This is a very simple package containing base R code for Logistic Regression. It implements binary logistic regression where the optimization is performed for two cost functions:
- Negative Log-Likelihood, corresponding to doing MLE.
- Negative Log-Likelihood plus a (ridge) regularization term. This is equivalent to do Maximum-A-Posteriori with an isotropic Gaussian prior. Notice that in this specific implementation regularization is done also on the intercept.
In both cases, 2 implementation methods are implemented in base R:
- Gradient Ascient (using a learning rate)
- Newton's Method (again using a learning rate)
A third optimization method is possible via a call to the optim function in R. The method is customizable, although it is
"BFGS" by default.