I have several classification models that used for image classification. The epochs is set to 100 for both. Model A gave me accuracy 99.7 and stopped at epoch 100 but Model B gave me 99.93 but take much time than Model A and the epochs exceeds the 100 until 150 epochs the code is training.

Should I need to stop training at 100 for the both model for fair comparing or make each one running until the accuracy improve. also the curve for the model A is enter image description here

and the curve for the model B is enter image description here

Should I need to cure the curve for the model B ?


1 Answer 1


It depends on the criteria you want to take into account when comparing your models. If your goal is to examine how fast a model is in training, you can stop here, but I guess it's not your objective.

To compare the accuracy, you should juxtapose models fairly. Currently, the range of y-axis labels is so different that we can't conclude whether it looks different in the last 10 epochs for Model A than Model B. Consider restricting your y-axis limits on both charts somewhere between 0 and 0.2.

Choosing a number of epochs might be perplexing. I can recommend automizing it to you. One of the common approaches is to stop training when your validation loss stops falling for a few epochs. If you are using TensorFlow, embrace the EarlyStopping mechanism.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.