Right now the model doesn't do any preprocessing. Which really should be done to improve accuracy. 