classifierwrapper.m

最新的模式识别分类工具箱,希望对朋友们有用!

% trains the classifier to distinguish between pairs of classes, computes
% majority vote, and labels accordingly.
% This is useful for classifiers such as Linear Discriminants, LS,
% perceptron etc.
% 
% Usage:
% [train_error, test_error, estTrainLabels, estTestLabels] = ...
%    classifierWrapper(train_features, train_labels,classifier, params,test_features, test_labels)
%
% where
% 	trainFeatures  - the training set vectors, one vector per column
%	trainLabels    - the labels of the above
%       classifier     - the name of the classification routine
%       params         - the parameters of the classifier
%       testFeatures   - test set, one column per vector
%       testLabels     - labels  for test set%
% Outputs
%	trainError     - the error rate on the training set (one entry per
%	                   class + total error)
%	testError      - the error rate on the test set (one entry per class
%	                  + total error)
%       estTrainLabels - the labels produced by the classifier on the
%                        training set
%       estTestLabels  - the labels produced by the classifier on the
%                        training set
%
%

function [trainError, testError, estTrainLabels, estTestLabels] = ...
  classifierWrapper(trainFeatures, trainLabels,classif, parameters,testFeatures, testLabels)

hm = findobj('Tag', 'Messages'); 
fprintf('Performing pairwise classification\n');
if (isempty(hm)==0)
  s = sprintf('Performing pairwise classification\n');
  set(hm,'String',s);
  refresh;
  pause(1);
end

[Nclasses, classes]  = find_classes([trainLabels(:);testLabels(:)]);
% Number of classes in labels

[Dim, NsamTrain]     = size(trainFeatures);
[Dim, NsamTest]      = size(testFeatures);

counts      = zeros(Nclasses, NsamTrain+NsamTest);
allFeatures = [trainFeatures testFeatures];
allLabels   = [trainLabels   testLabels];


for outclass =1:Nclasses-1,
  % extracts the outer class
  outclassIndex    = find(trainLabels == classes(outclass));
  outclassFeatures = trainFeatures(:,outclassIndex);
  outclassLabels   = zeros (1, max(size(outclassFeatures)));


  % extracts the inner class 
  for inclass = outclass+1:Nclasses,

    fprintf('Working on classes %d and %d\n',classes(outclass),classes(inclass));
    if (isempty(hm)==0)
      s = sprintf('Working on classes %d and %d\n',classes(outclass),classes(inclass));
      set(hm,'String',s);
      refresh;
      pause(.1);
    end
    
    inclassIndex    = find(trainLabels == classes(inclass));
    inclassFeatures = trainFeatures(:,inclassIndex);
    inclassLabels   = ones (1, max(size(inclassFeatures)));
    
    pairFeatures = [outclassFeatures inclassFeatures];
    pairLabels   = [outclassLabels   inclassLabels];
    
    % classifies
    [foo, foo1, hatTrLab, hatAllLab] = feval(classif, pairFeatures, ...
	pairLabels,parameters, allFeatures, allLabels);
    
    % collects votes
    counts(inclass,:) = counts(inclass,:)+(hatAllLab(:))';
    hatAllLab = 1-hatAllLab;
    counts(outclass,:) = counts(outclass,:) + (hatAllLab(:))';
  end
end

[vals,indexes] = max(counts);
vals = classes(indexes);
estTrainLabels = vals(1:NsamTrain);
estTestLabels  = vals((NsamTrain+1):end);

trainError = computeError(classes, trainLabels, estTrainLabels);
testError  = computeError(classes, testLabels , estTestLabels);