pagmo2
pagmo2 copied to clipboard
More MO optimization performance metrics
It would be useful to have some more performance metrics for MO optimization.
For instance, some that are often used in MO papers are:
- [ ] Diversity of the non-dominated front
- [ ] Closeness to the true Pareto-optimal front
These two can used only when the true Pareto optimal front is known (as it is the case for many test-suites such as zdt
, dtlz
, etc.).
The aforementioned two metrics are described in: Deb, K.: Multi-Objective Optimization using Evolutionary Algorithms, John Wiley & Sons, Chichester, UK (2001).
These metrics are a bit outdated, but I would propose to implement some of the quality indicators from:
Coello Coello and Falcon-Cardona,: Convergence and Diversity Analysis of Indicator-based Multi-Objective Evolutionary Algorithms, in GECCO 2019.
These metrics are a bit outdated, but I would propose to implement some of the quality indicators from: Coello Coello and Falcon-Cardona,: Convergence and Diversity Analysis of Indicator-based Multi-Objective Evolutionary Algorithms, in GECCO 2019.
Yes, those metrics are Pareto-compliant. Please don't add metrics that are not Pareto-compliant. Proper performance assessment for MO optimization is explained in these works:
- https://iridia-ulb.github.io/references/#ZitThiLauFon2003:tec
- https://iridia-ulb.github.io/references/#KnoThiZit06:tutorial
- https://iridia-ulb.github.io/references/#ZitKnoThi2008quality
Notable advances in metrics since those papers were published are:
- Averaged Hausdorff distance: https://iridia-ulb.github.io/references/#SchEsqLarCoe2012tec
- IGD+ : https://iridia-ulb.github.io/references/#IshMasTanNoj2015igd
- Weighted hypervolume: https://iridia-ulb.github.io/references/#AugBadBroZit2009gecco
You can find implementations (in C) of IGD+ and Average Hausdorff in https://mlopez-ibanez.github.io/eaf/