An approach to visualizing the 3D empirical attainment function
Tea Tušar, Bogdan Filipič · 2013
When analyzing the performance of a bi-objective optimization algorithm, the empirical attainment function (EAF) is often used to visualize the attained parts of the objective space. Similarly, when comparing two algorithms, the differences in EAF values can be used to show the parts of the objective space in which the first algorithm outperforms the second one, and vice versa. This paper proposes to visualize the EAF values and differences also when assessing algorithms that optimize three criteria. This can be achieved by cutting through the 3D EAFs using multiple cutting planes and presenting the resulting intersections in 2D. The approach is described in detail and demonstrated on two artificial Pareto front approximations.