1
$\begingroup$

I have trained my deep learning model. I also saved the validation loss to a file and plotted on a graph

enter image description here

I have $2$ questions for this:

  • Does the validation loss look normal? Is there any issue with the validation loss ?
  • How can I determine which epoch is the best epoch so I can choose ?

If you need any additional information, I can provide. Thanks

$\endgroup$

1 Answer 1

4
$\begingroup$
  1. The first question is not well defined. What does normal mean? It is clearly decreasing, so that is good. If you are asking if the fluctuations are normal, yes that is not uncommon. This is not an unusual looking validation curve, but if you want a more complex answer, you need to look at how your loss is defined, what sort of values are reasonable, how it compares to the training loss, and how it compares to the test loss (assuming a three fold train test validate split).

  2. The simplest answer is pick the one with the lowest loss as a good heuristic. The more complex answer, is to get a less biased estimate by doing either a three fold or cross validation. For more info on those two and their tradoffs, see:

https://stats.stackexchange.com/questions/410118/cross-validation-vs-train-validation-test

$\endgroup$
3
  • $\begingroup$ Let's say we train for 700 epochs, and the lowest validation loss occurs at epoch 530, should we choose the the last epoch 700 or the lowest validation loss epoch 530 ? $\endgroup$
    – user78437
    Dec 1, 2023 at 5:51
  • 1
    $\begingroup$ in that case best to pick 530. Also please mark answer as correct if it was helpful $\endgroup$ Dec 1, 2023 at 5:58
  • $\begingroup$ Thanks for your help $\endgroup$
    – user78437
    Dec 2, 2023 at 17:36

You must log in to answer this question.