pytorch model

2024-02-20 09:08:04

网络定义
model.named_children 返回名字和操作
model.modules() 可用于参数初始化
其他的可以参考：

网络定义

import torch as torch
import torch.nn as nn
class LeNet(nn.Module):
    def __init__(self):
        super(LeNet,self).__init__()
        layer1 = nn.Sequential()
        layer1.add_module('conv1',nn.Conv2d(1,6,5))
        layer1.add_module('pool1',nn.MaxPool2d(2,2))
        self.layer1 = layer1

        layer2 = nn.Sequential()
        layer2.add_module('conv2',nn.Conv2d(6,16,5))
        layer2.add_module('pool2',nn.MaxPool2d(2,2))
        self.layer2 = layer2

        layer3 = nn.Sequential()
        layer3.add_module('fc1',nn.Linear(16*5*5,120))
        layer3.add_module('fc2',nn.Linear(120,84))
        layer3.add_module('fc3',nn.Linear(84,10))
        self.layer3 = layer3

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = x.view(x.size(0),-1)#转换（降低）数据维度，进入全连接层
        x = self.layer3(x)
        return x

#代入数据检验
y = torch.randn(1,1,32,32)
model = LeNet()
out = model(y)
print(model)
print(out)

输出如下：

LeNet(
  (layer1): Sequential(
    (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))
    (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (layer2): Sequential(
    (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
    (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (layer3): Sequential(
    (fc1): Linear(in_features=400, out_features=120, bias=True)
    (fc2): Linear(in_features=120, out_features=84, bias=True)
    (fc3): Linear(in_features=84, out_features=10, bias=True)
  )
)
tensor([[ 0.0211,  0.1407, -0.1831, -0.1182,  0.0221,  0.1467, -0.0523, -0.0663,
         -0.0351, -0.0434]], grad_fn=<AddmmBackward>)

model.named_children 返回名字和操作

print("*"*50)
for name, module in model.named_children():
    print(name)
    print(module)

打印如下：

layer1
Sequential(
  (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))
  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
layer2
Sequential(
  (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
layer3
Sequential(
  (fc1): Linear(in_features=400, out_features=120, bias=True)
  (fc2): Linear(in_features=120, out_features=84, bias=True)
  (fc3): Linear(in_features=84, out_features=10, bias=True)
)

可以用于forward，直接对输入遍历操作

  def forward(self, x):
        for name, module in self.named_children():
            x = module(x)

model.modules() 可用于参数初始化

print("#"*200)
cnt = 0
for name in model.modules():
    cnt += 1
    print('-------------------------------------------------------cnt=',cnt)
    print(name)

输出如下：

########################################################################################################################################################################################################
-------------------------------------------------------cnt= 1
LeNet(
  (layer1): Sequential(
    (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))
    (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (layer2): Sequential(
    (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
    (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (layer3): Sequential(
    (fc1): Linear(in_features=400, out_features=120, bias=True)
    (fc2): Linear(in_features=120, out_features=84, bias=True)
    (fc3): Linear(in_features=84, out_features=10, bias=True)
  )
)
-------------------------------------------------------cnt= 2
Sequential(
  (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))
  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
-------------------------------------------------------cnt= 3
Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))
-------------------------------------------------------cnt= 4
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
-------------------------------------------------------cnt= 5
Sequential(
  (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
-------------------------------------------------------cnt= 6
Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
-------------------------------------------------------cnt= 7
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
-------------------------------------------------------cnt= 8
Sequential(
  (fc1): Linear(in_features=400, out_features=120, bias=True)
  (fc2): Linear(in_features=120, out_features=84, bias=True)
  (fc3): Linear(in_features=84, out_features=10, bias=True)
)
-------------------------------------------------------cnt= 9
Linear(in_features=400, out_features=120, bias=True)
-------------------------------------------------------cnt= 10
Linear(in_features=120, out_features=84, bias=True)
-------------------------------------------------------cnt= 11
Linear(in_features=84, out_features=10, bias=True)

model.modules()主要用于参数初始化

cnt = 0
for name in model.modules():
    cnt += 1
    print('-------------------------------------------------------cnt=',cnt)
    print(name)
    if isinstance(name, nn.Conv2d):
        print('------------------isinstance(name, nn.Conv2d)------------------')
        print(name.weight)
        print(name.bias)
        print('--end----------------isinstance(name, nn.Conv2d)------------end------')

    if isinstance(name, nn.Conv2d):
        nn.init.kaiming_normal_(name.weight)
    elif isinstance(name, (nn.BatchNorm2d, nn.GroupNorm)):
        nn.init.constant_(name.weight, 1)
        nn.init.constant_(name.bias, 0)

其中参数部分输出如下：

------------------isinstance(name, nn.Conv2d)------------------
Parameter containing:
tensor([[[[-0.1561, -0.0194, -0.0260, -0.0042,  0.1716],
          [ 0.1181, -0.1380, -0.0448,  0.0674, -0.1972],
          [-0.0197,  0.0359,  0.1186,  0.0876, -0.0395],
          [-0.0619,  0.0095, -0.0702,  0.0122,  0.1573],
          [ 0.1170,  0.1758, -0.1655,  0.1489, -0.0956]]],
       ...
  [[[-0.1337, -0.0562, -0.0624,  0.0885, -0.0640],
          [-0.0302, -0.1192, -0.0637,  0.0083,  0.0181],
          [ 0.1388, -0.1690,  0.1132,  0.1686, -0.1189],
          [-0.0246, -0.1649, -0.1817, -0.0330, -0.0430],
          [ 0.0672, -0.0671,  0.0469,  0.1284,  0.1420]]]], requires_grad=True)
Parameter containing:
tensor([ 0.0548,  0.0547,  0.1328, -0.0452,  0.1668, -0.1915],
       requires_grad=True)
--end----------------isinstance(name, nn.Conv2d)------------end------

其他的可以参考：

https://blog.csdn.net/MrR1ght/article/details/105246412
model.children(): 返回模型的所有子模块的迭代器
model.modules():返回模型的所有模块（不仅仅是子模块，还包含当前模块）
model.named_children():返回当前子模块的迭代器。名字：模块
model.named_modules():

码农公寓

网络定义

model.named_children 返回名字 和 操作

model.modules() 可用于参数初始化

其他的可以参考：

相关文章

model.named_children 返回名字和操作