BSW 0.1.2
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.