hySAINT - Hybrid Genetic Algorithm and Simulated Annealing for Variable
Selection with Interaction Effects
A hybrid genetic algorithm combined with simulated
annealing for variable selection in high-dimensional linear
regression with two-way interaction effects. The package uses
an indexed model formulation for efficient storage and
minimizes the ABC (Adaptive Best-subset with Complexity)
criterion from Ye and Yang (2019)
<doi:10.1109/TIT.2019.2913417>. Features include: (1) support
for all three heredity conditions (strong, weak, and no
heredity), (2) adaptive temperature control in simulated
annealing, (3) LASSO-based screening for ultra-high-dimensional
problems, and (4) automatic sigma and lambda tuning. The
package scales to problems with thousands of predictors.
