wineData.m

%% Load data and select features
% Data source: Wine data 
clear all 
mydata = csvread('data_wine_complete.csv', 1, 0);
features = mydata(:, 2:end);
labels = mydata(:, 1);
%figure(1)
idFeatures = [1 7];
XX = features(:, idFeatures);
YY = labels;
gscatter(XX(:,1),XX(:,2),YY);
xlabel('Achol level')
ylabel('Flavanoids')
title('Scatter Diagram of Wine Data')
clear features idFeatures labels

%% SVM for binary-class classification 
% prepare data, select wine type 1 and wine type 2 for SVM fitting 
figure(2)
X = XX(1:130, :);
Y = YY(1:130);
gscatter(X(:,1), X(:,2), Y);
xlabel('Achol level')
ylabel('Flavanoids')
title('Scatter Diagram of Wine Data')

%% 
% fit a SVM for classification 
figure(3)
SVM = fitcsvm(X, Y,'Standardize',true);
% Compute the scores over a grid
d = 0.02; % Step size of the grid
[x1Grid,x2Grid] = meshgrid(min(X(:,1)):d:max(X(:,1)),...
    min(X(:,2)):d:max(X(:,2)));
xGrid = [x1Grid(:),x2Grid(:)];        % The grid
[label,scores1] = predict(SVM,xGrid); % The scores
h(1:2) = gscatter(X(:,1),X(:,2),Y);
hold on

contour(x1Grid,x2Grid,reshape(scores1(:,2),size(x1Grid)),[0 0],'k', 'LineWidth', 2);
    % Decision boundary
xlabel('Achol level')
ylabel('Flavanoids')
title('SVM (linear kernel)')
legend({'1','2'},'Location','Best');
hold off

%% kernel function
% Sometimes the data cannot be seperated by linear functions. We can use
% kernel functions instead. Here we choose two types of kernel functions: polynomial and gaussian kernel. 
%
% *polynomial kernel function *
% 
figure(4)
SVM = fitcsvm(X, Y,'Standardize',true, 'KernelFunction','polynomial','Standardize',true); % polynomial, gaussian
% Compute the scores over a grid
d = 0.02; % Step size of the grid
[x1Grid,x2Grid] = meshgrid(min(X(:,1)):d:max(X(:,1)),...
    min(X(:,2)):d:max(X(:,2)));
xGrid = [x1Grid(:),x2Grid(:)];        % The grid
[~,scores1] = predict(SVM,xGrid); % The scores
h(1:2) = gscatter(X(:,1),X(:,2),Y);
hold on
contour(x1Grid,x2Grid,reshape(scores1(:,2),size(x1Grid)),[0 0],'k', 'LineWidth', 2);
    % Decision boundary
xlabel('Achol level')
ylabel('Flavanoids')
title('SVM (polynomial kernel)')
legend({'1', '2'},'Location','Best');
hold off

%% 
figure(5)
SVM = fitcsvm(X, Y,'Standardize',true, 'KernelFunction','gaussian','Standardize',true);
% Compute the scores over a grid
d = 0.02; % Step size of the grid
[x1Grid,x2Grid] = meshgrid(min(X(:,1)):d:max(X(:,1)),...
    min(X(:,2)):d:max(X(:,2)));
xGrid = [x1Grid(:),x2Grid(:)];        % The grid
[~,scores1] = predict(SVM,xGrid); % The scores
h(1:2) = gscatter(X(:,1),X(:,2),Y);
hold on
contour(x1Grid,x2Grid,reshape(scores1(:,2),size(x1Grid)),[0 0],'k', 'LineWidth', 2);
    % Decision boundary
xlabel('Achol level')
ylabel('Flavanoids')
title('SVM (gaussian kernel)')
legend({'1','2'},'Location','Best');
hold off

clear d h scores1 SVM X x1Grid x2Grid xGrid Y

%% SVM for three-class classification 
% In this section, we also want to use different kernel functions for
% classification. 
X = XX;
Y = YY;
classNameList = unique(Y);
SVMs = cell(3,1); % generate 3 SVM models for three classes 

for i = 1:numel(classNameList)
    Y_new = Y;
    id = find(Y_new ~= classNameList(i));
    Y_new(id) = -1;
    
    SVMs{i} = fitcsvm(X, Y_new,'ClassNames',[-1 i],'Standardize',true, 'KernelFunction', 'polynomial', 'BoxConstraint',1);
end

d = 0.02; 

[x1Grid,x2Grid] = meshgrid(min(X(:,1)):d:max(X(:,1)), min(X(:,2)):d:max(X(:,2)));
xGrid = [x1Grid(:),x2Grid(:)];
N = size(xGrid,1);
Scores = zeros(N,numel(classNameList));

for i = 1:numel(classNameList);
        [~, score] = predict(SVMs{i},xGrid);
        Scores(:,i) = score(:,2); % Second column contains positive-class scores
end

[~,maxScore] = max(Scores,[],2);

figure(6)
h(1:3) = gscatter(xGrid(:,1),xGrid(:,2),maxScore, [0.1 0.5 0.5; 0.5 0.1 0.5; 0.5 0.5 0.1]);
hold on
h(4:6) = gscatter(X(:,1),X(:,2),Y);
    %title('{\bf Iris Classification Regions}');
    % title(strcat('SVM with ', kernelList{k}))
title(['SVM classification with polynomial kernel']);
xlabel('Achol level')
ylabel('Flavanoids')
legend(h,{'region for type 1','region for type 2','region for type 3', 'observed type 1','observed type 2','observed type 3'},...
'Location','Northwest');
axis tight
hold off