Detailed explanation of the training and prediction process of deep learning [taking lenet model and cifar10 data set as examples]

Introduction to models and datasets

Model ：LeNet

Lenet It's a 7 Layer of neural network （ Does not include the input layer ）, contain 3 Convolution layers ,2 A pool layer ,2 All connection layers .
Use pytorch The structure is as follows ：

class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, 5)    # C1
        self.pool1 = nn.MaxPool2d(2, 2)     # S2
        self.conv2 = nn.Conv2d(16, 32, 5)   # C3
        self.pool2 = nn.MaxPool2d(2, 2)     # S4
        self.fc1 = nn.Linear(32*5*5, 120)   # C5（ Replace with full connection ）
        self.fc2 = nn.Linear(120, 84)       # F6
        self.fc3 = nn.Linear(84, 10)        # F7

    def forward(self, x):
        x = F.relu(self.conv1(x))    # input(3, 32, 32) output(16, 28, 28)
        x = self.pool1(x)            # output(16, 14, 14)
        x = F.relu(self.conv2(x))    # output(32, 10, 10)
        x = self.pool2(x)            # output(32, 5, 5)
        x = x.view(-1, 32*5*5)       # output(32*5*5)  Flattening 
        x = F.relu(self.fc1(x))      # output(120)
        x = F.relu(self.fc2(x))      # output(84)
        x = self.fc3(x)              # output(10)
        return x
        
model = LeNet()
print(model)

Data sets ：CIFAR10

Download address ：https://tensorflow.google.cn/datasets/catalog/cifar10
CIFAR10 Data set co ownership 60000 A color image , among 50000 Zhang is used for training ,5 Training batch , Every batch 10000 Pictures ;10000 Zhang for testing .
The size of the picture is 3X32X32, It is divided into 10 Categories , Each class 6000 Zhang .

Training process

The training of the model can be divided into the following steps ：

1. Dataset loading
2. Model loading
3. Iterative training
4. verification

The following is a detailed analysis combined with the code ：

1. Loading data and normalization

#  Data normalization 
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

#  Load training dataset 
#  Use it for the first time download Set to True To download the dataset automatically 
train_set = torchvision.datasets.CIFAR10(root='./data', train=True,
                                         download=False, transform=transform)
train_loader = torch.utils.data.DataLoader(train_set, batch_size=36,
                                           shuffle=True, num_workers=0)

#  Load test validation data set 
val_set = torchvision.datasets.CIFAR10(root='./data', train=False,
                                       download=False, transform=transform)
val_loader = torch.utils.data.DataLoader(val_set, batch_size=5000,
                                         shuffle=False, num_workers=0)
val_data_iter = iter(val_loader)
# val_image, val_label Respectively represent the original image and the label corresponding to the image （ Category ）
#  The training set should also be split during training 
val_image, val_label = val_data_iter.next()

2. Load model

#  Introducing models 
net = LeNet()
#  Define the loss function 
loss_function = nn.CrossEntropyLoss()
#  Define optimizer , Input model parameters and learning rate lr
optimizer = optim.Adam(net.parameters(), lr=0.001)

3. Iterative training

for epoch in range(100):  #  iteration 100 Time 

    running_loss = 0.0		#  Set the initial loss to 0
    for step, data in enumerate(train_loader, start=0):
        # inputs, labels Respectively represent the original image and the label corresponding to the image （ Category ）
        inputs, labels = data
        
        #  Every batch Set the gradient information to 0
        # （ You can have more than one batch  Call it once optimizer.zero_grad function . This is equivalent to increasing batch_size）
        optimizer.zero_grad()
        
        #  Transfer the original picture into the model , Get the prediction 
        outputs = net(inputs)
        #  Calculate the loss with the predicted result and the original label 
        loss = loss_function(outputs, labels)
        #  Back propagation 
        loss.backward()
        #  Update parameters 
        optimizer.step()

4. verification

for epoch in range(100):  #  iteration 100 Time 

    running_loss = 0.0		#  Set the initial loss to 0
    for step, data in enumerate(train_loader, start=0):
        # inputs, labels Respectively represent the original image and the label corresponding to the image （ Category ）
        inputs, labels = data
        #  Every batch Set the gradient information to 0
        # （ You can have more than one batch  Call it once optimizer.zero_grad function . This is equivalent to increasing batch_size）
        optimizer.zero_grad()
        #  Transfer the original picture into the model , Get the prediction 
        outputs = net(inputs)
        #  Calculate the loss with the predicted result and the original label 
        loss = loss_function(outputs, labels)
        #  Back propagation 
        loss.backward()
        #  Update parameters 
        optimizer.step()

		#  Calculate the total training loss 
        running_loss += loss.item()
        
    # The above is the training process , Verify from here 
	##########################################
    with torch.no_grad():		#  Verification is to stop calculating the gradient 
	    #  Pass in the original image of the validation set to the model , Get the prediction 
        outputs = net(val_image)  
        predict_y = torch.max(outputs, dim=1)[1]
        #  Calculate the same number of predicted values and tag values , And divide by the total , Get accuracy 
        accuracy = torch.eq(predict_y, val_label).sum().item() / val_label.size(0)
		#  Print training losses and accuracy 
        print('[%d, %5d] train_loss: %.3f test_accuracy: %.3f' %
              (epoch + 1, step + 1, running_loss / 500, accuracy))
        #  Reset loss is 0, Start the next iteration training 
        running_loss = 0.0
        
	print('Finished Training')
	#  End of training , Save the model 
    save_path = './Lenet.pth'
    torch.save(net.state_dict(), save_path)