Combining N pattern classifiers using weighted majority voting in Matlab

I want to combine some classifiers. The number of classifiers is 4 and there are 3 possible classes. I came a cross this code: "Efficient multiclass weighted majority voting implementation in MATLAB". It makes use of 3 classifiers and 3 possible classes. I have tried to customize it for use with 4 classifiers and 3 possible classes without success. How can this code be extended for use in my case, or to N classifiers. Or is there any other code applicable to my case. Please help.

Answer 1

[kh rezaee]

I think that your problem is near this code:

  voteWeightsSUM = sum(voteWeights);
  W = voteWeights/(voteWeightsSUM);
  outPut = (testPredictions(:,1)*W(1)+testPredictions(:,2)*W(2)+testPredictions(:,3)*W(3)+testPredictions(:,4)*W(4));
  VotingConfusionMatrix = confusionmat(TestLabel,outPut); 
  softVotingAccuracy = sum(diag(VotingConfusionMatrix))/sum(VotingConfusionMatrix(:));
  

Where, voteWeights and testPredictions are accuracy and predicted test labels of each classifier, respectively. Also, voteWeights comes from the training phase, but testPredictions is calculated based on trained models.