BN_Prediction

 EXAMPLE 2: Evaluation of the prediction capability of the models 
            learned during the evolution of an EDA for a multiobjective function
            The prediction is measured using the correlation between the probability
            given by the models to the solutions and its fitness values

0001  % EXAMPLE 2: Evaluation of the prediction capability of the models
0002  %            learned during the evolution of an EDA for a multiobjective function
0003  %            The prediction is measured using the correlation between the probability
0004  %            given by the models to the solutions and its fitness values
0005  
0006  %%%%% First, the EDA is defined
0007  clear edaparams;  
0008  PopSize = 1000; n = 50; cache = [1,1,1,1,1]; Card = 2*ones(1,n); MaxGen = 30;
0009  global FunctionTables;
0010  global FunctionStructure;
0011  global FunctionAccCard;
0012  global SelectedObjectives;
0013  
0014  [FunctionStructure] =  CreateListFactorsCircularNK(n,4); % The circular structure is created
0015  [FunctionTables] = ReadFunctionsFromData('testNK_fnt_N50_k4Inst_1.txt',FunctionStructure,Card); % Values are read from a file
0016  [FunctionAccCard] = FindListCard(FunctionStructure,Card); % Auxiliary structure for function evaluation
0017  SelectedObjectives = [1:4:48];
0018  
0019  F = 'PartialEvaluateGeneralFunction';  % General function that uses global variables FunctionTables, FunctionStructure and FunctionAccCard.
0020  selparams(1:2) = {0.5,'ParetoRank_ordering'};
0021  edaparams{1} = {'selection_method','truncation_selection',selparams};
0022  edaparams{2} = {'replacement_method','best_elitism',{'ParetoRank_ordering'}};
0023  edaparams{3} = {'stop_cond_method','max_gen',{MaxGen}};
0024  
0025  %%%%% Second, the EDA is executed and all the information is saved in Cache
0026  
0027  [AllStat,Cache]=RunEDA(PopSize,n,F,Card,cache,edaparams); 
0028  
0029  
0030  %%%%% Third the information is extracted and analyzed
0031  
0032  AllSols = []; AllVals = [];
0033  for i=1:MaxGen,
0034   AllSols = [AllSols;Cache{1,i}];   % All the populations
0035   AllVals = [AllVals;Cache{4,i}];   % All the evaluations
0036  end
0037  
0038  Index = FindParetoSet(AllSols,AllVals); % The set of Pareto solutions found by the EDA is extracted
0039  ParetoPop = AllSols(Index,:);
0040  ParetoVals = AllVals(Index,:);
0041  
0042  parallelcoords(ParetoVals); % The Pareto set is shown using parallel coordinates
0043  ObjectivesCorr = corr(ParetoVals); % The correlations between the objectives are computed.
0044  [idx,netsim,dpsim,expref]=apcluster(ObjectivesCorr,mean(ObjectivesCorr)); % Affinity propagation is used
0045                                                                            % to reduce the number of objectives
0046                                                                            % by selecting correlated variables
0047  
0048  for i=1:MaxGen,              % The correlations between the probabilities of each BN and each of
0049   bnet = Cache{3,i};          % the problem objectives are computed using the Pareto Set
0050   All_BN_Fit_Corr(i,:) =  BN_Fitness_Corr(bnet,ParetoPop,ParetoVals)
0051  end
0052