SRMPInferenceApprox {MCDA}

Approximative inference of an SRMP model

Description

Approximative inference approach from pairwise comparisons of alternatives for the SRMP ranking model. This method outputs an SRMP model that fulfils as many pairwise comparisons as possible. Neither the number of reference profiles, nor the lexicographic order are fixed beforehand, however a maximum value for the number of reference profiles needs to be provided.

Usage

SRMPInferenceApprox(
  performanceTable,
  criteriaMinMax,
  maxProfilesNumber,
  preferencePairs,
  indifferencePairs = NULL,
  alternativesIDs = NULL,
  criteriaIDs = NULL,
  timeLimit = 60,
  populationSize = 20,
  mutationProb = 0.1
)

Arguments

performanceTable

Matrix or data frame containing the performance table. Each row corresponds to an alternative, and each column to a criterion. Rows (resp. columns) must be named according to the IDs of the alternatives (resp. criteria).

criteriaMinMax

Vector containing the preference direction on each of the criteria. "min" (resp. "max") indicates that the criterion has to be minimized (maximized). The elements are named according to the IDs of the criteria.

maxProfilesNumber

The maximum number of reference profiles of the SRMP model.

preferencePairs

A two column matrix containing on each row a pair of alternative names where the first alternative is considered to be strictly preferred to the second.

indifferencePairs

A two column matrix containing on each row a pair of alternative names the two alternatives are considered to indifferent with respect to each other.

alternativesIDs

Vector containing IDs of alternatives, according to which the datashould be filtered.

criteriaIDs

Vector containing IDs of criteria, according to which the data should be filtered.

timeLimit

Allows to fix a time limit of the execution, in seconds (default 60).

populationSize

Allows to change the size of the population used by the genetic algorithm (default 20).

mutationProb

Allows to change the mutation probability used by the genetic algorithm (default 0.1).

Value

The function returns a list containing:

criteriaWeights

The inferred criteria weights.

referenceProfilesNumber

The number of inferred reference profiles.

referenceProfiles

The inferred reference profiles.

lexicographicOrder

The inferred lexicographic order of the reference profiles.

fitness

The percentage of fulfilled pair-wise relations.

References

A-L. OLTEANU, V. MOUSSEAU, W. OUERDANE, A. ROLLAND, Y. ZHENG, Preference Elicitation for a Ranking Method based on Multiple Reference Profiles, forthcoming 2018.

Examples

performanceTable <- rbind(c(10,10,9),c(10,9,10),c(9,10,10),c(9,9,10),c(9,10,9),c(10,9,9),
                          c(10,10,7),c(10,7,10),c(7,10,10),c(9,9,17),c(9,17,9),c(17,9,9),
                          c(7,10,17),c(10,17,7),c(17,7,10),c(7,17,10),c(17,10,7),c(10,7,17),
                          c(7,9,17),c(9,17,7),c(17,7,9),c(7,17,9),c(17,9,7),c(9,7,17))

criteriaMinMax <- c("max","max","max")

rownames(performanceTable) <- c("a1","a2","a3","a4","a5","a6","a7","a8","a9","a10","a11",
                                "a12","a13","a14","a15","a16","a17","a18","a19","a20",
                                "a21","a22","a23","a24")

colnames(performanceTable) <- c("c1","c2","c3")

names(criteriaMinMax) <- colnames(performanceTable)

# expected result for the tests below

expectedpreorder <- list("a16","a13",c("a3","a9"),"a14","a17",c("a1","a7"),"a18","a15")

# test - preferences and indifferences

preferencePairs <- matrix(c("a16","a13","a3","a14","a17","a1","a18","a15","a2","a11",
                            "a5","a10","a4","a12","a13","a3","a14","a17","a1","a18",
                            "a15","a2","a11","a5","a10","a4","a12","a6"),14,2)
indifferencePairs <- matrix(c("a3","a1","a2","a11","a11","a20","a10","a10","a19","a12",
                              "a12","a21","a9","a7","a8","a20","a22","a22","a19","a24",
                              "a24","a21","a23","a23"),12,2)

set.seed(1)

result<-SRMPInferenceApprox(performanceTable, criteriaMinMax, 3, preferencePairs,
                            indifferencePairs, alternativesIDs = c("a1","a3","a7",
                            "a9","a13","a14","a15","a16","a17","a18"))