training_with_early_stopping.py

# Early stopping details
n_epochs_stop = 5
min_val_loss = np.Inf
epochs_no_improve = 0

# Main loop
for epoch in range(n_epochs):
  # Initialize validation loss for epoch
  val_loss = 0
  
  # Training loop
  for data, targets in trainloader:
    # Generate predictions
    out = model(data)
    # Calculate loss
    loss = criterion(out, targets)
    # Backpropagation
    loss.backward()
    # Update model parameters
    optimizer.step()
    
# Validation loop
for data, targets in validloader:
  # Generate predictions 
  out = model(data)
  # Calculate loss
  loss = criterion(out, targets)
  val_loss += loss

# Average validation loss
val_loss = val_loss / len(trainloader)

# If the validation loss is at a minimum
if val_loss < min_val_loss:
  # Save the model
  torch.save(model, checkpoint_path)
  epochs_no_improve = 0
  min_val_loss = val_loss
  
else:
  epochs_no_improve += 1
  # Check early stopping condition
  if epochs_no_improve == n_epochs_stop:
    print('Early stopping!')
    
    # Load in the best model
    model = torch.load(checkpoint_path)

Is the indentation correct on this?? It seems like it goes through the entire epoch loop before entering the validation loop.

Yes, the validation runs once per epoch. This is generally how I've seen early stopping implemented; if the validation loss does not decrease for 5 epochs, the training stops and we use the best model.