Generate Plots

PEL208-Special_Learning_Topics/assignments.cpp

@@ -85,7 +85,27 @@ void mhorvath::runPCAExperimentEx(const MatrixXd &D, const char * const f_label
 	fclose(f);
 
 	cout << f_name << endl;
-
+
+
+	for (int i = 0; i < D.cols(); i++) {
+		sprintf(f_name, "%s_reduction_comp_%d.csv", f_label, i + 1);
+		out_file = filesystem::current_path() / out_path / f_name;
+
+		f = fopen(out_file.string().c_str(), "w");
+
+		for (int l = 0; l < reduction_D[i].rows(); l++) {
+			fprintf(f, "%f", reduction_D[i](l, 0));
+			for (int c = 1; c < reduction_D[i].cols(); c++) {
+				fprintf(f, ",%f", reduction_D[i](l, c));
+			}
+			fprintf(f, "\n");
+		}
+
+		fclose(f);
+		cout << f_name << endl;
+	}
+	cout << endl;
+
 
 	for (int i = 0; i < D.cols(); i++) {
 		sprintf(f_name, "%s_rebuilt_comp_%d.csv", f_label, i + 1);
@@ -252,6 +272,8 @@ void mhorvath::runLDAExperiment(const MatrixXd &X, const vector<string> &classes
 		filesystem::create_directory(out_path); // Create output path
 	}
 
+	cout << "######### " << f_label << " - LDA EXPERIMENT ########## " << endl << endl;
+
 	mhorvath::LDA lda(X, classes);
 
 	cout << "Data =" << endl << X.block(0, 0, p_m, p_n) << endl << endl;
@@ -270,6 +292,8 @@ void mhorvath::runLDAExperiment(const MatrixXd &X, const vector<string> &classes
 	// Compute dimensionality reduction
 	for (int i = 0; i < X.cols(); i++) {
 		reduction_D[i] = lda.transform(i + 1);
+
+		//cout << "Reduction " << i+1 << "=" << endl << reduction_D[i] << endl << endl;
 	}
 
 	// Vector of matrices to store rebuilt data using from 1 to n principal components
@@ -278,6 +302,8 @@ void mhorvath::runLDAExperiment(const MatrixXd &X, const vector<string> &classes
 	// Compute data rebuilding
 	for (unsigned int i = 0; i < n; i++) {
 		rebuilt_D[i] = lda.rebuild(reduction_D[i]);
+
+		//cout << "Rebuild " << i + 1 << "=" << endl << rebuilt_D[i] << endl << endl;
 	}
 
 	// Write all results in file
@@ -301,6 +327,25 @@ void mhorvath::runLDAExperiment(const MatrixXd &X, const vector<string> &classes
 	fclose(f);
 	cout << f_name << endl;
 
+	for (unsigned int i = 0; i < n; i++) {
+		sprintf(f_name, "%s_reduction_lda_%d.csv", f_label, i + 1);
+		out_file = filesystem::current_path() / out_path / f_name;
+
+		f = fopen(out_file.string().c_str(), "w");
+
+		for (int l = 0; l < reduction_D[i].rows(); l++) {
+			//fprintf(f, "%f", reduction_D[i](l, 0));
+			for (int c = 0; c < reduction_D[i].cols(); c++) {
+				fprintf(f, "%f,", reduction_D[i](l, c));
+			}
+			fprintf(f, "%s\n", classes[l].c_str());
+		}
+
+		fclose(f);
+		cout << f_name << endl;
+	}
+	cout << endl;
+
 	for (unsigned int i = 0; i < n; i++) {
 		sprintf(f_name, "%s_rebuilt_lda_%d.csv", f_label, i + 1);
 		out_file = filesystem::current_path() / out_path / f_name;
@@ -324,6 +369,75 @@ void mhorvath::runLDAExperiment(const MatrixXd &X, const vector<string> &classes
 	system("CLS");
 }
 
