NEWS

---

BSW 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.