1. 训练一个分类器

python">import torch.utils.data
import torchvision as tv
import torchvision.transforms as transforms
from torchvision.transforms import ToPILImage
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable# 定义网络结构
class myNet(nn.Module):def __init__(self):super(myNet, self).__init__()self.conv1 = nn.Conv2d(3,6,5)self.conv2 = nn.Conv2d(6,16,5)self.fc1 = nn.Linear(16*5*5, 120)  # 两次卷积、两次2*2的max_pool2d之后尺寸变成5*5self.fc2 = nn.Linear(120, 84)self.fc3 = nn.Linear(84, 10)def forward(self, x):x = F.max_pool2d(F.relu(self.conv1(x)), (2,2))x = F.max_pool2d(F.relu(self.conv2(x)), (2,2))x = x.view(x.size()[0], -1)x = F.relu(self.fc1(x))x = F.relu(self.fc2(x))x = self.fc3(x)return xnet = myNet()
#print(net)# 定义损失函数和优化器（loss and optimizer）
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr=0.01, momentum=0.9)# 预处理
transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5))])
# 加载训练集
trainset = tv.datasets.CIFAR10(root=r'D:\02.Work\06.LearnPyTorch\001\data', train=True, download=False, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=64,shuffle=True,num_workers=2)
# 加载验证集
testset = tv.datasets.CIFAR10(r'D:\02.Work\06.LearnPyTorch\001\data', train=False, download=False, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4,shuffle=False,num_workers=2)class_names = ('airplane', 'automobile', 'bird', 'cat',
'deer', 'dog', 'frog', 'horse', 'ship', 'truck')#(data, label) = trainset[100]
#print(class_names[label])
#print(type(data))show = ToPILImage()  # 可以把Tensor转成Image，方便可视化
if __name__ == '__main__':#img_forshow = show((data + 1)/2).resize((100,100))#img_forshow.show()#dataiter = iter(trainloader)#images, labels = next(dataiter)#imgs_forshow = show(tv.utils.make_grid((images + 1)/2)).resize((400,100))#imgs_forshow.show()# 训练for epoch in range(15):running_loss = 0.0for i, data in enumerate(trainloader, 0):# 输入数据inputs, labels = datainputs, labels = Variable(inputs), Variable(labels)if torch.cuda.is_available():net.cuda()inputs = inputs.cuda()labels = labels.cuda()# 每次前向都要清零梯度optimizer.zero_grad()# 前向传播、反向传播outputs = net(inputs)loss = criterion(outputs, labels)loss.backward()# 更新参数optimizer.step()running_loss += loss.item()# 打印lossif i % 2000 == 0:print('[epoch %d, step %d], loss = %f' % (epoch, i , running_loss / 2000))running_loss = 0.0print('Training finished.')# 测试一个test batch的结果dataiter = iter(testloader)images, labels = next(dataiter)print('groundtruth: ', ' '.join('%08s'%class_names[labels[j]] for j in range(4)))show(tv.utils.make_grid(images / 2 - 0.5)).resize((400, 100))if torch.cuda.is_available():images = images.cuda()outputs = net(Variable(images))_, predict = torch.max(outputs.data, 1)print('predicted results:', ' '.join('%5s'%class_names[predict[j]] for j in range(4)))# 测试整个测试集的效果correct = 0total = 0for data in testloader:images, labels = dataif torch.cuda.is_available():images = images.cuda()labels = labels.cuda()outputs = net(Variable(images))_, predict = torch.max(outputs.data, 1)total += labels.size(0)correct += (predict == labels).sum()print('1W张测试集中的准确率为: %d %%'%(100 * correct / total))print()

2. 训练一个线性回归

python"># y = 2*x + 3
import torch
import torch.nn as nn
from torch import optim
from matplotlib import pyplot as plt
#from IPython import displayclass LinearNet(nn.Module):def __init__(self):super(LinearNet, self).__init__()self.fc = nn.Linear(1, 1)def forward(self, input):return self.fc(input)#torch.manual_seed(1)def getFakeData(batch_size = 18):x = torch.rand(batch_size, 1)*30y = 2 * x + 3 + (torch.randn(batch_size, 1))  # 尾部是随机噪声return x,yif __name__ == '__main__':net = LinearNet()print(net.state_dict().keys())  # 打印所有参数名称# 随机初始化的参数print("随机初始化的参数：")print('net.fc.weight = ', net.fc.weight.detach().numpy())print('net.fc.bias = ', net.fc.bias.detach().numpy())criterion = nn.MSELoss()optimizer = optim.SGD(net.parameters(), lr=1e-3)for i in range(20000):x, y = getFakeData()y_pred = net(x)loss = criterion(y, y_pred)optimizer.zero_grad()loss.backward()optimizer.step()print('*'*30)print("训练后的参数：")print('net.fc.weight = ', net.fc.weight.detach().numpy())print('net.fc.bias = ', net.fc.bias.detach().numpy())if 10:torch.save(net, 'net.pth')   # 不推荐，因为这种保存方式依赖模型定义及文件路径结构等b = torch.load('net.pth')print('b.fc.weight = ', b.fc.weight.detach().numpy())print('b.fc.bias = ', b.fc.bias.detach().numpy())print('*'*30)if 1:torch.save(net.state_dict(), 'net2.pth')net2 = LinearNet()c = net2.load_state_dict(torch.load('net2.pth'))print('net2.fc.weight = ', net2.fc.weight.detach().numpy())print('net2.fc.bias = ', net2.fc.bias.detach().numpy())

输出：

odict_keys(['fc.weight', 'fc.bias'])
随机初始化的参数：
net.fc.weight =  [[0.37508917]]
net.fc.bias =  [0.81224954]
******************************
训练后的参数：
net.fc.weight =  [[2.0011635]]
net.fc.bias =  [2.9967082]
b.fc.weight =  [[2.0011635]]
b.fc.bias =  [2.9967082]
******************************
net2.fc.weight =  [[2.0011635]]
net2.fc.bias =  [2.9967082]