Hi everyone,
I have come across multiple examples that illustrate the working of a CNN foe classification tasks. However, there is very little out there that actually illustrates how a CNN can be modified for a regression task, particularly a ordinal regression tasks that can have outputs in the range of 0 to 4.
I understand that this problem can be treated as a classification problem by employing the cross entropy loss. Although, I think MSELoss() would work better since you would prefer a 0 getting miss-classified as a 1 rather than a 4.
I use the torchvision pre trained model for this task and then use the CrossEntropy loss.
model = models.resnet18(pretrained = True)
fc_in_features = model.fc.in_features
model.fc = nn.Linear(fc_in_features,5)
# DEFINE A FUNCTION TO TRAIN THE MODEL
def train_model(model, dataloaders, criterion, optimizer, lr_scheduler,model_path, num_epochs=25):
since = time.time()
val_acc_history = []
val_loss_history = []
train_acc_history = []
train_loss_history = []
best_model_wts = copy.deepcopy(model.state_dict())
best_acc = 0.0
for epoch in range(1, num_epochs+1):
print('Epoch {}/{}'.format(epoch, num_epochs))
print('-' * 10)
# Each epoch has a training and validation phase
for phase in ['train', 'val']:
if phase == 'train':
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_corrects = 0
all_preds = []
all_labels = []
# Iterate over data.
for inputs, labels 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 == 'train'):
# Get model outputs and calculate loss
outputs = model(inputs)
loss = criterion(outputs, labels)
# Get model predictions
_, preds = torch.max(outputs, 1)
# backward + optimize only if in training phase
if phase == 'train':
loss.backward()
optimizer.step()
# statistics
running_loss += loss.item() * inputs.size(0)
running_corrects += torch.sum(preds == labels.data)
all_preds.append(preds)
all_labels.append(labels)
epoch_loss = running_loss / len(dataloaders[phase].sampler)
epoch_acc = running_corrects.double() / len(dataloaders[phase].sampler)
all_labels = torch.cat(all_labels, 0)
all_preds = torch.cat(all_preds, 0)
lr_scheduler.step(epoch_loss)
print('{} Loss: {:.4f} - Acc: {:.4f} '.format(phase, epoch_loss, epoch_acc))
# deep copy the model
if phase == 'val' and epoch_acc > best_acc:
best_acc = epoch_acc
best_model_wts = copy.deepcopy(model.state_dict())
if phase == 'val':
val_acc_history.append(epoch_acc)
val_loss_history.append(epoch_loss)
if phase == 'train':
train_acc_history.append(epoch_acc)
train_loss_history.append(epoch_loss)
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, train_loss_history, val_loss_history, train_loss_history, val_acc_history
# TRAIN THE NETWORK
EPOCHS = 20
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.00001)
lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, patience=3,verbose = True)
model, train_loss_history, val_loss_history, train_acc_history, val_acc_history = train_model(model=model, dataloaders=data_loaders, criterion=criterion,optimizer = optimizer,lr_scheduler=lr_scheduler, num_epochs=EPOCHS)
I am confused how to convert this code so as to use MSELoss or L1SmoothLoss. If I merely change the loss function, the dimensions do not match.
The solution I tried was to convert the labels into one hot encoding and to add a Softmax function to the output layer of the network. This then gives model outputs and targets of the similar shape. And then I use torch.argmax() to convert the output and targets into the range 0-4. However the model only predicts one class always which is weird.
I would want to understand 4 things:
- Do we convert the labels into one hot encodings in this case?
- Do we have to make changes to the networks output layer to be able to use MSELoss or L1SmoothLoss?
- How do we handle the mismatch of the dimensions when we use these losses?
- Assuming that we apply these loss functions, how do we convert the output of the model to the range 0-4 so as to calculate the accuracy.
Thank you in advance. I hope that this discussion can finally outline a clear pipeline to use for regression tasks.