Add Gaussian generator in DataGenerator

and add unit test based on DataGenerator for SVMOcas
shogun-toolbox · Aug 10, 2012 · bebc04e · bebc04e
1 parent 529eea9
commit bebc04e
Show file tree

Hide file tree

Showing 3 changed files with 46 additions and 9 deletions.
diff --git a/src/shogun/features/DataGenerator.cpp b/src/shogun/features/DataGenerator.cpp
@@ -9,6 +9,7 @@
 
 #include <shogun/features/DataGenerator.h>
 #include <shogun/mathematics/Math.h>
+#include <shogun/distributions/Gaussian.h>
 
 using namespace shogun;
 
@@ -76,3 +77,31 @@ SGMatrix<float64_t> CDataGenerator::generate_sym_mix_gauss(index_t m,
 
 	return result;
 }
+
+SGMatrix<float64_t> CDataGenerator::generate_gaussians(index_t m, index_t n, index_t dim)
+{
+	/* evtl. allocate space */
+	SGMatrix<float64_t> result =
+		SGMatrix<float64_t>::get_allocated_matrix(dim, n*m);
+
+
+	for (index_t i = 0; i < n; ++i)
+	{
+		SGVector<float64_t> mean(dim);
+		SGMatrix<float64_t> cov = SGMatrix<float64_t>::create_identity_matrix(dim, 1.0);
+
+		mean.random(0.0, 10.0);
+
+		CGaussian* g = new CGaussian(mean, cov, DIAG);
+		for (index_t j = 0; j < m; ++j)
+		{
+			SGVector<float64_t> v = g->sample();
+			memcpy(result.matrix+((i+j)*result.num_rows), v.vector, dim*sizeof(float64_t));
+		}
+
+		SG_UNREF(g);
+	}
+
+	return result;
+}
+
diff --git a/src/shogun/features/DataGenerator.h b/src/shogun/features/DataGenerator.h
@@ -68,6 +68,14 @@ class CDataGenerator: public CSGObject
 			float64_t d, float64_t angle,
 			SGMatrix<float64_t> target=SGMatrix<float64_t>());
 
+	/*
+	 *
+	 * @param m number of samples
+	 * @param n number of gaussians
+	 * @
+	 */
+	static SGMatrix<float64_t> generate_gaussians(index_t m, index_t n, index_t dim);
+
 	inline virtual const char* get_name() const { return "DataGenerator"; }
 
 private:

diff --git a/tests/unit/classifier/svm/SVMOcas_unittest.cc b/tests/unit/classifier/svm/SVMOcas_unittest.cc
@@ -10,23 +10,23 @@ using namespace shogun;
 
 TEST(SVMOcasTest,train)
 {
-	index_t num_samples = 100, dim = 10;
-	float64_t mean_shift = 1.0;
+	index_t num_samples = 50;
+	float64_t dist = 1.0, angle = 0.0;
 	CMath::init_random(5);
 	SGMatrix<float64_t> data =
-		CDataGenerator::generate_mean_data(num_samples, dim, mean_shift);
+		CDataGenerator::generate_gaussians(num_samples, 2, 2);
 	CDenseFeatures<float64_t> features(data);
 
-	SGVector<index_t> train_idx(100), test_idx(100);
-	SGVector<float64_t> labels(100);
+	SGVector<index_t> train_idx(num_samples), test_idx(num_samples);
+	SGVector<float64_t> labels(num_samples);
 	for (index_t i = 0, j = 0; i < data.num_cols; ++i)
 	{
 		if (i % 2 == 0)
 			train_idx[j] = i;
 		else
 			test_idx[j++] = i;
 
-		labels[i/2] = (i < 100) ? 1.0 : -1.0;
+		labels[i/2] = (i < data.num_cols/2) ? 1.0 : -1.0;
 	}
 
 	CDenseFeatures<float64_t>* train_feats = (CDenseFeatures<float64_t>*)features.copy_subset(train_idx);
@@ -38,12 +38,12 @@ TEST(SVMOcasTest,train)
 	ocas->train();
 
 	CLabels* pred = ocas->apply(test_feats);
-	CROCEvaluation roc;
-	float64_t err = CMath::abs(roc.evaluate(pred, ground_truth)-0.7684);
-	EXPECT_TRUE(err < 10E-12);
+	for (int i = 0; i < num_samples; ++i)
+		EXPECT_EQ(((CBinaryLabels*)pred)->get_int_label(i), ((CBinaryLabels)labels).get_int_label(i));
 
 	SG_UNREF(ocas);
 	SG_UNREF(train_feats);
 	SG_UNREF(test_feats);
+	SG_UNREF(pred);
 }