Mutual Subspace Method

Author: Santos Safrao

data/CVLABFace2.mat

@@ -0,0 +1,17 @@
+load("data/CVLABFace2.mat")
+training_data = X1;
+testing_data = X2;
+
+[~, num_samples, ~] = size(training_data);
+[num_dim, num_samples_per_set, num_sets, num_classes] = size(testing_data);
+
+num_dim_reference_subspaces = 20;
+num_dim_input_subpaces = 5;
+
+reference_subspaces = computeBasisVectors(training_data, num_dim_reference_subspaces);
+input_subspaces = computeBasisVectors(testing_data, num_dim_input_subpaces);
+similarities = computeSubspacesSimilarities(reference_subspaces, input_subspaces);
+
+model_evaluation = ModelEvaluation(similarities(:, :, end, end), generateLabels(size(testing_data, 3), num_classes));
+
+displayModelResults('Mutual Subspace Methods', model_evaluation);

examples/msm.m

@@ -1,2 +1 @@
 addpath(genpath('..')) % Add the subfolders of the project to the path
-addPackageFoldersToPath

install.m

@@ -1,5 +1,6 @@
 classdef ModelEvaluation
  properties
+ accuracy; % Accuracy
  error_rate; % Overall error rate
  equal_error_rate; % Equal error rate (point where FAR and FRR are approximately equal)
  classification_threshold; % Threshold used for classification
@@ -43,11 +44,13 @@
  else
  [~, predicted_labels] = min(evaluation_values, [], 1);
  end
- 
- obj.error_rate = 1 - mean(predicted_labels == labels);
+ obj.accuracy = mean(predicted_labels == labels); 
+ obj.error_rate = 1 - obj.accuracy;
  else
  binary_labels = zeros(size(labels));
  binary_labels(labels ~= 0) = 1;
+ predicted_labels = evaluation_values >= obj.classification_threshold;
+ obj.accuracy = mean(predicted_labels == binary_labels);
  end
 
  evaluation_values = evaluation_values(:);

src/classes/@ModelEvaluation/ModelEvaluation.m

@@ -0,0 +1,127 @@
+
+classdef OrzEval
+ properties (SetAccess = public)
+ ER;
+ EER;
+ Thres;
+ 
+ A;
+ FAR;
+ FRR;
+ 
+ nP;
+ nN;
+ 
+ flgSIM;
+ 
+ end% properties
+ 
+ methods
+ function OB = OrzEval(VAL, Label, varargin)
+%function OB = OrzEval(VAL, Label, varargin)
+% VAL: �ގ��x�������͔�ގ��x�i�����j���������s��A�������͍s�x�N�g��
+% VAL���s��̏ꍇ�A���N���X���i�Q�N���X�ȏ�j�Ɣ��f
+% VAL���s�x�N�g���̏ꍇ�A�P�N���X�Ɣ��f��ER���v�Z���Ȃ�
+% Label: VAL�̗񐔂Ɠ����T�C�Y�̍s�x�N�g��
+% VAL�̐������x����ێ�
+% ���N���X���̏ꍇ�A�P�`�N���X���̒l
+% �P�N���X���̏ꍇ�A�P�iPositive�j�ƂO�iNegative�j
+% ��O�����F VAL�̒l���ގ��x����ގ��x�i�����j�����肷��
+% �f�t�H���g�ł́A�ގ��x
+% ����'D'����O�����ɓ��͂��ꂽ�ꍇ�A��ގ��x�i�����j�Ƃ��Čv�Z
+% 
+% PlotEER�F False Reject Rate��False Alarm Rate��Figure(10)�ɕ`��
+% �����ɂ��A�ԍ���ύX�”\
+% PlotROC�F ROC curve��Figure(100)�ɕ`��
+% �����ɂ��A�ԍ���ύX�”\
+ 
+ VAL=VAL(:,:);
+ % �ގ��x����ގ��x��
+ OB.flgSIM=true;
+ if nargin == 3
+ if varargin{1}=='D';
+ OB.flgSIM=false;
+ end
+ end
+ 
+ % One-Class ��肩�ǂ���
+ if size(VAL,1)>=2
+ B=zeros(size(VAL));
+ Lu = unique(Label);
+ for I=1:size(Lu,2)
+ B(I,Label==Lu(I))=1;
+ end
+ 
+ if OB.flgSIM
+ [v ind] = max(VAL,[],1);
+ else
+ [v ind] = min(VAL,[],1);
+ end
+ OB.ER = 1-mean(ind == Label);
+ 
+ else
+ B = zeros(size(Label));
+ B(Label~=0)=1;
+ end
+ VAL=VAL(:);
+ B=B(:);
+ 
+ OB.nP = sum(B==1);
+ OB.nN = sum(B==0);
+ 
+ if OB.flgSIM
+ [OB.A C]= sort(VAL,'ascend');
+ else
+ [OB.A C]= sort(VAL,'descend');
+ end
+ D = B(C);
+ 
+ OB.FAR = 1-cumsum(D==0)/OB.nN;
+ OB.FRR = cumsum(D==1)/OB.nP;
+ 
+ [val ind] = min((abs(OB.FAR-OB.FRR)));
+ OB.EER = (OB.FAR(ind)+OB.FRR(ind))/2;
+ OB.Thres = OB.A(ind);
+ end
+ 
+ function PlotEER(OB,varargin)
+ if nargin == 2
+ No = varargin{1};
+ else
+ No = 10;
+ end
+ 
+ figure(No)
+ clf;
+ hold on
+ plot(OB.A,OB.FRR,'b');
+ plot(OB.A,OB.FAR,'r');
+ title('FRR - FAR');
+ legend('False Reject Rate','False Alarm Rate' );
+ xlabel('Threshold')
+ ylabel('Rate')
+ hold off
+ end
+ 
+ function PlotROC(OB,varargin)
+ if nargin == 2
+ No = varargin{1};
+ color = 'r';
+ elseif nargin == 3
+ No = varargin{1};
+ color = varargin{2}; 
+ else
+ No = 100;
+ color = 'r';
+ end
+ 
+ figure(No)
+ %clf;
+ hold on
+ plot(OB.FAR,1-OB.FRR,color);
+ xlabel('False Positive Rate')
+ ylabel('True Positive Rate')
+ hold off
+ end
+ end
+end

