I didn’t quite understand why a random sample from the training dataset was taken, normalized, and then fit the image generator which was then used to create the data generators for the validation and test set. Coul someone clarify the reasoning for this?