Where does the Sigmoid in Logistic Regression come from? (Rather, why log-odds?)