记：关于Pytorch中Linear结构与参数权重查看

代码段：

import torch
x = torch.ones(1,3)
y = torch.nn.Linear(3,3,bias=True)
print(x)
print(y)
out = y.forward(x)
print(out)
print(y.state_dict().keys())   # 查看有哪些参量
print(y.weight)                # 输出weight参量
print(y.bias)                  # 输出bias参量

运行结果：

tensor([[1., 1., 1.]])
Linear(in_features=3, out_features=3, bias=True)
tensor([[ 0.0577, -0.6237, -0.2998]], grad_fn=<AddmmBackward>)
odict_keys(['weight', 'bias'])
Parameter containing:
tensor([[ 0.0177,  0.4276, -0.1507],
        [-0.0755, -0.2118,  0.1403],
        [ 0.1560, -0.3528, -0.4536]], requires_grad=True)
Parameter containing:
tensor([-0.2369, -0.4766,  0.3506], requires_grad=True)