classify_test_ds2.asv

function classify_test_ds2(featureSet, featSel, classifier, param1, param2, param3)
%function classify_test_ds2(featureSet, featSel, classifier, C, gamma)
% Classificação de Arritmia. Treino o modelo com DS1 e valida em DS2.
%
% >> featureSet: conjunto de características
% Pode assumir os valores:
% 'YuCN','YuFourier', '2005Chazal', '2005Guler', 2005Song, 2006Yu, 2007Yu, 2010YeCoimbra, VCGComplexNet
%
% >> featSel: vetor para seleção de características
% Deve ser no formato: fs = [1 1 0 0 0 1 0 0 1], do dimensão das
% características onde:
% 1 - usar a característica
% 0 - ignorar a característica
% passe um vetor vazio para utilizar todas as características
%
% >> classifier: classificador a ser utilizado
% Pode assumir os valores:
%
% 'SVM', 'LD', 'MLP', 'PNN'
%
% param1
% C: parâmetro para SVM
% 0: escolhe default (0.05)
% -1: utiliza script para grid selection
%
% param2
% gamma: parâmetro para SVM
% 0: escolhe default (1/8*num features)
% 
% Autor: Eduardo Luz
%
if nargin < 4
   %C = 0;
   %gamma = 0; % default
   % default
   param1 = 0;
   param2 = 0;
   param3 = 0;  
elseif nargin < 5
   param2 = 0;
   param3 = 0;  
elseif nargin < 6
   param3 = 0;  
end

s = char(featureSet);

if (~isempty(featSel))
    numfs = size(featSel,2);
    fileNamed = ['results\',s,'\',classifier,'_','_test_ds2_results_FS_' num2str(numfs) '.tex'];
else
    fileNamed = ['results\',s,'\',classifier,'_','_test_ds2_results_allFeatures.tex'];
end

arq = fopen(fileNamed,'w');

% Tabela latex dos resultados
fprintf(arq,'\\documentclass{article}\n');
fprintf(arq,'\\usepackage{graphicx}\n');
fprintf(arq,'\\usepackage[latin1]{inputenc}\n');
fprintf(arq,'\\usepackage{tabularx}\n');
fprintf(arq,'\\usepackage{multirow}\n');
fprintf(arq,'\\newcommand{\\citep}{\\cite}\n');
fprintf(arq,'\\newcommand{\\citet}{\\cite}\n');
fprintf(arq,'\\newcommand{\\TFigure}{Fig.}\n');
fprintf(arq,'\\begin{document}\n');

fprintf(arq,'\n');
fprintf(arq,'\\begin{table*} \n');
fprintf(arq,'\\footnotesize \n');
s2 = char('\\caption{Tebela de resultados por paciente ');
s2 = [s2 s(10:end-1) '} \n'];
fprintf(arq,s2);
%fprintf(arq,' \\caption{Tebela dos registros do método} \n');
fprintf(arq,' \\label{tab:regtable} \n');
fprintf(arq,' \\begin{center} \n');
fprintf(arq,'  \\begin{tabular}{|c|c|c|c|c|c|c|} \n');
fprintf(arq,'   \\hline \n');
fprintf(arq,'    Registro & Acc & N Se/+P/FPR & SVEB Se/+P/FPR & VEB Se/+P/FPR & F Se/+P/FPR & Q Se/+P/FPR  \\\\ \n');
fprintf(arq,'   \\hline \n');

% iniciliza variaveis
finalCM = zeros(5,5);

% Inicializa os registros
DS1 = {'101';'106';'108';'109';'112';'114';'115';'116';'118';'119';'122';'124';'201';'203';'205';'207';'208';'209';'215';'220';'223';'230'};
DS2 = {'100';'103';'105';'111';'113';'117';'121';'123';'200';'202';'210';'212';'213';'214';'219';'221';'222';'228';'231';'232';'233';'234'};   
%DS1 = {'101';'106'};
%DS2 = {'100';'103'};   

