Consider the following two approaches to avoiding overfitting when applying logistic regression to a machine learning problem: (1) do feature selection prior to running logistic regression, as we did with the Gini impurity method or (2) use a regularized version of logistic regression, so that the error function being minimized isn't just the cross entropy error but (cross entropy error) + (norm of weight vector). Briefly compare and contrast these two methods, not just in terms of their effect on overfitting, but in other ways one method is different or better than the other.