Plotting to two sets of boxes instead of one

[jbh1128d1]

Hello,

I copied the script from this GIT account and end up getting testing and training in two separate plots instead of together and the legend in both say 'training'. See script and output below to spot any mistakes I'm missing:

max_trn_batch =  1000
print_interval = 100

def train_model(model, criterion, optimizer, num_epochs=10):
    liveloss = PlotLosses()

    for epoch in range(num_epochs):
        logs = {}
        for phase in ['train', 'test']:
            if phase == 'train':
                model.train()
            else:
                model.eval()

            running_loss = 0.0
            running_corrects = 0

            for b, (image, label, policy, categorical_data) in enumerate(dataloaders[phase]):
                image = image.cuda()
                label = label.cuda()
                #numerical_data = numerical_data.cuda()
                categorical_data = categorical_data.cuda()

                outputs = model(image, categorical_data)
                loss = criterion(outputs, label)

                if phase == 'train':
                    optimizer.zero_grad()
                    loss.backward()
                    optimizer.step()

                _, preds = torch.max(outputs, 1)
                running_loss += loss.detach() * image.size(0)
                running_corrects += torch.sum(preds == label.data)

                #count batches
                b += 1
                #throttle the batches
                if b % print_interval == 0:
                    print(epoch, b)
                if b == max_trn_batch:
                    break

            epoch_loss = running_loss / len(dataloaders[phase].dataset)
            epoch_acc = running_corrects.float() / len(dataloaders[phase].dataset)

            prefix = ''
            if phase == 'test':
                prefix = 'test_'

            logs[prefix + 'log loss'] = epoch_loss.item()
            logs[prefix + 'accuracy'] = epoch_acc.item()

        liveloss.update(logs)
        liveloss.draw()

train_model(model = combined_model, criterion = criterion, optimizer = optimizer, num_epochs=500)

[stared]

Use val_ or in creating PlotLosses object: PlotLosses(series_fmt={'training': '{}', 'test':'test_{}'}).

[jbh1128d1]

I did the above:

def train_model(model, criterion, optimizer, num_epochs=10):
    liveloss = PlotLosses(series_fmt={'training': '{}', 'test':'test_{}'})
    max_accuracy = 0
    for epoch in range(num_epochs):
        logs = {}
        for phase in ['train', 'test']:
            if phase == 'train':
                model.train()
            else:
                model.eval()

            running_loss = 0.0
            running_corrects = 0

            for b, (image, label, policy, categorical_data) in enumerate(dataloaders[phase]):
                image = image.cuda()
                label = label.cuda()
                #numerical_data = numerical_data.cuda()
                categorical_data = categorical_data.cuda()

                outputs = model(image, categorical_data)
                loss = criterion(outputs, label)

                if phase == 'train':
                    optimizer.zero_grad()
                    loss.backward()
                    optimizer.step()

                _, preds = torch.max(outputs, 1)
                running_loss += loss.detach() * image.size(0)
                running_corrects += torch.sum(preds == label.data)

                b += 1
                if b % print_interval == 0:
                    print(epoch, b)

                if b == max_trn_batch:
                    break

            epoch_loss = running_loss / len(dataloaders[phase].dataset)
            epoch_acc = running_corrects.float() / len(dataloaders[phase].dataset)

            prefix = ''
            if phase == 'test':
                prefix = 'test_'
                if epoch_acc > max_accuracy:
                    max_accuracy = epoch_acc
                    torch.save(combined_model.state_dict()
                               , 'D:\\CIS inspection images 0318\\self_build\\combined_model_1.pt') 
                #if es.step(epoch_acc):
                    #break

            else:
                prefix = 'train_'

            logs[prefix + 'log loss'] = epoch_loss.item()
            logs[prefix + 'accuracy'] = epoch_acc.item()
            print(logs)

        liveloss.update(logs)
        liveloss.draw()

    print(max_accuracy)
    scheduler.step(loss)

Still the same issue. Same with `val_'