普通邻接矩阵（类型torch.Tensor）转为torch.sparse_coo稀疏矩阵

普通矩阵：

adj <class 'torch.Tensor'> tensor([[10.0527,  0.7901,  4.7365,  ..., -2.9740, -1.2431,  2.6618],
        [ 0.7901,  7.8155,  2.0635,  ..., -0.7821,  0.8580, -1.8541],
        [ 4.7365,  2.0635,  6.2384,  ..., -2.2005, -0.2560,  2.0816],
        ...,
        [-2.9740, -0.7821, -2.2005,  ..., 18.2876, -2.9486,  0.2639],
        [-1.2431,  0.8580, -0.2560,  ..., -2.9486, 16.9452, -3.5839],
        [ 2.6618, -1.8541,  2.0816,  ...,  0.2639, -3.5839, 17.8889]],
       grad_fn=<MmBackward>)

变为COO稀疏矩阵：

idx = torch.nonzero(adj).T  
data = adj[idx[0],idx[1]]
adj_coo = torch.sparse_coo_tensor(idx, data, adj.shape)  # 转换成COO矩阵

coo矩阵：

adj_coo <class 'torch.Tensor'> tensor(indices=tensor([[   0,    0,    0,  ..., 2707, 2707, 2707],
                       [   0,    1,    2,  ..., 2705, 2706, 2707]]),
       values=tensor([10.0527,  0.7901,  4.7365,  ...,  0.2639, -3.5839,
                      17.8889]),
       size=(2708, 2708), nnz=7333264, layout=torch.sparse_coo,
       grad_fn=<SparseCooTensorWithDimsAndTensorsBackward>)