Sorting by NDS-rank and Hypervolume Contribution, with known points

Share:

Description

Sorts the large design for the purpose of multi objective optimization with SPOT. First non dominated sorting (NDS) rank is used. If the choice of points for the next sequential step is not clear by NDS rank, the hypervolume contribution of the competing points is recalculated sequentially to remove those with the smallest contribution.

Usage

1
2
spotMcoSelectionHypervol(largeDesign, designY, newsize, mergedX, mergedY,
  ref = NULL)

Arguments

largeDesign

the design matrix in the parameter space, to be sorted by the associated y-values for each objective

designY

objective value matrix. Contains objective values associated to largeDesign

newsize

this is the number of points that need to be selected, i.e. the seq.design.new.size

mergedX

position of the already known points in parameter space (vector of parameter values)

mergedY

y-values of the already known points (vector of objective values)

Details

In contrast to spotMcoSort, this function considers the known points in mergedX and mergedY so that new points will rather be chosen in between known points, thus producing a better Pareto front. To do so, the known points are added to the set of solutions. To ensure that they are not removed, they receive infinite hypervolume contribution, and are not counted when determining the number of NDS ranks to be considered.

Value

largeDesign
- The sorted large design

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