LARS float32 / floatmax support clean up

Replacing if-SGERROR sets with REQUIRE statements

Fixing memory leaks / implicit type conversion warnings

Adding comment for LARS float support cleanup patch

LARS templated float support cleanup patch

LARS templated float support cleanup patch

src/shogun/labels/DenseLabels.h

@@ -122,8 +122,11 @@ class CDenseLabels : public CLabels
 		template<typename ST>
 		ST get_label_t(int32_t idx)
 		{
+			REQUIRE(idx<get_num_labels(), "The provided index (%d) is out of bounds (the last label has index (%d)).  "
+				"Please ensure that you're using a valid index number.", idx, get_num_labels())
+			REQUIRE(m_labels.vector, "You're attempting to get a label when there are in fact none!  "
+				"Please ensure that you initialized the labels correctly.")
 			int32_t real_num=m_subset_stack->subset_idx_conversion(idx);
-			ASSERT(m_labels.vector && idx<get_num_labels())
 			return m_labels.vector[real_num];
 		}
 
@@ -156,7 +159,7 @@ class CDenseLabels : public CLabels
 				return get_labels_copy_t<ST>();
 
 			SGVector<ST> labels_copy(m_labels.vlen);
-			for(int i = 0; i < m_labels.vlen; ++i)
+			for(index_t i = 0; i < m_labels.vlen; ++i)
 				labels_copy[i] = (ST) m_labels[i];
 
 			return labels_copy;
@@ -174,7 +177,7 @@ class CDenseLabels : public CLabels
 			if (!m_subset_stack->has_subsets())
 			{
 				SGVector<ST> labels_copy(m_labels.vlen);
-				for(int i = 0; i < m_labels.vlen; ++i)
+				for(index_t i = 0; i < m_labels.vlen; ++i)
 					labels_copy[i] = (ST) m_labels[i];
 
 				return labels_copy;

src/shogun/regression/LeastAngleRegression.cpp

@@ -45,7 +45,7 @@ void CLeastAngleRegression::find_max_abs(const std::vector<ST> &vec, const std::
 {
 	imax = -1;
 	vmax = -1;
-	for (index_t i=0; i < vec.size(); ++i)
+	for (size_t i=0; i < vec.size(); ++i)
 	{
 		if (ignore_mask[i])
 			continue;
@@ -88,30 +88,23 @@ void CLeastAngleRegression::plane_rot(ST x0, ST x1,
 
 bool CLeastAngleRegression::train_machine(CFeatures* data)
 {
-	if (!m_labels)
-		SG_ERROR("No labels set\n")
-
-	if (m_labels->get_label_type() != LT_REGRESSION)
-		SG_ERROR("Provided labels (%s) are of type (%d) - they should be regression labels (%d) instead.\n",
-			m_labels->get_name(), m_labels->get_label_type(), LT_REGRESSION, m_labels->get_label_type())
+	REQUIRE(m_labels->get_label_type() == LT_REGRESSION, "Provided labels (%s) are of type (%d) - they should be regression labels (%d) instead.\n"
+		, m_labels->get_name(), m_labels->get_label_type(), LT_REGRESSION, m_labels->get_label_type())
 
 	if (!data)
 	{
-		if(!features)
-			SG_ERROR("No features provided.\n")
-
-		if(features->get_feature_class() != C_DENSE)
-			SG_ERROR("Feature-class (%d) must be of type C_DENSE (%d)\n", features->get_feature_class(), C_DENSE)
+		REQUIRE(features, "No features provided.\n")
+		REQUIRE(features->get_feature_class() == C_DENSE,
+			"Feature-class (%d) must be of type C_DENSE (%d)\n", features->get_feature_class(), C_DENSE)
 
 		data = features;
 	}
 	else
-		if (data->get_feature_class() != C_DENSE)
-			SG_ERROR("Feature-class (%d) must be of type C_DENSE (%d)\n", features->get_feature_class(), C_DENSE)
+		REQUIRE(data->get_feature_class() == C_DENSE,
+			"Feature-class must be of type C_DENSE (%d)\n", data->get_feature_class(), C_DENSE)
 
-	if (data->get_num_vectors() != m_labels->get_num_labels())
-		SG_ERROR("Number of training vectors (%d) does not match number of labels (%d)\n", 
-			data->get_num_vectors(), m_labels->get_num_labels())
+	REQUIRE(data->get_num_vectors() == m_labels->get_num_labels(), "Number of training vectors (%d) does not match number of labels (%d)\n"
+		, data->get_num_vectors(), m_labels->get_num_labels())
 
