One can perform glm model fit with logistic link function for response from binomial family and then can apply step function to extract the best subset of explanatory features in terms of information criterion such as AIC or BIC.
Have anyone performed such procedure for an output model from cv.glmnetfunction or glmnet from glmnet package? Can you suggest how one can reproduce the same methodology used on regular glm model and cv.glmnet model?
This below doesn't look like it would work:
modelAIC <-step( object= model$glmnet.fit, direction = "backward")
