Early stopping Criteria on Valid set or Training set?

1

I just need to know how to decide early stopping criteria do I use early stopping on Validation set or training set.

Like:-
Training set is not converging but validation set still converging or Vice versa ??

2

Usually you stop when the training loss remains constant over the time.

1 Like
3

So my training loss became constant for last 4-5 epochs but my validation set still converging.I can stop it.?

4

Yeah you should stop it. Because validation set performance should be checked once you stopped the training.

1 Like
5

That doesn’t seem correct to me. You measure both validation, and training performance during the training. Ideally, you should stop once validation loss converges, i.e., doesn’t decrease for a few epochs.
Then, you measure the final performance on a test set.