avoid specialisation of labels in stratified xvalidation (#4313)

examples/meta/src/evaluation/cross_validation_support_vector_machine.sg

@@ -2,14 +2,14 @@ CSVFile f_feats("../../data/fm_train_real.dat")
 CSVFile f_labels("../../data/label_train_twoclass.dat")
 
 Features feats = features(f_feats)
-BinaryLabels labels(f_labels)
+Labels labs = labels(f_labels)
 
 Machine svm = machine("LibLinear")
 svm.put("liblinear_solver_type", enum LIBLINEAR_SOLVER_TYPE.L2R_L2LOSS_SVC)
 
-StratifiedCrossValidationSplitting splitting_strategy(labels, 5)
+StratifiedCrossValidationSplitting splitting_strategy(labs, 5)
 AccuracyMeasure evaluation_criterion()
-CrossValidation cross(svm, feats, labels, splitting_strategy, evaluation_criterion)
+CrossValidation cross(svm, feats, labs, splitting_strategy, evaluation_criterion)
 cross.set_num_runs(2)
 
 CrossValidationResult result = CrossValidationResult:obtain_from_generic(cross.evaluate())

src/shogun/evaluation/StratifiedCrossValidationSplitting.cpp

@@ -4,9 +4,10 @@
  * Authors: Heiko Strathmann, Soeren Sonnenburg, Thoralf Klein, Viktor Gal
  */
 
+#include <shogun/base/range.h>
 #include <shogun/evaluation/StratifiedCrossValidationSplitting.h>
-#include <shogun/labels/Labels.h>
 #include <shogun/labels/BinaryLabels.h>
+#include <shogun/labels/Labels.h>
 #include <shogun/labels/MulticlassLabels.h>
 
 using namespace shogun;
@@ -25,44 +26,31 @@ CStratifiedCrossValidationSplitting::CStratifiedCrossValidationSplitting(
 	 * if there are of a class less labels than num_subsets, the class will not
 	 * appear in every subset, leading to subsets of only one class in the
 	 * extreme case of a two class labeling. */
-	SGVector<float64_t> classes;
 
-	int32_t num_classes=2;
-	if (labels->get_label_type() == LT_MULTICLASS)
-	{
-		num_classes=((CMulticlassLabels*) labels)->get_num_classes();
-		classes=((CMulticlassLabels*) labels)->get_unique_labels();
-	}
-	else if (labels->get_label_type() == LT_BINARY)
-	{
-		classes=SGVector<float64_t>(2);
-		classes[0]=-1;
-		classes[1]=+1;
-	}
-	else
-	{
-		SG_ERROR("Multiclass or binary labels required for stratified crossvalidation\n")
-	}
+	auto dense_labels = labels->as<CDenseLabels>();
+	auto classes = dense_labels->get_labels().unique();
 
-	SGVector<index_t> labels_per_class(num_classes);
+	SGVector<index_t> labels_per_class(classes.size());
 
-	for (index_t i=0; i<num_classes; ++i)
+	for (auto i : range(classes.size()))
 	{
-		labels_per_class.vector[i]=0;
-		for (index_t j=0; j<labels->get_num_labels(); ++j)
+		labels_per_class[i] = 0;
+		for (auto j : range(labels->get_num_labels()))
 		{
-			if (classes.vector[i]==((CDenseLabels*) labels)->get_label(j))
-				labels_per_class.vector[i]++;
+			if (classes[i] == dense_labels->get_label(j))
+				labels_per_class[i]++;
 		}
 	}
 
-	for (index_t i=0; i<num_classes; ++i)
+	for (index_t i = 0; i < classes.size(); ++i)
 	{
-		if (labels_per_class.vector[i]<num_subsets)
+		if (labels_per_class[i] < num_subsets)
 		{
-			SG_WARNING("There are only %d labels of class %.18g, but %d "
-					"subsets. Labels of that class will not appear in every "
-					"subset!\n", labels_per_class.vector[i], classes.vector[i], num_subsets);
+			SG_WARNING(
+			    "There are only %d labels of class %.18g, but %d "
+			    "subsets. Labels of that class will not appear in every "
+			    "subset!\n",
+			    labels_per_class[i], classes[i], num_subsets);
 		}
 	}
 