+void mhorvath::runPCA_LDAExperiment(const MatrixXd &X, const vector<string> &classes, const char * const f_label = "", const char * const out_path = "")
+{
+	unsigned const int n((unsigned int)X.cols()); // Number of features
+	unsigned const int m((unsigned int)X.rows()); // Number of observations
+	unsigned const int p_m((unsigned int)std::min((unsigned int)10, m)); // Limit of lines to print
+	unsigned int p_n((unsigned int)std::min((unsigned int)10, n)); // Limit of lines to print
+	FILE *f; // File used
+	filesystem::path out_file; // Used to build output path
+	char f_name[256]; // Used to build output file name
+	MatrixXd reduction_pca(X.rows(), 2);
+	MatrixXd reduction_lda(X.rows(), 2);
+
+	if (strcmp(out_path, "")) {
+		cout << filesystem::create_directory(out_path); // Create output path
+	}
+
+	cout << "######### " << f_label << " - PCA+LDA EXPERIMENT ########## " << endl << endl;
+
+	mhorvath::PCA pca(X); // Compute PCA
+	reduction_pca = pca.transform(3);
+
+	cout << "Data =" << endl << X.block(0, 0, p_m, p_n) << endl << endl;
+	cout << "Data Mean =" << endl << pca.getOriginalMean().segment(0, p_n) << endl << endl;
+	cout << "DataAdjust =" << endl << pca.getDataAdjust().block(0, 0, p_m, p_n) << endl << endl;
+	cout << "Covariance =" << endl << pca.covariance().block(0, 0, p_n, p_n) << endl << endl;
+	cout << "Eigenvalues =" << endl << pca.values().segment(0, p_n) << endl << endl;
+	cout << "Eigenvectors =" << endl << pca.components().block(0, 0, p_n, p_n) << endl << endl;
+	cout << "Explained Variance =" << endl << pca.explained_variace_ratio().segment(0, p_n) << endl << endl;
+	cout << "Sum of Explained Variance =" << endl << pca.explained_variace_ratio().sum() << endl << endl;
+
+	mhorvath::LDA lda(reduction_pca, classes);
+
+	p_n = 2;
+	reduction_lda = lda.transform(p_n);
+
+	cout << "Data =" << endl << reduction_pca.block(0, 0, p_m, p_n) << endl << endl;
+	cout << "Sb =" << endl << lda.getSb() << endl << endl;
+	cout << "Sw =" << endl << lda.getSw() << endl << endl;
+	cout << "Data Mean =" << endl << lda.data_mean() << endl << endl;
+	cout << "Classes Mean =" << endl << lda.classes_mean() << endl << endl;
+	cout << "Eigenvalues =" << endl << lda.values().segment(0, p_n) << endl << endl;
+	cout << "Eigenvectors =" << endl << lda.components().block(0, 0, p_n, p_n) << endl << endl;
+	cout << "Explained Variance =" << endl << lda.explained_variace_ratio().segment(0, p_n) << endl << endl;
+	cout << "Sum of Explained Variance =" << endl << lda.explained_variace_ratio().sum() << endl << endl;
+
+	cout << "GENERATED FILES: " << endl;
+
+	sprintf(f_name, "%s_pca_lda.csv", f_label);
+	out_file = filesystem::current_path() / out_path / f_name;
+
+	f = fopen(out_file.string().c_str(), "w");
+
+	for (int l = 0; l < reduction_lda.rows(); l++) {
+		fprintf(f, "%f", reduction_lda(l, 0));
+		for (int c = 1; c < reduction_lda.cols(); c++) {
+			fprintf(f, ",%f", reduction_lda(l, c));
+		}
+		fprintf(f, "\n");
+	}
+
+	fclose(f);
+	cout << f_name << endl;
+
+	cout << endl;
+
+	system("PAUSE");
+	system("CLS");
+}
+
 void mhorvath::inClassExample()
 {
 	// In class example: Some Data
@@ -547,7 +661,9 @@ void mhorvath::iris()
 
 	fclose(f); // Close file
 
-	mhorvath::runLDAExperiment(D, classes, ex_label, output_folder);
+	//mhorvath::runLDAExperiment(D, classes, ex_label, output_folder);
+	//mhorvath::runPCAExperimentEx(D, ex_label, output_folder);
+	mhorvath::runPCA_LDAExperiment(D, classes, ex_label, output_folder);
 }
 
 void mhorvath::inClassExampleLDA()

