In lesson 4 Jeremy discuss how to chose which data augmentation and how much of it to use (to try each data augmentation type separately with 3-4 different values and choose the best one).
I didn’t understand how you decide which of the 3-4 options are the best one.
Is it the one with the smallest difference between training and validation accuracy? The one with the best results on the validation? Or maybe the one that reduce the training accuracy by something reasonable?
1 Like
Best results on validation accuracy, just realize you may be fitting your validation set if you try too many options.
1 Like
I would go with best validation accuracy. However note you have to reset the model each time you train otherwise it is not comparable.