Overfitting or underfitting

the concept of overfitting and undercutting is still quite confusing to me. Is this plot of training/validation loss below overfitting or underfitting?

I would claim it’s neither as the validation loss is still “close” to the training loss, doesn’t change a lot and is thus also quite noisy. Plot both curves into the same plot and the gap should be small, if I interpret the values correctly.

Thanks for your response. This is both plots together. So can this be considered a “reasonable” loss and means the model is actually learning?

The model reduced the loss initially at least and might not be stuck on a plateau. You might want to continue with the training to see if the loss would still decrease or not.

1 Like

There is also learning schedulers that may be useful later on


Im using pytorch lightning EarlyStopping with a patience of 8, so I guess the training stopped when the validation loss wasn’t improving?

Is this similar to using Early stopping with pytorch lightning?

I try to keep deps to the minimum and haven’t used lightining, but I think it is not. Early Stopping does not change the learning rate, but stops the training process when the validation loss is stagnant.