PEL208-Special_Learning_Topics/assignments.h

@@ -6,10 +6,9 @@
 namespace mhorvath {
 	void runPCAExperimentEx(const Eigen::MatrixXd &, const char * const, const char * const);
 	void runPCAExperiment(const Eigen::MatrixXd &, const char * const, const char * const);
-
 	void runLeastSquaresExperiment(const Eigen::MatrixXd &, const Eigen::VectorXd &, const char * const, const char * const);
-
 	void runLDAExperiment(const Eigen::MatrixXd &, const std::vector<std::string> &, const char * const, const char * const);
+	void runPCA_LDAExperiment(const Eigen::MatrixXd &, const std::vector<std::string> &, const char * const, const char * const);
 
 	void inClassExample();
 

PEL208-Special_Learning_Topics/lda.cpp

@@ -14,7 +14,7 @@ bool compEigen2(const tuple<int, double *> &A, const tuple<int, double *> &B) {
 }
 
 mhorvath::LDA::LDA(const MatrixXd &M, const vector<string> &class_vec)
-	: DataAdjust(M.rows(), M.cols()), EigenVectors(M.cols(), M.cols()),
+	: DataAdjust(M), EigenVectors(M.cols(), M.cols()),
 	EigenValues(M.cols()), ExplainedVariance(M.cols()), DataMean(M.colwise().mean()),
 	Sb(MatrixXd::Zero(M.cols(), M.cols())), Sw(MatrixXd::Zero(M.cols(), M.cols()))
 {
@@ -33,8 +33,8 @@ mhorvath::LDA::LDA(const MatrixXd &M, const vector<string> &class_vec)
 	VectorXd evals(n); // Unordered eigenvalues
 	double evalues_sum(0.0); // Sum of eigenvalues (Used to compute proportional explained variance)
 
-	// Subtract the mean
-	this->DataAdjust = M.rowwise() - this->DataMean;
+	//// Subtract the mean
+	//this->DataAdjust = M.rowwise() - this->DataMean;
 
 	// Check number of classes
 	for (unsigned int i = 0; i < m; i++) {
@@ -86,8 +86,8 @@ mhorvath::LDA::LDA(const MatrixXd &M, const vector<string> &class_vec)
 	// Compute Sb matrix
 	it_class = classes.begin();
 	for (unsigned int i = 0; i < g; i++, it_class++) {
-		//RowVectorXd temp = this->ClassMean[i] - this->DataMean;
-		RowVectorXd temp = this->ClassMean[i];
+		RowVectorXd temp = this->ClassMean[i] - this->DataMean;
+		//RowVectorXd temp = this->ClassMean[i];
 		Sb += MatrixXd((temp.transpose() * temp).array() * it_class->second);
 	}
 
@@ -161,11 +161,11 @@ MatrixXd mhorvath::LDA::rebuild(const MatrixXd &A)
 	//return (temp.transpose().inverse().block(0, 0, temp.rows(), A.cols()) *
 	//	A.transpose()).transpose();
 
-	//return (this->EigenVectors.transpose().inverse().block(0, 0, this->EigenVectors.rows(), A.cols()) *
-	//	A.transpose()).transpose();
-
 	return (this->EigenVectors.transpose().inverse().block(0, 0, this->EigenVectors.rows(), A.cols()) *
-		A.transpose()).transpose().rowwise() + this->DataMean;
+		A.transpose()).transpose();
+
+	//return (this->EigenVectors.transpose().inverse().block(0, 0, this->EigenVectors.rows(), A.cols()) *
+	//	A.transpose()).transpose().rowwise() + this->DataMean;
 }
 
 MatrixXd mhorvath::LDA::getSb()

PEL208-Special_Learning_Topics/lda_data/inClassExample.py

@@ -92,36 +92,39 @@
 plt.grid(linestyle=':')
 # plt.plot(comp_X + data_mean[0], comp_Y + data_mean[1], 'b')
 # plt.plot(comp_X + data_mean[0], comp_Y2 + data_mean[1], 'g')
-x = X_orig['1']
-plt.plot(x[:, 0], x[:, 1], 'rx', label='classe 1')
 x = X_orig['2']
 plt.plot(x[:, 0], x[:, 1], 'bo', label='classe 2')
+x = X_orig['1']
+plt.plot(x[:, 0], x[:, 1], 'rx', label='classe 1')
 plt.legend()
 # plt.plot(data_mean[0], data_mean[1], 'kX')
+plt.savefig(os.path.join('plots', 'inClassExample.png'))
 
