【Pytorch】torch.einsum

torch.einsum(‘i,j->ij’, x, y)

Example

• 初始化两个Tensor，计算torch.einsum

x = torch.randn(5)
y = torch.randn(4)
z = torch.einsum('i,j->ij', x, y)

• x, y, z的值

x = tensor([ 0.3231, -1.5957, -1.0785, -3.3953,  0.8070])
y = tensor([ 0.3149,  0.1090,  0.1193, -0.4916])
z = tensor([[ 0.1018,  0.0352,  0.0385, -0.1589],
        [-0.5025, -0.1739, -0.1903,  0.7845],
        [-0.3396, -0.1176, -0.1286,  0.5302],
        [-1.0692, -0.3701, -0.4050,  1.6692],
        [ 0.2541,  0.0880,  0.0963, -0.3968]])

• x, y, z的shape

x.shape = torch.Size([5])
y.shape = torch.Size([4])
z.shape = torch.Size([5, 4])

理解

实现outer product功能
$$\begin{gathered} x=tensor([x_0,x_1,x_2,x_3,x_4]) \\ y=tensor([y_0,y_1,y_2,y_3]) \end{gathered}$$

$$z=\left[\begin{array}{ccccc}{x}_0{y}_0& x_0{y}_1& x_0{y}_2& x_0{y}_2& x_0{y}_3\\ x_1{y}_0& x_1{y}_1& x_1{y}_2& x_1{y}_2& x_1{y}_3\\ x_2{y}_0& x_2{y}_1& x_2{y}_2& x_2{y}_2& x_2{y}_3\\ x_3{y}_0& x_3{y}_1& x_3{y}_2& x_3{y}_2& x_3{y}_3\\ x_4{y}_0& x_4{y}_1& x_4{y}_2& x_4{y}_2& x_4{y}_3\end{array}\right]$$