% imprime a tabela de registros em arquvo .tex
printRegisterTable(['features\' featureSet '\']);
printRegisterTable_15_classes(['features\' featureSet '\']);

fprintf('\n Registro | Acc | N_Se SVEB_Se VEB_Se N+P SVEB+P VEB+P \n')

test_da = [];
train_ds = [];
    
for j=1:size(DS1,1)
    train_ds(j) = str2double(DS1(j));
end

first_time=1;
    
for k=1:size(DS2,1) % numero de registros
    
    test_ds = str2double(DS2(k));
       
%% Carrega os dados

feat_folder = ['features\' featureSet '\'];

[p1d p1t p2d p2t] = loadDataAAMI(0,feat_folder,train_ds,test_ds);

if ~isempty(featSel)
    fsel = find(featSel==1);

    p1d = p1d(:,fsel) ; 
    p2d = p2d(:,fsel) ; 
end

%primeira etapa com DS1 treino DS2 teste
fs1.train = p1d;
fs1.test = p2d;
target.train = p1t;
target.test = p2t;

if(strcmp(classifier,'LD'))
    %[fs1.train fs1.test] = normaliza(fs1.train, fs1.test);
elseif(strcmp(classifier,'PNN'))
	% normalização feita em classifyPNN(..)
else
    [fs1.train, scale_factor] = mapminmax(fs1.train');   
    fs1.test = mapminmax('apply',fs1.test',scale_factor);

    fs1.train = fs1.train';
    fs1.test = fs1.test';
end


%% aplica o classificador
if(strcmp(classifier,'LD'))
    cd('LD_classifier')

    cm1 = ld_Classifier_chazal(fs1,fs1,target);
    
    %fprintf('\n----------------------- Classificador LD ---------------------\n')
    cd ..
elseif(strcmp(classifier,'SVM'))
    cd('svm')
    
    tic
    
    if(first_time==1)
        first_time=0;
        
        if param1 == -1
            [newData newTarget] = undersampling_all_class(20, fs1.train,target.train);
            [best_c,best_g,best_cv,hC] = parameter_optimization(newData,newTarget);
        else
            if param1 == 0
                best_c=0.05;
            else
                best_c=param1;
            end

            if param2 ==0
                best_g=1/(8*size(fs1.train,2));
            else
                best_g= param2;
            end
        end
         
        %[newData newTarget] = SMOTE(fs1.train, target.train, 2, 4000, 5);
        [cm1 model] = svm_Classifier(fs1.train,target.train,fs1.test,target.test,best_c,best_g);
    else
        testingLabels = [];
        targetsTest = [];
        % formata para uso em SVM - para cada classe
        for t=1:size(target.test,2)
            targetsTest(find(target.test(:,t)==1),t)=t;     
        end
        testingLabels = sum(targetsTest');
        
        [predict_label, accuracy, dec_values] = svmpredict(testingLabels', fs1.test, model); % test the training data
        %cm1 = confusionmat(testingLabels',predict_label);
        clear predict_matrix;
        
            for i=1: size(predict_label,1)
            switch(predict_label(i))
                case 1,
                    predict_matrix(i,:) = [1 0 0 0 0];            
                case 2,
                    predict_matrix(i,:) = [0 1 0 0 0];            
                case 3,
                    predict_matrix(i,:) = [0 0 1 0 0];            
                case 4,
                    predict_matrix(i,:) = [0 0 0 1 0]; 
                case 5,
                    predict_matrix(i,:) = [0 0 0 0 1];
            end
            end

            clear label_matrix;
            for i=1: size(testingLabels,2)
            switch(testingLabels(i))
                case 1,
                    label_matrix(i,:) = [1 0 0 0 0];            
                case 2,
                    label_matrix(i,:) = [0 1 0 0 0];            
                case 3,
                    label_matrix(i,:) = [0 0 1 0 0];            
                case 4,
                    label_matrix(i,:) = [0 0 0 1 0]; 
                case 5,
                    label_matrix(i,:) = [0 0 0 0 1];
            end
            end
            %[C,CM,IND,PER] = CONFUSION(TARGETS,OUTPUTS)
            [c cm1] = confusion(label_matrix', predict_matrix');
    end
    
