I am modifying the 'train model' function below so to plot loss and accuracy graphs at every epochs during traning

def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
    since = time.time()

    best_model_wts = copy.deepcopy(model.state_dict())
    best_acc = 0.0
    losses=[]
    accuracies=[]

    y_loss = {}  # loss history
    y_loss['aug1_train'] = []
    y_loss['valid'] = []
    y_acc = {}
    y_acc['aug1_train'] = []
    y_acc['valid'] = []

    x_epoch = []
    fig = plt.figure()
    ax0 = fig.add_subplot(121, title="loss")
    ax1 = fig.add_subplot(122, title="accuracy")


    for epoch in range(num_epochs):
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        print('-' * 10)

        # Each epoch has a training and validation phase
        for phase in ['aug1_train', 'valid']:
            if phase == 'aug1_train':
                scheduler.step()
                model.train()  # Set model to training mode
            else:
                model.eval()   # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0

            # Iterate over data.
            for inputs, labels,paths in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)
               

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                # track history if only in train
                with torch.set_grad_enabled(phase == 'aug1_train'):
                    outputs = model(inputs)
                  
                    _, preds = torch.max(outputs, 1)
                    loss = criterion(outputs, labels)
                 

                    # backward   optimize only if in training phase
                    if phase == 'aug1_train':
                        loss.backward()
                        optimizer.step()

                # statistics
                running_loss  = loss.item() * inputs.size(0)
                running_corrects  = torch.sum(preds == labels.data)
                

            epoch_loss = running_loss / dataset_sizes[phase]
            epoch_acc = running_corrects.double() / dataset_sizes[phase]
          

            print('{} Loss: {:.4f} Acc: {:.4f} '.format(
                phase, epoch_loss, epoch_acc))
            
            y_loss[phase].append(epoch_loss)
            y_acc[phase].append(epoch_acc)


            # deep copy the model
            if phase == 'valid' and epoch_acc > best_acc:
                best_acc = epoch_acc
                best_model_wts = copy.deepcopy(model.state_dict())
                def draw_curve(current_epoch):
                    x_epoch.append(current_epoch)
                    ax0.plot(x_epoch, y_loss['aug1_train'], 'bo-', label='train')
                    ax0.plot(x_epoch, y_loss['valid'], 'ro-', label='val')
                    ax1.plot(x_epoch, y_acc['aug1_train'], 'bo-', label='train')
                    ax1.plot(x_epoch, y_acc['valid'], 'ro-', label='val')

                    if current_epoch == 0:
                        ax0.legend()
                        ax1.legend()

                    fig.savefig(os.path.join('/content/drive/My Drive/Stanford40/Graphs', 'train.jpg'))
                
                draw_curve(epoch)

            if phase=='aug1_train':
              losses.append(epoch_loss)
              accuracies.append(epoch_acc)


        print()

    time_elapsed = time.time() - since
    print('Training complete in {:.0f}m {:.0f}s'.format(
        time_elapsed // 60, time_elapsed % 60))
    print('Best val Acc: {:4f}'.format(best_acc))

    # load best model weights
    model.load_state_dict(best_model_wts)
    return model,losses,accuracies

and I load the Densenet161 for traning as below

#Load Pretrained Densenet161 model
model_ft = models.densenet161(pretrained=True)
model_ft.classifier=nn.Linear(2208,11)

model_ft = model_ft.to(device)

criterion = nn.CrossEntropyLoss()

# Observe that all parameters are being optimized
opt = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)

# Decay LR by a factor of 0.1 every 7 epochs
sched = lr_scheduler.StepLR(opt, step_size=5, gamma=0.1)

Finally I run the code below to start training:

model_ft,losses,accuracies = train_model(model_ft, criterion,opt ,sched,num_epochs=30)

and got this error as in the picture below:

How can I modify the code to get away from this error by using tensor.cpu() ?

CodePudding user response：

It's hard to say without a detailed error backtrace. It is vague but the information it gives is that something somewhere is detecting a tensor and trying to convert it to numpy array not properly. My intuition tells me it comes from the matplotlib code in your visualization step. I believe it is trying to convert your loss terms.

You should convert them to lists after having performed back propagation...

y_loss[phase].append(epoch_loss.item())
y_acc[phase].append(epoch_acc.item())

CodePudding user response：

What if try to get item() here

running_corrects  = torch.sum(preds == labels.data).item()

and remove double() when dividing?

epoch_acc = running_corrects / dataset_sizes[phase]