close all; clear; clc;

load data-1800
X=data(:,1:end-1);
Y=categorical(data(:,end));
rng(1)
clear xapp yapp xtest ytest

% 10 distances pour KNN
 Dist= {'euclidean' 'Mahalanobis' 'cosine' 'correlation' 'spearman' 'hamming' 'jaccard' 'minkowski' 'cityblock' 'chebychev' };

nbre=1;
for d=1:10
    for k_nn=1:5
        CVO = cvpartition(Y,'holdout',0.4);
           trIdx = CVO.training;
            tsIdx = CVO.test;
            
            % base d'apprentissage et de test
            XTrain = X(trIdx,:);
            YTrain = categorical(Y(trIdx)');
            XTest = X(tsIdx,:);
            YTest = categorical(Y(tsIdx)');
            
            % Phase de training
            tic;
            MonModele = fitcknn(X(trIdx,:), Y(trIdx),'NSMethod','exhaustive','Distance',Dist{d},...
                'NumNeighbors',k_nn,'Standardize',1);
            accuracy_tr = (1-resubLoss(MonModele))*100;
            time_tr =toc;
            
            % Phase de testing
            tic;
            Y_ts_pred = predict(MonModele,X(tsIdx,:));
            %Y_ts_pred = Y_ts_pred';
            accuracy_ts = sum(Y(tsIdx) == Y_ts_pred)/length(Y(tsIdx))*100;
            time_ts=toc;
            
            % stocker les résultats
            Performance_KNN=[accuracy_tr time_tr accuracy_ts time_ts];
        
        acc_tr=mean(Performance_KNN(:,1));
        t_tr=mean(Performance_KNN(:,2));
        acc_ts=mean(Performance_KNN(:,3));
        t_ts=mean(Performance_KNN(:,4));
        Performance_KNN_global(nbre,:)=[d k_nn acc_tr t_tr acc_ts t_ts];
        nbre=nbre+1;
    end
    
end
[val , ordre]=max(Performance_KNN_global(:,5));
% Meilleur resultat sous forme [distance k_nn acc_tr time_tr acc_ts time_ts]
Performance_KNN_global(ordre,:)