--- license: mit --- Libraries needed: ``` import torch import torchvision import torchvision.transforms as transforms from tqdm import tqdm from torch import nn import matplotlib.pyplot as plt ``` to define a data loader ``` transformRes = transforms.Compose([ transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]), # transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]) ]) trainsetRes = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transformRes) trainloaderRes64 = torch.utils.data.DataLoader(trainsetRes, batch_size=64, shuffle=True, num_workers=10) testsetRes = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transformRes) testloaderRes64 = torch.utils.data.DataLoader(testsetRes, batch_size=64, shuffle=False, num_workers=10) ``` The model itself and training ``` import torchvision.models as models # Load the pretrained model from pytorch resnet50v2 = models.resnet50(pretrained=True) # Freeze the parameters of the model for param in resnet50v2.parameters(): param.requires_grad = True # Change the final layer to match the number of classes in the CIFAR-10 dataset num_ftrs = resnet50v2.fc.in_features resnet50v2.fc = nn.Sequential( nn.Linear(num_ftrs, 500), nn.ReLU(), nn.Linear(500, 200), nn.Dropout(0.5), nn.Linear(200,40), nn.ReLU(), nn.Dropout(0.3), nn.Linear(40,10), nn.ReLU() ) print("Model Info:") print("ResNet50,Pretrained,weight adj. LR=0.01,Mom=0.3,WD=0.0001") print("Schedule step=1,gamma=0.7, 20 epoches") # Move the model to the GPU resnet50v2 = resnet50v2.to(device, dtype=torch.float32) optimizer = torch.optim.SGD(resnet50v2.parameters(), lr=0.01,momentum=0.3,weight_decay=0.0001) criterion = nn.CrossEntropyLoss() scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.5) train_losses = [] test_losses = [] accuracies = [] train_acc = [] for epoch in range(20): # loop over the dataset multiple times running_loss = 0.0 correctTrain = 0 totalTrain = 0 pbar = tqdm(enumerate(trainloaderRes16, 0), total=len(trainloaderRes16), desc="Epoch {}".format(epoch+1)) for i, data in pbar: # get the inputs; data is a list of [inputs, labels] inputs, labels = data[0].to(device,dtype=torch.float32), data[1].to(device) # zero the parameter gradients optimizer.zero_grad() # forward + backward + optimize outputs = resnet50v2(inputs) loss = criterion(outputs, labels) loss.backward() optimizer.step() # print statistics running_loss += loss.item() _, predicted_train = torch.max(outputs.data, 1) totalTrain += labels.size(0) correctTrain += (predicted_train == labels).sum().item() pbar.set_postfix({'loss': running_loss/(i+1)}) train_accuracy = 100 * correctTrain / totalTrain train_acc.append(train_accuracy) print(f'Epoch {epoch + 1} loss: {running_loss / len(trainloaderRes16):.3f}') # Start of testing phase resnet50v2.eval() # Set the model to evaluation mode test_loss = 0.0 correct = 0 total = 0 with torch.no_grad(): for data in testloaderRes16: images, labels = data[0].to(device,dtype=torch.float32), data[1].to(device) outputs = resnet50v2(images) loss = criterion(outputs, labels) test_loss += loss.item() _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print(f'Epoch {epoch + 1} Test loss: {test_loss / len(testloaderRes16):.3f}, Accuracy: {100 * correct / total:.2f}%') #print the learning rate print(f'Epoch {epoch + 1} Learning rate: {optimizer.param_groups[0]["lr"]}') train_losses.append(running_loss / len(trainloaderRes16)) test_losses.append(test_loss / len(testloaderRes16)) accuracies.append(100 * correct / total) resnet50v2.train() # Set the model back to training model scheduler.step() print('Finished Training') plt.figure(figsize=(10, 5)) plt.plot(train_losses, label='Training Loss') plt.plot(test_losses, label='Test Loss') plt.xlabel('Epochs') plt.ylabel('Loss') plt.legend() plt.show() plt.figure(figsize=(10, 5)) plt.plot(accuracies, label='Accuracy') plt.plot(train_acc, label='Training Accuracy') plt.xlabel('Epochs') plt.ylabel('Accuracy (%)') plt.legend() plt.show() ```