@@ -75,35 +63,21 @@ void CStratifiedCrossValidationSplitting::build_subsets()
 	reset_subsets();
 	m_is_filled=true;
 
-	SGVector<float64_t> unique_labels;
-
-	if (m_labels->get_label_type() == LT_MULTICLASS)
-	{
-		unique_labels=((CMulticlassLabels*) m_labels)->get_unique_labels();
-	}
-	else if (m_labels->get_label_type() == LT_BINARY)
-	{
-		unique_labels=SGVector<float64_t>(2);
-		unique_labels[0]=-1;
-		unique_labels[1]=+1;
-	}
-	else
-	{
-		SG_ERROR("Multiclass or binary labels required for stratified crossvalidation\n")
-	}
+	auto dense_labels = m_labels->as<CDenseLabels>();
+	auto classes = dense_labels->get_labels().unique();
 
 	/* for every label, build set for indices */
 	CDynamicObjectArray label_indices;
-	for (index_t i=0; i<unique_labels.vlen; ++i)
+	for (auto i : range(classes.size()))
 		label_indices.append_element(new CDynamicArray<index_t> ());
 
 	/* fill set with indices, for each label type ... */
-	for (index_t i=0; i<unique_labels.vlen; ++i)
+	for (auto i : range(classes.size()))
 	{
 		/* ... iterate over all labels and add indices with same label to set */
-		for (index_t j=0; j<m_labels->get_num_labels(); ++j)
+		for (auto j : range(m_labels->get_num_labels()))
 		{
-			if (((CDenseLabels*) m_labels)->get_label(j)==unique_labels.vector[i])
+			if (dense_labels->get_label(j) == classes[i])
 			{
 				CDynamicArray<index_t>* current=(CDynamicArray<index_t>*)
 						label_indices.get_element(i);
@@ -127,7 +101,7 @@ void CStratifiedCrossValidationSplitting::build_subsets()
 
 	/* distribute labels to subsets for all label types */
 	index_t target_set=0;
-	for (index_t i=0; i<unique_labels.vlen; ++i)
+	for (auto i : range(classes.size()))
 	{
 		/* current index set for current label */
 		CDynamicArray<index_t>* current=(CDynamicArray<index_t>*)

src/shogun/lib/SGVector.cpp

@@ -884,6 +884,22 @@ int32_t SGVector<complex128_t>::unique(complex128_t* output, int32_t size)
 	return j;
 }
 
+template <class T>
+SGVector<T> SGVector<T>::unique()
+{
+	SGVector<T> result = clone();
+	auto new_size = unique(result.data(), result.size());
+	result.resize_vector(new_size);
+	return result;
+}
+
+template <>
+SGVector<complex128_t> SGVector<complex128_t>::unique()
+{
+	SG_SNOTIMPLEMENTED
+	return SGVector<complex128_t>();
+}
+
 template <class T>
 SGVector<index_t> SGVector<T>::find(T elem)
 {

src/shogun/lib/SGVector.h

@@ -473,6 +473,9 @@ template<class T> class SGVector : public SGReferencedData
 		 */
 		static int32_t unique(T* output, int32_t size);
 
+		/** Returns new vector with sorted unique elements of current */
+		SGVector<T> unique();
+
 		/** Display array size */
 		void display_size() const;
 

tests/unit/lib/SGVector_unittest.cc

@@ -428,3 +428,16 @@ TEST(SGVectorTest,unique)
     for (index_t i = 0; i < num_unique; ++i)
 		EXPECT_EQ(i+1, vec[i]);
 }
+
+TEST(SGVectorTest, unique_method)
+{
+	SGVector<int32_t> vec{1, 4, 3, 1, 4, 3, 3};
+	auto vec_unique = vec.unique();
+
+	ASSERT_NE(vec_unique.data(), nullptr);
+	ASSERT_EQ(3, vec_unique.size());
+	EXPECT_NE(vec_unique.data(), vec.data());
+	EXPECT_EQ(vec_unique[0], 1);
+	EXPECT_EQ(vec_unique[1], 3);
+	EXPECT_EQ(vec_unique[2], 4);
+}