This'll give you a chance to experiment and make sure everything works before spending more time training on the full dataset.