0

I'm training an LSTM model. I'm confusing about the validation loss of the model. Which value better represents the validation loss of the model? Is it the last value I obtain in the floowing loop, or I should calculate the mean of all the history values?

This is my model

for epoch in range(n_epochs):
        lstm.train()
        outputs = lstm.forward(X_train) # forward pass
        optimiser.zero_grad() # calculate the gradient, manually setting to 0
        # obtain the loss function
        loss = loss_fn(outputs, y_train)
        loss.backward() # calculates the loss of the loss function
        optimiser.step() # improve from loss, i.e backprop
        train_loss_history.append(loss.item())
        # test loss
        lstm.eval()
        test_preds = lstm(X_test)
        MSEtest_loss = loss_fn(test_preds, y_test)
        val_loss_history.append(MSEtest_loss.item())
        if epoch % 100 == 0:
            print("Epoch: %d, train loss: %1.5f, val MSE loss: %1.5f " % (epoch, 
                                                                      loss.item(), 
                                                                      MSEtest_loss.item(),
                                                                      ))

Now, is the last value of MSEtest_loss.item() represents the validation loss of the model? or Should I calculate the val_loss_history to represent the validation loss of the model ?

Mouna Ahmen
  • 25
  • 3
  • Because this question is less concerned with programming and more with classification methodology and/or training parameters it is better suited for the [Data Science](https://datascience.stackexchange.com/). Please read the relevant guidance related to how to properly ask questions on this site before posting your question. – itprorh66 Jun 03 '23 at 12:51

0 Answers0