inst/examples/exampleFLOSS.R

In MCDA: Support for the Multicriteria Decision Aiding Process

#################################
#------------ Install ----------#
#################################

# check that glpk is already on your systen, else the command below will fail
install.packages("MCDA")

# if available, install cplexAPI too - here the path is for the standard location of CPLEX 12.8 - update based on your local configuration
install.packages("cplexAPI",configure.args = "--with-cplex-dir=/opt/ibm/ILOG/CPLEX_Studio128/cplex")

#################################
#------------- Init ------------#
#################################

# load the library
library(MCDA)

# set up criteria parameters
criteria <- c("c1","c2","c3","c4","c5")
criteriaMinMax <- c("max","max","max","max","max")
names(criteriaMinMax) <- criteria

# set up categories parameters
categories <- c("Good","Neutral","Bad")
categoriesRanks <- c(1,2,3)
names(categoriesRanks) <- categories

# load performance table
f <- system.file("datasets", "dataFLOSS.csv", package = "MCDA")
pT <- read.csv(f, TRUE, ",", "\"", ".")
pT <- data.matrix(pT)
rownames(pT) <- 1:dim(pT)[1]
colnames(pT) <- criteria

#################################
#-------- First iteration-------#
#################################

# get assignments of first 25 alternatives
assig <-c("Bad", "Neutral", "Bad", "Bad", "Neutral","Good",
          "Neutral", "Good", "Bad", "Bad", "Good", "Neutral",
          "Bad", "Bad", "Bad", "Bad", "Bad", "Neutral",
          "Bad", "Neutral", "Bad", "Bad", "Neutral", "Bad",
          "Bad")
names(assig) <- 1:25

# generate MR-Sort model
m1 <- MRSortInferenceExact(pT, assig, categoriesRanks, 
                           criteriaMinMax, veto = FALSE,
                           readableWeights = TRUE, 
                           readableProfiles = TRUE,
                           solver = "cplex",
                           alternativesIDs = names(assig))

# check if we found an optimal solution and that the model fits
print(m1)
# it does not

# generate incompatible assignments
incomp <- MRSortIdentifyIncompatibleAssignments(pT, assig,
                                                categoriesRanks,
                                                criteriaMinMax,
                                                veto = FALSE,
                                                solver = "cplex",
                                                alternativesIDs = names(assig))

# generate alternative assignments for each set of incompatible assignments
for(incAlt in incomp$incompatibleSets)
{
  m1pr <- MRSortInferenceExact(pT, assig, categoriesRanks,
                               criteriaMinMax, veto = FALSE,
                               readableWeights = TRUE,
                               readableProfiles = TRUE,
                               alternativesIDs = names(assig)[!names(assig) %in% incAlt],
                               solver = "cplex")
  
  print(MRSort(pT, m1pr$profilesPerformances, categoriesRanks, m1pr$weights, criteriaMinMax,
         m1pr$lambda, alternativesIDs = incAlt))
}

# DM expresses posibility to update alternative 2 but not the rest -> trying MR-Sort with vetoes

m1 <- MRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                           veto = TRUE, readableWeights = TRUE, 
                           readableProfiles = TRUE, solver = "cplex",
                           alternativesIDs = names(assig))

# model fits
print(m1)

# plot the model
criteriaLBs=c(-2,-2,-2,-2,-2)
names(criteriaLBs) <- criteria
criteriaUBs=c(2,2,2,2,2)
names(criteriaUBs) <- criteria

plotMRSortSortingProblem(pT, m1$profilesPerformances, categoriesRanks, assig,
                         criteriaMinMax, criteriaUBs, criteriaLBs, 
                         NULL, m1$vetoPerformances, 'V',
                         m1$weights, m1$lambda,
                         alternativesIDs = "no alternatives")

#################################
#------- Second iteration-------#
#################################

# add assignments for the next 10 alternatives
# the performance table already contains their evaluations, however we filter the input to our algorithmsusing assig
assig <-c(assig, "Bad", "Bad", "Neutral", "Bad", "Bad",
          "Bad", "Bad", "Neutral", "Good", "Neutral")
names(assig) <- 1:35

# generate MR-Sort model with vetoes
m2 <- MRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                           veto = TRUE, readableWeights = TRUE, 
                           readableProfiles = TRUE, solver = "cplex",
                           alternativesIDs = names(assig))

# model does not fit
print(m2)

# generate incompatible assignments
incomp <- MRSortIdentifyIncompatibleAssignments(pT, assig, categoriesRanks,
                                                criteriaMinMax, veto = TRUE,
                                                solver = "cplex",
                                                alternativesIDs = names(assig))

# generate alternative assignments for each set of incompatible assignments
for(incAlt in incomp$incompatibleSets)
{
  m2pr <- MRSortInferenceExact(pT, assig, categoriesRanks,
                               criteriaMinMax, veto = TRUE,
                               readableWeights = TRUE,
                               readableProfiles = TRUE,
                               alternativesIDs = names(assig)[!names(assig) %in% incAlt],
                               solver = "cplex")
  
  print(MRSort(pT, m2pr$profilesPerformances, categoriesRanks, m2pr$weights, criteriaMinMax,
         m2pr$lambda, m2pr$vetoPerformances, alternativesIDs = incAlt))
}

