pytorch 邻接矩阵转稀疏矩阵 (dense matrix to coo matrix)

如何将一个dense矩阵，比如：一个 $N\times N$ 的邻接矩阵a 转成 pytorch的sparse coo 矩阵。

方法一：

import numpy as np
import torch

a = np.array([[0, 1.2, 0],[2, 3.1, 0],[0.5, 0, 0]])
idx = a.nonzero() # (row, col)
data = a[idx]

# to torch tensor
idx_t = torch.LongTensor(np.vstack(idx))
data_t = torch.FloatTensor(data)
coo_a = torch.sparse_coo_tensor(idx_t, data_t, a.shape)
print(coo_a)

方法二：

import scipy.sparse as sp
import numpy as np
import torch

a = np.array([[0, 1.2, 0],[2, 3.1, 0],[0.5, 0, 0]])
coo_np = sp.coo_matrix(a)
data = coo_np.data
idx_t = torch.LongTensor(np.vstack((coo_np.row, coo_np.col)))
data_t = torch.FloatTensor(data)
coo_a=torch.sparse_coo_tensor(idx_t,data_t,a.shape)

方法三（从tensor到sparse tensor）：

import numpy as np
import torch

a = torch.tensor([[0, 1.2, 0],[2, 3.1, 0],[0.5, 0, 0]])
idx = torch.nonzero(a).T  # 这里需要转置一下
data = a[idx[0],idx[1]]

coo_a = torch.sparse_coo_tensor(idx, data, a.shape)
print(coo_a)

如果想转回dense的tensor：coo_a.to_dense()