Hypervolume Lower Confidence Bound Infill Criterion

Share:

Description

This multi objective infill criterion is similar to the SMS-EGO infill criterion by Ponweiser (2008). It aggregates the objective values for each point by calculating the hypervolume contribution. As a first step the lower confidence bound is calculated, decreasing the predicted objective values by their predicted variance. Unlike SMS-EGO, epsilon dominance is not employed here. Also, the penalties for dominated points are calculated differently: The hypervolume between the dominated points and the current true Pareto front is used.

Usage

1
spotInfillLcbHyperVolume(resy, resvar, y, ref = NULL)

Arguments

resy

predicted objective values

resvar

predicted variance

y

the current Pareto front

ref

reference point, if not given will be chosen as maximum of observed values plus one

Value

returns the contribution (or penalty) for each row in resy

Note

An optimizer like pso will work signif. better than cmaes with this infill criterion.

References

W. Ponweiser, T. Wagner, D. Biermann, and M. Vincze. Multiobjective optimization on a limited budget of evaluations using model-assisted -metric selection. In PPSN, pages 784-794, 2008.

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.