Pytorch构建网络模型四步骤：

​①利用pytorch进行深度学习

• 准备数据集（Prepare dataset）
• 设计用于计算最终结果的模型（Design model）
• 构造损失函数及优化器（Construct loss and optimizer）
• 设计循环周期（Training cycle）——前馈、反馈、更新
  

1. import torch
   
2. from torchvision import transforms
   
3. from torchvision import datasets
   
4. from torch.utils.data import DataLoader
   
5. import torch.nn.functional as F
   
6. import torch.optim as optim
   
7. import matplotlib.pyplot as plt
   
8. 
   
9. # prepare dataset
   
10. 
    
11. batch_size = 64
    
12. transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])
    
13. 
    
14. train_dataset = datasets.MNIST(root='../dataset/mnist/', train=True, download=True, transform=transform)
    
15. train_loader = DataLoader(train_dataset, shuffle=True, batch_size=batch_size)
    
16. test_dataset = datasets.MNIST(root='../dataset/mnist/', train=False, download=True, transform=transform)
    
17. test_loader = DataLoader(test_dataset, shuffle=False, batch_size=batch_size)
    
18. 
    
19. 
    
20. # design model using class
    
21. 
    
22. 
    
23. class Net(torch.nn.Module):
    
24.     def __init__(self):
    
25.         super(Net, self).__init__()
    
26.         self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
    
27.         self.conv2 = torch.nn.Conv2d(10, 20, kernel_size=5)
    
28.         self.pooling = torch.nn.MaxPool2d(2)
    
29.         self.fc = torch.nn.Linear(320, 10)
    
30. 
    
31.     def forward(self, x):
    
32.         # flatten data from (n,1,28,28) to (n, 784)
    
33. 
    
34.         batch_size = x.size(0)
    
35.         x = F.relu(self.pooling(self.conv1(x)))
    
36.         x = F.relu(self.pooling(self.conv2(x)))
    
37.         x = x.view(batch_size, -1)  # -1 此处自动算出的是320
    
38.         # print("x.shape",x.shape)
    
39.         x = self.fc(x)
    
40. 
    
41.         return x
    
42. 
    
43. 
    
44. model = Net()
    
45. device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    
46. model.to(device)
    
47. 
    
48. # construct loss and optimizer
    
49. criterion = torch.nn.CrossEntropyLoss()
    
50. optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)
    
51. 
    
52. 
    
53. # training cycle forward, backward, update
    
54. 
    
55. 
    
56. def train(epoch):
    
57.     running_loss = 0.0
    
58.     for batch_idx, data in enumerate(train_loader, 0):
    
59.         inputs, target = data
    
60.         inputs, target = inputs.to(device), target.to(device)
    
61.         optimizer.zero_grad()
    
62. 
    
63.         outputs = model(inputs)
    
64.         loss = criterion(outputs, target)
    
65.         loss.backward()
    
66.         optimizer.step()
    
67. 
    
68.         running_loss += loss.item()
    
69.         if batch_idx % 300 == 299:
    
70.             print('[%d, %5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / 300))
    
71.             running_loss = 0.0
    
72. 
    
73. 
    
74. def test():
    
75.     correct = 0
    
76.     total = 0
    
77.     with torch.no_grad():
    
78.         for data in test_loader:
    
79.             images, labels = data
    
80.             images, labels = images.to(device), labels.to(device)
    
81.             outputs = model(images)
    
82.             _, predicted = torch.max(outputs.data, dim=1)
    
83.             total += labels.size(0)
    
84.             correct += (predicted == labels).sum().item()
    
85.     print('accuracy on test set: %d %% ' % (100 * correct / total))
    
86.     return correct / total
    
87. 
    
88. 
    
89. if __name__ == '__main__':
    
90.     epoch_list = []
    
91.     acc_list = []
    
92. 
    
93.     for epoch in range(10):
    
94.         train(epoch)
    
95.         acc = test()
    
96.         epoch_list.append(epoch)
    
97.         acc_list.append(acc)
    
98. 
    
99.     plt.plot(epoch_list, acc_list)
    
100.     plt.ylabel('accuracy')
     
101.     plt.xlabel('epoch')
     
102.     plt.show()
     

复制代码