    toc
    %fprintf('\n----------------------- Classificador SVM
    %---------------------\n')
    
    cd ..
%elseif(strcmp(classifier,'PNN'))
%    cd pnn
%    
%    [acc1 sensitivityN1 sensitivitySVEB1 sensitivityVEB1 sensitivityF1 sensitivityQ1 specificitySVEB1 specificityVEB1 specificityF1 specificityQ1] = pnn_Classifier(fs.train,target.train,fs.test,target.test);
%    [acc2 sensitivityN2 sensitivitySVEB2 sensitivityVEB2 sensitivityF2
%    sensitivityQ2 specificitySVEB2 specificityVEB2 specificityF2 specificityQ2] = svm_Classifier(fs.train,target.train,fs.test,target.test);%

%    cd ..
elseif(strcmp(classifier,'MLP'))
    cd('mlp')
    
    tic
    
   if(first_time==1)
         first_time=0;
   	 [cm1 net1 net2 net3 net4 combNet] = mlp_Classifier(fs1.train,target.train,fs1.test,target.test);
   else
	%% Testa o método com input2 e target2 (dados de teste)
	outputs1 = sim(net1,fs1.test');
	outputs2 = sim(net2,fs1.test');
	outputs3 = sim(net3,fs1.test');
	outputs4 = sim(net4,fs1.test');

	newInputTest = [outputs1;outputs2;outputs3;outputs4];

	predictedTarget2 = sim(combNet,newInputTest);

	% cria matriz de confusão
	% c = total accuracy
	% cm = matriz de confusão
	% per = S x 3 matrix, where each ith row represents the percentage of false
	% negatives, false positives, and true positives for the ith category
	[c,cm1,ind,per] = confusion(target.test',predictedTarget2);
   end
    
    %fprintf('\n----------------------- Classificador MLP comb ---------------------\n')
    
    toc
    
    cd ..
elseif(strcmp(classifier,'PNN')) 
	cd('pnn')

	if(first_time==1)
        first_time=0;
		if param1 == 0
			spread = 0.1;
		else
			spread = param1;		
		end
	  	[cm1 net] = classifyPNN(fs1.train,target.train,fs1.test,target.test, spread);
    else

        %formata para pnn
        testingLabels = [];
        targetsTest = [];
        for t=1:size(target.test,2)
            targetsTest(find(target.test(:,t)==1),t)=t;     
        end
        testingLabels = sum(targetsTest');

		for(i=1:length(fs1.test(:,1)))
		    for(j=1:length(fs1.test(1,:)))
			%normaliza tb a matriz de teste
			fs1.test(i,j) = tansig((fs1.test(i,j)-mean(fs1.test(i,:)))/std(fs1.test(i,:)));
		    end
		end

		predict_label = sim(net,fs1.test');
		%[c cm] = confusion(target.test', predict_label);

        predict_label = predict_label';
            
        clear predict_matrix;
        predict_matrix=[];
        for t=1:2
            predict_matrix(find(predict_label(:,t)==1),t)=1;     
        end

        clear label_matrix;
        label_matrix=[];
        for i=1: size(testingLabels,2)
        switch(testingLabels(i))
            case 1,
                label_matrix(i,:) = [1 0];            
            case 2,
                label_matrix(i,:) = [0 1];            

        end
        end
        %[C,CM,IND,PER] = CONFUSION(TARGETS,OUTPUTS)
        [c cm] = confusion(label_matrix', predict_matrix');

		if size(cm,1)<5
		    cm_N = zeros(5,5);
		    for i=1:5
			for j=1:5
			    if i <= size(cm,1) && j <= size(cm,2)
				cm_N(i,j) = cm(i,j);
			    end
			end
		    end
		    
		    cm = cm_N;
		end

	end

	cd ..
end


% Calcula estatíSticas
 % Calcula estatíSticas
   cm1
   
   finalCM = finalCM + cm1;
   
   acc_num=0;
   acc_den=0;
   den1=0;
   den2=0;
   num=0;
   t=0;
   
    if(size(cm1,1)>=1)
       t = 1;
       
       num = cm1(t,t);
       den1 = sum(cm1(t,:));
       den2 = sum(cm1(:,t));
        
