Skip to main content
. 2020 Oct 23;22(11):1198. doi: 10.3390/e22111198
Algorithm 3 Gene purging algorithm
Require:

  • Current Pareto front front

  • Evaluation function with ability to randomly shuffle features evalF

  • 1:

    functionpurgeParetoFront

  • 2:

        newFront = ← Dict.empty      ▹ Empty dictionary representing new Pareto Front

  • 3:

        for item in front do

  • 4:

            if |item| == 1 then

  • 5:

               continue            ▹ Items with only one feature cannot be reduced

  • 6:

            end if

  • 7:

            baseResult ← item.result        ▹ Score acquired by using all features from item

  • 8:

            scoreDecrease ← []             ▹ Score decrease for each feature

  • 9:

            for geneInd in item.genes WHERE ISSET(item.genes[geneInd]) do  ▹ Check only set genes

  • 10:

               newScore ← evalF(item.model, item.genes, geneInd)

    ▹ Evaluate existing trained model on test data set, where values of feature geneInd are randomly shuffled

  • 11:

               scoreDecrease[geneInd] ← baseResult - newScore

  • 12:

            end for

  • 13:

            bestInd ← argmin(scoreDecrease)

             ▹ Take the gene which showed smallest score decrease when shuffled

  • 14:

            newItem ← Item(genes=unsetBit(item.genes, bestInd))

            ▹ Create new item where the gene with the smallest score decrease is unset

  • 15:

            newFront[|newItem|] ← Evaluate(newItem, evalF)

                      ▹ Evaluate newly created item (train and test model)

  • 16:

        end for

  • 17:

        front ← front ∪ newFront       ▹ Update current front with newly calculated items

  • 18:

        front ←removeDominated(front)       ▹ Remove unoptimal items from Pareto front

  • 19:

        return front

  • 20:

    end function