1.3.0
Updates:
- Implemented problem-agnostic corrector algorithms to use if
all_discrete_x
is available - Switched to LHS as fallback sampler
- Bug fixes
ArchSBO-specific updates:
- Updated default settings for hidden constraint strategies: support KPLS and auto-scale random forest
- Updated default infill selection: only use ensembles
- Added gradient-based infill optimization after the evolutionary optimization
- Updated to SMT 2.2
- Do not force cont relax kernel for PLS for SMT>=2.2
- Added more logging