 	//check for type of CFeatures, then call the appropriate template method
 	if(data->get_feature_type() == F_DREAL)
@@ -123,12 +116,13 @@ bool CLeastAngleRegression::train_machine(CFeatures* data)
 	else
 		SG_ERROR("Feature-type (%d) must be of type F_SHORTREAL (%d), F_DREAL (%d) or F_LONGREAL (%d).\n", 
 			data->get_feature_type(), F_SHORTREAL, F_DREAL, F_LONGREAL)
+
+	return false;
 }
 
 template <typename ST>
 bool CLeastAngleRegression::train_machine_templated(CDenseFeatures<ST> * data)
 {
-
 	std::vector<std::vector<ST>> m_beta_path_t;		
 
 	int32_t n_fea = data->get_num_features();
@@ -200,7 +194,7 @@ bool CLeastAngleRegression::train_machine_templated(CDenseFeatures<ST> * data)
 		map_corr = map_Xy - (map_Xr*map_mu);
 
 		// corr_sign = sign(corr)
-		for (index_t i=0; i < corr.size(); ++i)
+		for (size_t i=0; i < corr.size(); ++i)
 			corr_sign[i] = CMath::sign(corr[i]);
 
 		// find max absolute correlation in inactive set
@@ -351,10 +345,10 @@ bool CLeastAngleRegression::train_machine_templated(CDenseFeatures<ST> * data)
 	}
 
 	//copy m_beta_path_t (of type ST) into m_beta_path
-	for(index_t i = 0; i < m_beta_path_t.size(); ++i)
+	for(size_t i = 0; i < m_beta_path_t.size(); ++i)
 	{
 		std::vector<float64_t> va;
-		for(index_t p = 0; p < m_beta_path_t[i].size(); ++p){
+		for(size_t p = 0; p < m_beta_path_t[i].size(); ++p){
 			va.push_back((float64_t) m_beta_path_t[i][p]);			
 		}
 		m_beta_path.push_back(va);

src/shogun/regression/LeastAngleRegression.h

@@ -124,10 +124,9 @@ class CLeastAngleRegression: public CLinearMachine
 
 	void switch_w(int32_t num_variable)
 	{
-		if (w.vlen <= 0)
-			SG_ERROR("Please train the model before updating its parameters")
-		if (size_t(num_variable) >= m_beta_idx.size() || num_variable < 0)
-			SG_ERROR("cannot switch to an estimator of %d non-zero coefficients", num_variable)
+		REQUIRE(w.vlen > 0,"Please train the model (i.e. run the model's train() method) before updating its weights.\n")
+		REQUIRE(size_t(num_variable) < m_beta_idx.size() && num_variable >= 0,
+			"Cannot switch to an estimator of %d non-zero coefficients.\n", num_variable)
 		if (w.vector == NULL)
 			w = SGVector<float64_t>(w.vlen);
 
@@ -188,13 +187,12 @@ class CLeastAngleRegression: public CLinearMachine
 
 protected:
 	/**
-	* An interface method to - this is called by the superclass's (CLinearMachine)
-	* train() method.  This method checks to see if data is a dense feature vector,
-	* and that it's elements are of type float64_t or float32_t.  It then calls 
-	* train_machine_templated with the appropriate template parameters (either float64_t or
-	* float32_t)
-	* @param data the data being passed to LARs to be trained on
-	* @see train
+	* An interface method used call train_machine_templated - 
+	* this is called by the superclass's train method (@see CLinearMachine::train).  
+	* Checks to see if data is a dense feature vector,
+	* and that it's elements are floating point types.  It then calls 
+	* train_machine_templated with the appropriate template parameters
+	* @param data training data
 	* @see train_machine_templated
 	*/
 	bool train_machine(CFeatures * data);
@@ -210,14 +208,16 @@ class CLeastAngleRegression: public CLinearMachine
 	static void plane_rot(ST x0, ST x1,
 		ST &y0, ST &y1, SGMatrix<ST> &G);
 
+	#ifndef SWIG
 	template <typename ST>
 	static void find_max_abs(const std::vector<ST> &vec, const std::vector<bool> &ignore_mask,
 		int32_t &imax, ST& vmax);
+	#endif
 
 private:
 	/**
 	* A templated specialization of the train_machine method
-	* @param data the data being passed to LARs to be trained on
+	* @param data training data
 	* @see train_machine
 	*/
 	template <typename ST>

tests/unit/regression/lars_unittest.cc

@@ -252,8 +252,23 @@ void lars_n_less_than_d_feature_test_templated()
 	CRegressionLabels* labels=new CRegressionLabels(lab);
 	SG_REF(labels);
 	CLeastAngleRegression* lars=new CLeastAngleRegression(false);
+	SG_REF(lars)
+
 	lars->set_labels(labels);
-	lars->train(features);
+
+	//Catch exceptions thrown when training, clean up
+	try
+	{
+		lars->train(features);
+	}
+	catch(...)
+	{
+		SG_UNREF(lars);
+		SG_UNREF(features);
+		SG_UNREF(labels);
+
+		throw;
+	}
 
 	SGVector<float64_t> active2 = lars->get_w_for_var(2);
 	SGVector<float64_t> active1 = lars->get_w_for_var(1);