torch.nn.Sequential

简单介绍

nn.Sequential是一个有序的容器，该类将按照传入构造器的顺序，依次创建相应的函数，并记录在Sequential类对象的数据结构中，同时以神经网络模块为元素的有序字典也可以作为传入参数。

因此，Sequential可以看成是有多个函数运算对象，串联成的神经网络，其返回的是Module类型的神经网络对象。

构建实例

参数列表

• 以参数列表的方式来实例化

print("利用系统提供的神经网络模型类：Sequential,以参数列表的方式来实例化神经网络模型对象")
# A sequential container. Modules will be added to it in the order they are passed in the constructor. 
# Example of using Sequential
model_c = nn.Sequential(
						nn.Linear(28*28, 32), 
						nn.ReLU(), 
						nn.Linear(32, 10), 
						nn.Softmax(dim=1)
					)
print(model_c)
 
print("\n显示网络模型参数")
print(model_c.parameters)
 
print("\n定义神经网络样本输入")
x_input = torch.randn(2, 28, 28, 1)
print(x_input.shape)
 
print("\n使用神经网络进行预测")
y_pred = model.forward(x_input.view(x_input.size()[0],-1))
print(y_pred)

利用系统提供的神经网络模型类：Sequential,以参数列表的方式来实例化神经网络模型对象
Sequential(
  (0): Linear(in_features=784, out_features=32, bias=True)
  (1): ReLU()
  (2): Linear(in_features=32, out_features=10, bias=True)
  (3): Softmax(dim=1)
)

显示网络模型参数
<bound method Module.parameters of Sequential(
  (0): Linear(in_features=784, out_features=32, bias=True)
  (1): ReLU()
  (2): Linear(in_features=32, out_features=10, bias=True)
  (3): Softmax(dim=1)
)>

定义神经网络样本输入
torch.Size([2, 28, 28, 1])

使用神经网络进行预测
tensor([[-0.1526,  0.0437, -0.1685,  0.0034, -0.0675,  0.0423,  0.2807,  0.0527,
         -0.1710,  0.0668],
        [-0.1820,  0.0860,  0.0174,  0.0883,  0.2046, -0.1609,  0.0165, -0.2392,
         -0.2348,  0.1697]], grad_fn=<AddmmBackward>)

字典

• 以字典的方式实例化

# Example of using Sequential with OrderedDict
print("利用系统提供的神经网络模型类：Sequential,以字典的方式来实例化神经网络模型对象")
model = nn.Sequential(OrderedDict([('h1', nn.Linear(28*28, 32)),
                                     ('relu1', nn.ReLU()),
                                     ('out', nn.Linear(32, 10)),
                                     ('softmax', nn.Softmax(dim=1))]))
 
print(model)
 
print("\n显示网络模型参数")
print(model.parameters)
 
print("\n定义神经网络样本输入")
x_input = torch.randn(2, 28, 28, 1)
print(x_input.shape)
 
print("\n使用神经网络进行预测")
y_pred = model.forward(x_input.view(x_input.size()[0],-1))
print(y_pred)

利用系统提供的神经网络模型类：Sequential,以字典的方式来实例化神经网络模型对象
Sequential(
  (h1): Linear(in_features=784, out_features=32, bias=True)
  (relu1): ReLU()
  (out): Linear(in_features=32, out_features=10, bias=True)
  (softmax): Softmax(dim=1)
)

显示网络模型参数
<bound method Module.parameters of Sequential(
  (h1): Linear(in_features=784, out_features=32, bias=True)
  (relu1): ReLU()
  (out): Linear(in_features=32, out_features=10, bias=True)
  (softmax): Softmax(dim=1)
)>

定义神经网络样本输入
torch.Size([2, 28, 28, 1])

使用神经网络进行预测
tensor([[0.1249, 0.1414, 0.0708, 0.1031, 0.1080, 0.1351, 0.0859, 0.0947, 0.0753,
         0.0607],
        [0.0982, 0.1102, 0.0929, 0.0855, 0.0848, 0.1076, 0.1077, 0.0949, 0.1153,
         0.1029]], grad_fn=<SoftmaxBackward>)

基本操作

• 查看结构

通过打印 Sequential对象来查看它的结构

print(net)
# Sequential(
#   (0): Linear(in_features=20, out_features=10, bias=True)
#   (1): ReLU()
#   (2): Linear(in_features=10, out_features=5, bias=True)
# )

• 索引

我们可以使用索引来查看其子模块

print(net[0])
# Linear(in_features=20, out_features=10, bias=True)
print(net[1])
# ReLU()

• 长度

print(len(net))
# 3

• 修改子模块

net[1] = nn.Sigmoid()
print(net)
# Sequential(
#   (0): Linear(in_features=20, out_features=10, bias=True)
#   (1): Sigmoid()
#   (2): Linear(in_features=10, out_features=5, bias=True)
# )

• 删除子模块

del net[2]
print(net)
# Sequential(
#   (0): Linear(in_features=20, out_features=10, bias=True)
#   (1): Sigmoid()
# )

• 添加子模块

net.append(nn.Linear(10, 2))  # 均会添加到末尾
print(net)
# Sequential(
#   (0): Linear(in_features=20, out_features=10, bias=True)
#   (1): Sigmoid()
#   (2): Linear(in_features=10, out_features=2, bias=True)
# )

• 遍历

net = nn.Sequential(
    nn.Linear(20, 10),
    nn.ReLU(),
    nn.Linear(10, 5)
)

for sub_module in net:
    print(sub_module)
# Linear(in_features=20, out_features=10, bias=True)
# ReLU()
# Linear(in_features=10, out_features=5, bias=True)

• 嵌套

'''在一个 Sequential 中嵌套两个 Sequential'''
seq_1 = nn.Sequential(nn.Linear(15, 10), nn.ReLU(), nn.Linear(10, 5))
seq_2 = nn.Sequential(nn.Linear(25, 15), nn.Sigmoid(), nn.Linear(15, 10))
seq_3 = nn.Sequential(seq_1, seq_2)
print(seq_3)
# Sequential(
#   (0): Sequential(
#     (0): Linear(in_features=15, out_features=10, bias=True)
#     (1): ReLU()
#     (2): Linear(in_features=10, out_features=5, bias=True)
#   )
#   (1): Sequential(
#     (0): Linear(in_features=25, out_features=15, bias=True)
#     (1): Sigmoid()
#     (2): Linear(in_features=15, out_features=10, bias=True)
#   )
# )
''''使用多级索引进行访问'''
print(seq_3[1])
# Sequential(
#   (0): Linear(in_features=25, out_features=15, bias=True)
#   (1): Sigmoid()
#   (2): Linear(in_features=15, out_features=10, bias=True)
# )
print(seq_3[0][1])
# ReLU()
'''使用双重循环进行遍历'''
for seq in seq_3:
    for module in seq:
        print(module)
# Linear(in_features=15, out_features=10, bias=True)
# ReLU()
# Linear(in_features=10, out_features=5, bias=True)
# Linear(in_features=25, out_features=15, bias=True)
# Sigmoid()
# Linear(in_features=15, out_features=10, bias=True)

参考

PyTorch学习笔记（六）–Sequential类、参数管理与GPU_Lareges的博客-CSDN博客_sequential类

[Pytorch系列-30]：神经网络基础 - torch.nn库五大基本功能：nn.Parameter、nn.Linear、nn.functioinal、nn.Module、nn.Sequentia_文火冰糖的硅基工坊的博客-CSDN博客_torch库nn