torch.nn.Module.apply(fn)
# 遞歸的調(diào)用weights_init函數(shù),遍歷nn.Module的submodule作為參數(shù)
# 常用來對(duì)模型的參數(shù)進(jìn)行初始化
# fn是對(duì)參數(shù)進(jìn)行初始化的函數(shù)的句柄,fn以nn.Module或者自己定義的nn.Module的子類作為參數(shù)
# fn (Module -> None) – function to be applied to each submodule
# Returns: self
#　Return type: Module

例子:

def weights_init(m):
 classname = m.__class__.__name__
 if classname.find('Conv') != -1:
  m.weight.data.normal_(0.0, 0.02) 
  # m.weight.data是卷積核參數(shù), m.bias.data是偏置項(xiàng)參數(shù)
 elif classname.find('BatchNorm') != -1:
  m.weight.data.normal_(1.0, 0.02)
  m.bias.data.fill_(0)

netG = _netG(ngpu)　# 生成模型實(shí)例
netG.apply(weights_init)　# 遞歸的調(diào)用weights_init函數(shù),遍歷netG的submodule作為參數(shù)

#-*-coding:utf-8-*-
import torch
from torch.autograd import Variable

#　對(duì)模型參數(shù)進(jìn)行初始化
# 官方論壇鏈接：https://discuss.pytorch.org/t/weight-initilzation/157/3

# 方法一
# 單獨(dú)定義一個(gè)weights_init函數(shù),輸入?yún)?shù)是m(torch.nn.module或者自己定義的繼承nn.module的子類)
#　然后使用net.apply()進(jìn)行參數(shù)初始化
#　m.__class__.__name__　獲得nn.module的名字
#　https://github.com/pytorch/examples/blob/master/dcgan/main.py#L90-L96
def weights_init(m):
 classname = m.__class__.__name__
 if classname.find('Conv') != -1:
  m.weight.data.normal_(0.0, 0.02)
 elif classname.find('BatchNorm') != -1:
  m.weight.data.normal_(1.0, 0.02)
  m.bias.data.fill_(0)

netG = _netG(ngpu)　# 生成模型實(shí)例
netG.apply(weights_init)　# 遞歸的調(diào)用weights_init函數(shù),遍歷netG的submodule作為參數(shù)

# function to be applied to each submodule

# 方法二
# 1. 使用net.modules()遍歷模型中的網(wǎng)絡(luò)層的類型 2. 對(duì)其中的m層的weigth.data(tensor)部分進(jìn)行初始化操作
# Another initialization example from PyTorch Vision resnet implementation.
# https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py#L112-L118
class ResNet(nn.Module):
 def __init__(self, block, layers, num_classes=1000):
  self.inplanes = 64
  super(ResNet, self).__init__()
  self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
        bias=False)
  self.bn1 = nn.BatchNorm2d(64)
  self.relu = nn.ReLU(inplace=True)
  self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
  self.layer1 = self._make_layer(block, 64, layers[0])
  self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
  self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
  self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
  self.avgpool = nn.AvgPool2d(7, stride=1)
  self.fc = nn.Linear(512 * block.expansion, num_classes)
  #　權(quán)值參數(shù)初始化
  for m in self.modules():
   if isinstance(m, nn.Conv2d):
    n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
    m.weight.data.normal_(0, math.sqrt(2. / n))
   elif isinstance(m, nn.BatchNorm2d):
    m.weight.data.fill_(1)
    m.bias.data.zero_()

# 方法三
# 自己知道網(wǎng)絡(luò)中參數(shù)的順序和類型, 然后將參數(shù)依次讀取出來,調(diào)用torch.nn.init中的方法進(jìn)行初始化
net = AlexNet(2)
params = list(net.parameters()) #　params依次為Conv2d參數(shù)和Bias參數(shù)
# 或者
conv1Params = list(net.conv1.parameters())
# 其中,conv1Params[0]表示卷積核參數(shù), conv1Params[1]表示bias項(xiàng)參數(shù)
# 然后使用torch.nn.init中函數(shù)進(jìn)行初始化
torch.nn.init.normal(tensor, mean=0, std=1)
torch.nn.init.constant(tensor, 0)

# net.modules()迭代的返回: AlexNet,Sequential,Conv2d,ReLU,MaxPool2d,LRN,AvgPool3d....,Conv2d,...,Conv2d,...,Linear,
# 這里,只有Conv2d和Linear才有參數(shù)
# net.children()只返回實(shí)際存在的子模塊: Sequential,Sequential,Sequential,Sequential,Sequential,Sequential,Sequential,Linear

# 附AlexNet的定義
class AlexNet(nn.Module):
 def __init__(self, num_classes = 2): # 默認(rèn)為兩類，貓和狗
#   super().__init__() # python3
  super(AlexNet, self).__init__()
  # 開始構(gòu)建AlexNet網(wǎng)絡(luò)模型，5層卷積，3層全連接層
  # 5層卷積層
  self.conv1 = nn.Sequential(
   nn.Conv2d(in_channels=3, out_channels=96, kernel_size=11, stride=4),
   nn.ReLU(inplace=True),
   nn.MaxPool2d(kernel_size=3, stride=2),
   LRN(local_size=5, bias=1, alpha=1e-4, beta=0.75, ACROSS_CHANNELS=True)
  )
  self.conv2 = nn.Sequential(
   nn.Conv2d(in_channels=96, out_channels=256, kernel_size=5, groups=2, padding=2),
   nn.ReLU(inplace=True),
   nn.MaxPool2d(kernel_size=3, stride=2),
   LRN(local_size=5, bias=1, alpha=1e-4, beta=0.75, ACROSS_CHANNELS=True)
  )
  self.conv3 = nn.Sequential(
   nn.Conv2d(in_channels=256, out_channels=384, kernel_size=3, padding=1),
   nn.ReLU(inplace=True)
  )
  self.conv4 = nn.Sequential(
   nn.Conv2d(in_channels=384, out_channels=384, kernel_size=3, padding=1),
   nn.ReLU(inplace=True)
  )
  self.conv5 = nn.Sequential(
   nn.Conv2d(in_channels=384, out_channels=256, kernel_size=3, padding=1),
   nn.ReLU(inplace=True),
   nn.MaxPool2d(kernel_size=3, stride=2)
  )
  # 3層全連接層
  # 前向計(jì)算的時(shí)候，最開始輸入需要進(jìn)行view操作，將3D的tensor變?yōu)?D
  self.fc6 = nn.Sequential(
   nn.Linear(in_features=6*6*256, out_features=4096),
   nn.ReLU(inplace=True),
   nn.Dropout()
  )
  self.fc7 = nn.Sequential(
   nn.Linear(in_features=4096, out_features=4096),
   nn.ReLU(inplace=True),
   nn.Dropout()
  )
  self.fc8 = nn.Linear(in_features=4096, out_features=num_classes)

 def forward(self, x):
  x = self.conv5(self.conv4(self.conv3(self.conv2(self.conv1(x)))))
  x = x.view(-1, 6*6*256)
  x = self.fc8(self.fc7(self.fc6(x)))
  return x

  pretrained_dict = torch.load(“model.pth”)
  model_dict = et.state_dict()
  pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
  model_dict.update(pretrained_dict)
  net.load_state_dict(model_dict)