Imagine you have a huge dataset of billions of samples, and you learn f1 with logistic regression and f4 with polynomial logistic regression with p = 4. Do you think f1 or f4 will have lower reducible error? Explain your answer in a few sentences.