 plt.show()
 
 plt.grid(linestyle=':')
 # plt.plot(comp_X + data_mean[0], comp_Y + data_mean[1], 'b')
 # plt.plot(comp_X + data_mean[0], comp_Y2 + data_mean[1], 'g')
-x = X_pca['1']
-plt.plot(x[:, 0], x[:, 1], 'rx', label='classe 1')
 x = X_pca['2']
 plt.plot(x[:, 0], x[:, 1], 'bo', label='classe 2')
+x = X_pca['1']
+plt.plot(x[:, 0], x[:, 1], 'rx', label='classe 1')
 plt.legend()
 # plt.plot(data_mean[0], data_mean[1], 'kX')
+plt.savefig(os.path.join('plots', 'inClassExample_pca.png'))
 
 plt.show()
 
 plt.grid(linestyle=':')
 # plt.plot(comp_X + data_mean[0], comp_Y + data_mean[1], 'b')
 # plt.plot(comp_X + data_mean[0], comp_Y2 + data_mean[1], 'g')
-x = X_lda['1']
-plt.plot(x[:, 0], x[:, 1], 'rx', label='classe 1')
 x = X_lda['2']
 plt.plot(x[:, 0], x[:, 1], 'bo', label='classe 2')
+x = X_lda['1']
+plt.plot(x[:, 0], x[:, 1], 'rx', label='classe 1')
 plt.legend()
 # plt.plot(data_mean[0], data_mean[1], 'kX')
+plt.savefig(os.path.join('plots', 'inClassExample_lda.png'))
 
 plt.show()
 

PEL208-Special_Learning_Topics/lda_data/inClassExample_rebuilt_lda_1.csv

@@ -1,11 +1,11 @@
-2.967759,3.754755,1
-2.948741,3.846042,1
-3.105416,3.094005,1
-2.910705,4.028618,1
-3.067379,3.276580,1
-3.319145,2.068104,2
-3.300127,2.159391,2
-3.456801,1.407354,2
-3.281108,2.250679,2
-3.418765,1.589929,2
-3.224054,2.524543,2
+-0.194711,0.934613,1
+-0.213729,1.025901,1
+-0.057055,0.273863,1
+-0.251766,1.208477,1
+-0.095091,0.456439,1
+0.156675,-0.752038,2
+0.137656,-0.660750,2
+0.294331,-1.412788,2
+0.118638,-0.569462,2
+0.256294,-1.230212,2
+0.061583,-0.295599,2

PEL208-Special_Learning_Topics/lda_data/inClassExample_reduction_comp_1.csv

@@ -0,0 +1,11 @@
+2.027224
+0.613573
+-0.073318
+-2.213730
+-2.900621
+3.480745
+2.067094
+1.380203
+0.653442
+-1.447100
+-3.587512

PEL208-Special_Learning_Topics/lda_data/inClassExample_reduction_comp_2.csv

@@ -0,0 +1,11 @@
+2.027224,1.086102
+0.613573,1.046233
+-0.073318,0.319472
+-2.213730,0.966493
+-2.900621,0.239733
+3.480745,-0.287679
+2.067094,-0.327549
+1.380203,-1.054310
+0.653442,-0.367419
+-1.447100,-1.134049
+-3.587512,-0.487028

PEL208-Special_Learning_Topics/lda_data/inClassExample_reduction_lda_1.csv

@@ -0,0 +1,11 @@
+-0.827137,1
+-0.907927,1
+-0.242370,1
+-1.069508,1
+-0.403951,1
+0.665557,2
+0.584767,2
+1.250324,2
+0.503977,2
+1.088743,2
+0.261606,2

PEL208-Special_Learning_Topics/lda_data/inClassExample_reduction_lda_2.csv

@@ -0,0 +1,11 @@
+-0.827137,-1.386889,1
+-0.907927,-2.569824,1
+-0.242370,-3.548804,1
+-1.069508,-4.935693,1
+-0.403951,-5.914673,1
+0.665557,-0.978980,2
+0.584767,-2.161915,2
+1.250324,-3.140896,2
+0.503977,-3.344850,2
+1.088743,-5.506765,2
+0.261606,-6.893654,2