MultiObjectiveMetrics.jl
Performance indicators for comparing approximations of the Pareto front produced by multiobjective optimization solvers. All objectives are minimized, and every indicator is implemented from its primary source.
| Indicator | Function | Better | Source |
|---|---|---|---|
| Purity | purity | higher | [1, 2] |
| Γ-spread | gamma_spread | lower | [2] |
| Δ-spread | delta_spread | lower | [2, 3] |
| Covering | covering | higher | [4, 5] |
| Hypervolume | hypervolume, normalized_hypervolume | higher | [4, 6] |
They rest on dominance with an explicit tolerance (Tolerance, compare) and on non-dominated filtering (nondominated). The package has no runtime dependencies and accepts any vector of objective vectors, including the columns or rows of a matrix.
- Installation and Quick Start: get started.
- Point sets and tolerance: input formats and comparisons.
- Performance profiles: comparing several solvers on many problems.
- Conventions: edge cases and differences from other implementations.
- Metrics: Purity, Spread metrics, Covering, Hypervolume.
- API Reference, Acknowledgements and References.
For a survey of performance indicators, see [7].