The link function is the logistic function e^log(odds)/ (1+e^log(odds)) The default regularization for logistic regression is L2. It is basically like Ridge regression with the penalty being the sum of the squares of the coefficients. Liblinear and Saga kernels also allow for L1 penalty.