Changes in version 0.1.2 (2025-10-06)                  

New

  - Added option conswitch in bsw():
      - 1 (default): all possible min/max combinations of predictors
        (bounds predictions).
      - 0: raw design matrix (suitable for risk factor identification
        only).
  - New function obj_value() to compute the log-likelihood value used in
    the Armijo line search.
  - Implemented a robust bootstrap procedure bootBSW() for bsw objects.
    The function completes execution even if individual bootstrap
    samples fail to converge.
  - Implemented variable selection methods (variable_selection_bsw()),
    supporting both backward elimination and forward selection. User can
    specify significance level (alpha) and maximum iterations (maxit).

Improved

  - Optimized Hessian matrix calculation by replacing explicit loops
    with vectorized operations.
  - Armijo line search implemented to ensure stable convergence.