src/classes/@OrzEval/OrzEval.m

@@ -0,0 +1,22 @@
+function displayModelResults(model_name, model_evaluation)
+% This function displays the model evaluation results.
+%
+% Parameters:
+% - model_name: Name of the model being evaluated.
+% - model_evaluation: Object of ModelEvaluation class containing evaluation results.
+
+% Assertions to check input types
+assert(ischar(model_name) || isstring(model_name), 'The model name should be a string.');
+assert(isa(model_evaluation, 'ModelEvaluation'), 'The model_evaluation should be an instance of the ModelEvaluation class.');
+
+% Display the model name
+fprintf('\nModel: %s\n', model_name);
+
+% Display the results
+disp('---------- Model Evaluation Results ----------');
+fprintf('Model Accuracy: %.2f%%\n', model_evaluation.accuracy * 100); % Display as percentage
+fprintf('Model Error Rate: %.2f%%\n', model_evaluation.error_rate * 100); % Display as percentage
+fprintf('Equal Error Rate (EER): %.2f%%\n', model_evaluation.equal_error_rate * 100); % Display as percentage
+fprintf('Classification Threshold: %.2f\n', model_evaluation.classification_threshold);
+disp('----------------------------------------------');
+end

src/functions/displayModelResults.m

@@ -0,0 +1,61 @@
+function labels = generateLabels(num_samples, varargin)
+ % generateLabels Generates a vector of labels for dataset samples.
+ %
+ % This function creates a vector of integer labels for a dataset based on
+ % the specified number of samples, classes, and sets. Labels are assigned
+ % in a repeating sequential manner across the classes for each set.
+ %
+ % Parameters:
+ % num_samples (integer): The number of samples in each class.
+ % varargin: Optional input arguments to specify number of classes and sets.
+ % - If one additional argument is provided, it specifies the number of classes.
+ % - If two additional arguments are provided, the first specifies the number
+ % of classes, and the second specifies the number of sets.
+ %
+ % Returns:
+ % labels (1 x N integer array): The generated vector of labels, where N is the
+ % total number of samples across all classes and sets.
+ %
+ % Usage:
+ % labels = generateLabels(100);
+ % % Generates 100 labels, all assigned to a single class and set.
+ %
+ % labels = generateLabels(100, 5);
+ % % Generates 500 labels for 5 classes, each class having 100 samples,
+ % % all within a single set.
+ %
+ % labels = generateLabels(100, 5, 3);
+ % % Generates 1500 labels for 5 classes and 3 sets, each class having
+ % % 100 samples in each set.
+ %
+ % Errors:
+ % - An error is thrown if the number of input arguments is less than 1 or more than 3.
+ %
+ % Example:
+ % labels = generateLabels(10, 3, 2)
+ % % Returns: [1 2 3 1 2 3 1 2 3 1 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3]
+ 
+ % Validate the number of input arguments
+ if nargin < 1
+ error('Not enough input arguments.');
+ elseif nargin == 1
+ num_classes = 1; % Set default number of classes to 1
+ num_sets = 1; % Set default number of sets to 1
+ elseif nargin == 2
+ num_classes = varargin{1}; % Get number of classes from input
+ num_sets = 1; % Set default number of sets to 1
+ elseif nargin == 3
+ num_classes = varargin{1}; % Get number of classes from input
+ num_sets = varargin{2}; % Get number of sets from input
+ else
+ error('Too many input arguments.');
+ end
+ 
+ % Create a matrix of class labels, repeating each class label 'num_samples' times
+ % and then repeating the whole sequence for each set.
+ A = repmat(1:num_classes, [num_samples, num_sets]);
+ 
+ % Convert the matrix of class labels into a single row vector
+ labels = A(:)';
+ end
+ 

src/functions/generateLabels.m

@@ -0,0 +1,28 @@
+function L = orzLabel(nNum,varargin)
+
+
+if nargin <1
+ error('error');
+end
+if nargin == 1
+ nClass = 1;
+ nSet = 1;
+end
+
+if nargin == 2
+ nClass = varargin{1};
+ nSet = 1;
+end
+
+if nargin == 3
+ nClass = varargin{1};
+ nSet = varargin{2};
+end
+
+if nargin > 3
+ error('error');
+end
+
+A=repmat(1:nClass,[nNum,nSet]);
+L=A(:)';
+

src/functions/orzLabel.m

@@ -1,8 +1,2 @@
-% These are the similarity scores for each data point, representing likelihood of belonging to class 1
-evaluation_values = [0.9, 0.8, 0.2, 0.1, 0.9, 0.7];
-% 1 for class 1 and 2 for class 2
-labels = [1, 1, 2, 2, 1, 1];
-model_eval = ModelEvaluation(evaluation_values, labels);
-disp("Error Rate: " + model_eval.error_rate);
-disp("EER: " + model_eval.equal_error_rate);
-
+L = generateLabels(10,3, 2);
+L