       TN = sum(sum(cm1(:,:))) - den1 - den2 + cm1(t,t);
       FP = den2 - cm1(t,t);
       
        if(den1~=0)
            sensitivityN = (num/den1) * 100;
        else
            sensitivityN = -1;
        end
        
        if(den2~=0)
            specificityN = (num/den2) * 100;
        else
            specificityN = 0;
        end
        
        if(TN + FP > 0)
            FPR_N = 100*FP/(TN+FP);
        else
            FPR_N=0;
        end
        
        acc_num = acc_num + num;
        acc_den = acc_den + den1;
        
    else
        sensitivityN = -1;
        specificityN = -1;
        FPR_N=-1;
    end
   
    % caso especial para LD classifier
    %size(cm1,1)==5
    
    if(size(cm1,1)>=2)
       t = 2;
     
       num = cm1(t,t);
        den1 = sum(cm1(t,:));
        den2 = sum(cm1(:,t));
        
        TN = sum(sum(cm1(:,:))) - den1 - den2 + cm1(t,t);
        FP = den2 - cm1(t,t);
        
        if(den1~=0)
            sensitivitySVEB = (num/den1) * 100;
        else
            sensitivitySVEB = -1;
        end
        
        if(den2~=0)
            specificitySVEB = (num/den2) * 100;
        else
            specificitySVEB = -1;
        end
        
        if(TN + FP > 0)
            FPR_SVEB = 100*FP/(TN+FP);
        else
            FPR_SVEB=-1;
        end
        
        acc_num = acc_num + num;
        acc_den = acc_den + den1;
        
    else
        sensitivitySVEB = -1;
        specificitySVEB = -1;
        FPR_SVEB=-1;
    end
   
    if(size(cm1,1)>=3)
       t = 3;
        num = cm1(t,t);
        den1 = sum(cm1(t,:));
        den2 = sum(cm1(:,t));
        
        TN = sum(sum(cm1(:,:))) - den1 - den2 + cm1(t,t);
        FP = den2 - cm1(t,t);
        
        if(den1~=0)
            sensitivityVEB = (num/den1) * 100;
        else
            sensitivityVEB = -1;
        end
        
        if(den2~=0)
            specificityVEB = (num/den2) * 100;
        else
            specificityVEB = -1;
        end
        
        if(TN + FP > 0)
            FPR_VEB = 100*FP/(TN+FP);
        else
            FPR_VEB=-1;
        end
        
        acc_num = acc_num + num;
        acc_den = acc_den + den1;
        
    else
        sensitivityVEB = -1;
        specificityVEB = -1;
        FPR_VEB=-1;
    end
   
    if(size(cm1,1)>=4)
       t = 4;
       
       num = cm1(t,t);
        den1 = sum(cm1(t,:));
        den2 = sum(cm1(:,t));
        
        TN = sum(sum(cm1(:,:))) - den1 - den2 + cm1(t,t);
        FP = den2 - cm1(t,t);
        
        if(den1~=0)
            sensitivityF = (num/den1) * 100;
        else
            sensitivityF = -1;
        end
        if(den2~=0)
        specificityF = (num/den2) * 100;
        else
            specificityF = -1;
        end
        
        if(TN + FP > 0)
            FPR_F = 100*FP/(TN+FP);
        else
            FPR_F=-1;
        end
        
        acc_num = acc_num + num;
        acc_den = acc_den + den1;
        
    else
        sensitivityF = -1;
        specificityF = -1;
        FPR_F=-1;
    end
   
    if(size(cm1,1)>=5)
       t = 5;
       
        num = cm1(t,t);
        den1 = sum(cm1(t,:));
        den2 = sum(cm1(:,t));
        
