How does best first search generate feature subset in case of correlation based feature selection (CFS)?

Question

I am trying to get important subset of features using CFS. I understood most of the part of the algorithm, stuck on the part where feature subsets are generated for evaluation.