Improvd downward branch and bound algorithm for regression variable selection

Improved downward branch and bound to select the best subset for least squares regression problems.

Subset (feature) selection for least squares regression is a common problem, which is combinartorial, hence is computationally NP hard. This code provides a tool using the improved downward branch and bound approach to solve this problem efficiently. One of the applications of this algorithm is to select globally optimal controlled variables for self-optimizing control.