Orthant search is "one code to rule them all" for many operations in multiobjective evolutionary algorithms.
This repository accompanies the paper:
Buzdalov M. Generalized Offline Orthant Search: One Code for Many Problems in Multiobjective Optimization // Proceedings of Genetic and Evolutionary Computation. - 2018. - P. 593-600.
The following reductions to orthant search are implemented and tested:
- Domination count (the number of points which a given point dominates), used in SPEA and SPEA2.
- Domination rank (the number of points that dominate a given point), used in MOGA and VEGA.
- Non-dominated sorting (used in NSGA-II, NSGA-III and many other algorithms).
- A "buggy" version of non-dominated sorting that assigns increasing ranks to several identical solutions.
- The additive binary epsilon-indicator (used mainly in assessing the performance of multiobjective optimization algorithms).
- Initial fitness assignment for the IBEA algorithm (the version that uses the additive binary epsilon-indicator).
- NEW: the R2 indicator, including the very recent version with an arbitrary power (typically equal to the dimension) applied to the addends.
The following contributors would like to acknowledge the support of this research by the Russian Scientific Foundation, agreement 17-71-20178: