, 3],
        [4, 0, 5]])

Suppose we want to create a (2 + 1)-dimensional tensor with the entry [3, 4] at location (0, 2), entry [5, 6] at location (1, 0), and entry [7, 8] at location (1, 2). We would write

>>> i = [[0, 1, 1],
         [2, 0, 2]]
>>> v =  [[3, 4], [5, 6], [7, 8]]
>>> s = torch.sparse_coo_tensor(i, v, (2, 3, 2))
>>> s
tensor(indices=tensor([[0, 1, 1],
                       [2, 0, 2]]),
       values=tensor([[3, 4],
                      [5, 6],
                      [7, 8]]),
       size=(2, 3, 2), nnz=3, layout=torch.sparse_coo)
>>> s.to_dense()
tensor([[[0, 0],
         [0, 0],
         [3, 4]],
        [[5, 6],
         [0, 0],
         [7, 8]]])