It seems like it probably should be removed (it seems like you'll always want to train on all of the training data).