Repeat training over epochs

Use iteration for one mini-batch update and epoch for one full pass through the training data. You’ll reason why many repeated updates usually improve predictions more safely than one giant change.