        TN = sum(sum(cm1(:,:))) - den1 - den2 + cm1(t,t);
        FP = den2 - cm1(t,t);
        
        if(den1~=0)
            sensitivityQ = (num/den1) * 100;
        else
            sensitivityQ = -1;
        end
        
        if(den2~=0)
            specificityQ = (num/den2) * 100;
        else
            specificityQ = -1;
        end
        
        if(TN + FP > 0)
            FPR_Q = 100*FP/(TN+FP);
        else
            FPR_Q=-1;
        end
        
        acc_num = acc_num + num;
        acc_den = acc_den + den1;
        
    else
        sensitivityQ = -1;
        specificityQ = -1;
        FPR_Q=-1;
    end
    
    fprintf(arq,'%6d & %6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f \\\\ \n',...
        str2double(DS2(k)), 100*acc_num/acc_den, sensitivityN,specificityN,FPR_N,sensitivitySVEB,specificitySVEB,FPR_SVEB,sensitivityVEB,specificityVEB,FPR_VEB,...
        sensitivityF,specificityF,FPR_F, sensitivityQ,specificityQ,FPR_Q);         
    fprintf(arq,'\n');
    
   fprintf('\n');
    
   fprintf('Registro=%6.0f | Acc=%6.1f |', str2double(DS2(k)), 100*acc_num/acc_den);
    
   fprintf(' N_Se=%6.1f SVEB_Se=%6.1f VEB_Se=%6.1f N+P=%6.1f SVEB+P=%6.1f VEB+P=%6.1f \n\n',...
       sensitivityN, sensitivitySVEB,sensitivityVEB,  specificityN, specificitySVEB,specificityVEB)
 
   
end % for t

fprintf(arq,'    \\hline \n');
%fprintf(arq,'    \\hline \n');

TN = sum(sum(finalCM(:,:))) - sum(finalCM(1,:)) - sum(finalCM(:,1)) + finalCM(1,1);
FP = sum(finalCM(:,1)) - finalCM(1,1);
FPR_N = FP/(TN+FP);
        
TN = sum(sum(finalCM(:,:))) - sum(finalCM(2,:)) - sum(finalCM(:,2)) + finalCM(2,2);
FP = sum(finalCM(:,2)) - finalCM(2,2);
FPR_SVEB = FP/(TN+FP);

TN = sum(sum(finalCM(:,:))) - sum(finalCM(3,:)) - sum(finalCM(:,3)) + finalCM(3,3) ;
FP = sum(finalCM(:,3)) - finalCM(3,3);
FPR_VEB = FP/(TN+FP);

if size(finalCM,1) > 3

TN = sum(sum(finalCM(:,:))) - sum(finalCM(4,:)) - sum(finalCM(:,4)) + finalCM(4,4);
FP = sum(finalCM(:,4)) - finalCM(4,4);
FPR_F = FP/(TN+FP);

TN = sum(sum(finalCM(:,:))) - sum(finalCM(5,:)) - sum(finalCM(:,5)) + finalCM(5,5);
FP = sum(finalCM(:,5)) - finalCM(5,5);
FPR_Q = FP/(TN+FP);

fprintf(arq, ' Gross & %6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f \\\\ \n',...
    100*(finalCM(1,1)+finalCM(2,2)+finalCM(3,3)+finalCM(4,4)+finalCM(5,5))/(sum(finalCM(1,:))+sum(finalCM(2,:))+sum(finalCM(3,:))+sum(finalCM(4,:))+sum(finalCM(5,:))) ,...
       100*finalCM(1,1)/sum(finalCM(1,:)), 100*finalCM(1,1)/sum(finalCM(:,1)),100*FPR_N,...
       100*finalCM(2,2)/sum(finalCM(2,:)),100*finalCM(2,2)/sum(finalCM(1:end-1,2)),100*FPR_SVEB,...
       100*finalCM(3,3)/sum(finalCM(3,:)),100*finalCM(3,3)/sum(finalCM(1:end-2,3)),100*FPR_VEB,...
       100*finalCM(4,4)/sum(finalCM(4,:)),100*finalCM(4,4)/sum(finalCM(:,4)),100*FPR_F,...
       100*finalCM(4,4)/sum(finalCM(5,:)),100*finalCM(5,5)/sum(finalCM(:,5)),100*FPR_Q);
   