# DM updates assignments
assig[2] <- "Bad"
assig[35] <- "Bad"

# generate updated model
m2 <- MRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                           veto = TRUE, readableWeights = TRUE, 
                           readableProfiles = TRUE, solver = "cplex",
                           alternativesIDs = names(assig))

# plot model
plotMRSortSortingProblem(pT, m2$profilesPerformances, categoriesRanks, assig,
                         criteriaMinMax, criteriaUBs, criteriaLBs, 
                         NULL, m2$vetoPerformances, 'V',
                         m2$weights, m2$lambda,
                         alternativesIDs = "no alternatives")

#################################
#-------- Third iteration-------#
#################################

# add assignments for the next 10 alternatives
assig <-c(assig, "Bad", "Neutral", "Bad", "Neutral",
          "Good", "Bad", "Bad", "Good", "Bad", "Bad")
names(assig) <- 1:45

# generate MR-Sort model with vetoes
m3 <- MRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                           veto = TRUE, readableWeights = TRUE, 
                           readableProfiles = TRUE, solver = "cplex",
                           alternativesIDs = names(assig))

# model does not fit
print(m3)

# generate incompatible assignments
incomp <- MRSortIdentifyIncompatibleAssignments(pT, assig, categoriesRanks,
                                                criteriaMinMax, veto = TRUE,
                                                solver = "cplex",
                                                alternativesIDs = names(assig))

# generate alternative assignments for each set of incompatible assignments
for(incAlt in incomp$incompatibleSets)
{
  m3pr <- MRSortInferenceExact(pT, assig, categoriesRanks,
                               criteriaMinMax, veto = TRUE,
                               readableWeights = TRUE,
                               readableProfiles = TRUE,
                               alternativesIDs = names(assig)[!names(assig) %in% incAlt],
                               solver = "cplex")
  
  print(MRSort(pT, m3pr$profilesPerformances, categoriesRanks, m3pr$weights, criteriaMinMax,
         m3pr$lambda, m3pr$vetoPerformances, alternativesIDs = incAlt))
}

# DM does not update any assignemnts

# generate other MR-Sort variants

# MR-Sort with dictators
m3 <- LPDMRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                              majorityRule = "D", readableWeights = TRUE,
                              readableProfiles = TRUE, minmaxLPD = TRUE, 
                              solver = "cplex", alternativesIDs = names(assig))

# model does not fit
print(m3)

# MR-Sort with dominating vetoes
m3 <- LPDMRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                              majorityRule = "dV", readableWeights = TRUE,
                              readableProfiles = TRUE, minmaxLPD = TRUE, 
                              solver = "cplex", alternativesIDs = names(assig))

# model does not fit
print(m3)

# MR-Sort with dominating dictators
m3 <- LPDMRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                              majorityRule = "Dv", readableWeights = TRUE,
                              readableProfiles = TRUE, minmaxLPD = TRUE, 
                              solver = "cplex", alternativesIDs = names(assig))

# model does not fit
print(m3)

# MR-Sort with vetoes weakened by dictators
m3 <- LPDMRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                              majorityRule = "v", readableWeights = TRUE,
                              readableProfiles = TRUE, minmaxLPD = TRUE, 
                              solver = "cplex", alternativesIDs = names(assig))

# model does not fit
print(m3)

# MR-Sort with dictators weakened by vetoes
m3 <- LPDMRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                              majorityRule = "d", readableWeights = TRUE,
                              readableProfiles = TRUE, minmaxLPD = TRUE, 
                              solver = "cplex", alternativesIDs = names(assig))

# model does not fit
print(m3)

# MR-Sort with conflicting vetoes and dictators
m3 <- LPDMRSortInferenceExact(pT, assig, categoriesRanks, criteriaMinMax,
                              majorityRule = "dv", readableWeights = TRUE,
                              readableProfiles = TRUE, minmaxLPD = TRUE, 
                              solver = "cplex", alternativesIDs = names(assig))

# model fits
print(m3)

# plot model
plotMRSortSortingProblem(pT, m3$profilesPerformances, categoriesRanks, assig,
                         criteriaMinMax, criteriaUBs, criteriaLBs, 
                         m3$dictatorPerformances, m3$vetoPerformances, 'dv',
                         m3$weights, m3$lambda,
                         alternativesIDs = "no alternatives")

#################################
#-------------- End ------------#
#################################

# we stop the protocol and extract the assignments rules

# this is currently not implemented
#extractLPDMRSortRules(m3$lambda,m3$weights,criteriaMinMax,criteriaUBs,criteriaLBs,m3$profilesPerformances,categoriesRanks,m3$vetoPerformances,m3$dictatorPerformances,"dv")