1. """嵌套块访问参数"""
    2. def block1():
    3.     return  nn.Sequential(nn.Linear(4,8),nn.ReLU(),nn.Linear(8,4),nn.ReLU())
    4. def block2():
    5.     net =nn.Sequential()
    6.     for i in range(4):
    7.         net.add_module(f"block{i}",block1())
    8.     return net
    9. rgnet = nn.Sequential(block2(),nn.Linear(4,1))
    10. """Sequential(
    11.   (0): Sequential(
    12.     (block0): Sequential(
    13.       (0): Linear(in_features=4, out_features=8, bias=True)
    14.       (1): ReLU()
    15.       (2): Linear(in_features=8, out_features=4, bias=True)
    16.       (3): ReLU()
    17.     )
    18.     (block1): Sequential(
    19.       (0): Linear(in_features=4, out_features=8, bias=True)
    20.       (1): ReLU()
    21.       (2): Linear(in_features=8, out_features=4, bias=True)
    22.       (3): ReLU()
    23.     )
    24.     (block2): Sequential(
    25.       (0): Linear(in_features=4, out_features=8, bias=True)
    26.       (1): ReLU()
    27.       (2): Linear(in_features=8, out_features=4, bias=True)
    28.       (3): ReLU()
    29.     )
    30.     (block3): Sequential(
    31.       (0): Linear(in_features=4, out_features=8, bias=True)
    32.       (1): ReLU()
    33.       (2): Linear(in_features=8, out_features=4, bias=True)
    34.       (3): ReLU()
    35.     )
    36.   )
    37.   (1): Linear(in_features=4, out_features=1, bias=True)
    38. )
    40. Process finished with exit code 0
    41. """
    1. """内置初始化"""
    2. def init_normal(m):
    3.     if type(m)==nn.Linear:
    4.         nn.init.normal_(m.weight,mean=0,std=0.01)  # nn.init.constant(m.weight,1) 如果全初始化为1 等价于一个神经元
    5.         nn.init.zeros_(m.bias)
    6. net.apply(init_normal)
    7. print(net[0].weight.data[0])
    8. # tensor([ 0.0123,  0.0025, -0.0029,  0.0007])