else
    fprintf(arq, ' Gross & %6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & %6.1f/%6.1f/%6.1f & - & - \\\\ \n',...
    100*(finalCM(1,1)+finalCM(2,2)+finalCM(3,3))/(sum(finalCM(1,:))+sum(finalCM(2,:))+sum(finalCM(3,:))) ,...
       100*finalCM(1,1)/sum(finalCM(1,:)), 100*finalCM(1,1)/sum(finalCM(:,1)),100*FPR_N,...
       100*finalCM(2,2)/sum(finalCM(2,:)),100*finalCM(2,2)/sum(finalCM(:,2)),100*FPR_SVEB,...
       100*finalCM(3,3)/sum(finalCM(3,:)),100*finalCM(3,3)/sum(finalCM(:,3)),100*FPR_VEB);
end

fprintf(arq,'   \\hline \n');
fprintf(arq,'  \\end{tabular} \n');
fprintf(arq,' \\end{center} \n');
fprintf(arq,'\\end{table*} \n');
fprintf(arq,'\n');
fprintf(arq,'\n');

fprintf(arq,'\n');
fprintf(arq,'\\begin{table*} \n');
fprintf(arq,' \\caption{Matriz e confusão} \n');
fprintf(arq,' \\label{tab:regtable} \n');
fprintf(arq,' \\begin{center} \n');
fprintf(arq,'  \\begin{tabular}{|');
for tt=1:size(finalCM,1)
    fprintf(arq,'c|');
end
fprintf(arq,'} \n');
fprintf(arq,'\n');
fprintf(arq,'   \\hline \n');
for tt=1:size(finalCM,1)
    for uu=1:size(finalCM,2)
        if uu==5
            fprintf(arq,'%6.0f',finalCM(tt,uu));
        else
            fprintf(arq,'%6.0f & ',finalCM(tt,uu));
        end
    end
 fprintf(arq,'    \\\\ \n');
 fprintf(arq,'   \\hline \n');
 end
fprintf(arq,'   \\hline \n');
fprintf(arq,'  \\end{tabular} \n');
fprintf(arq,' \\end{center} \n');
fprintf(arq,'\\end{table*} \n');
fprintf(arq,'\n');

if(strcmp(classifier,'SVM'))
    fprintf(arq,'SVM parameters - C = %6.2f, gamma= %6.2f',best_c, best_g);
end

fprintf(arq,'\n');
fprintf(arq,'\\end{document}\n');
fclose(arq);

fprintf('\n Gross Statistics:\n');
fprintf(' Gross & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f & %6.1f \\\\ \n',...
       100*(finalCM(1,1)+finalCM(2,2)+finalCM(3,3)+finalCM(4,4)+finalCM(5,5))/(sum(finalCM(1,:))+sum(finalCM(2,:))+sum(finalCM(3,:))+sum(finalCM(4,:))+sum(finalCM(5,:))) ,...
       100*finalCM(1,1)/sum(finalCM(1,:)), 100*finalCM(2,2)/sum(finalCM(2,:)), 100*finalCM(3,3)/sum(finalCM(3,:)), 100*finalCM(4,4)/sum(finalCM(4,:)), 100*finalCM(5,5)/sum(finalCM(5,:)),...
       100*finalCM(1,1)/sum(finalCM(:,1)), 100*finalCM(2,2)/sum(finalCM(:,2)), 100*finalCM(3,3)/sum(finalCM(:,3)), 100*finalCM(4,4)/sum(finalCM(:,4)), 100*finalCM(5,5)/sum(finalCM(:,5)));
  
   
end