Cleanup dense preprocessors and rename init to fit

src/shogun/metric/LMNNImpl.cpp

@@ -365,7 +365,7 @@ SGMatrix<float64_t> CLMNNImpl::compute_pca_transform(CDenseFeatures<float64_t>*
 			new CDenseFeatures<float64_t>(features->get_feature_matrix().clone());
 	CPruneVarSubMean* mean_substractor =
 			new CPruneVarSubMean(false); // false to avoid variance normalization
-	mean_substractor->init(cloned_features);
+	mean_substractor->fit(cloned_features);
 	mean_substractor->apply_to_feature_matrix(cloned_features);
 
 	// Obtain the linear transform applying PCA

src/shogun/preprocessor/HomogeneousKernelMap.cpp

@@ -43,13 +43,6 @@ CHomogeneousKernelMap::~CHomogeneousKernelMap()
 {
 }
 
-bool CHomogeneousKernelMap::init(CFeatures* features)
-{
-	ASSERT(features->get_feature_class()==C_DENSE)
-	ASSERT(features->get_feature_type()==F_DREAL)
-
-	return true;
-}
 
 void CHomogeneousKernelMap::cleanup()
 {
@@ -163,7 +156,7 @@ void CHomogeneousKernelMap::init()
 
 SGMatrix<float64_t> CHomogeneousKernelMap::apply_to_feature_matrix (CFeatures* features)
 {
-	CDenseFeatures<float64_t>* simple_features = (CDenseFeatures<float64_t>*)features;
+	auto simple_features = features->as<CDenseFeatures<float64_t>>();
 	int32_t num_vectors = simple_features->get_num_vectors ();
 	int32_t num_features = simple_features->get_num_features ();
 

src/shogun/preprocessor/HomogeneousKernelMap.h

@@ -69,9 +69,6 @@ namespace shogun
 			/** destructor */
 			virtual ~CHomogeneousKernelMap();
 
-			/** initialize preprocessor from features */
-			virtual bool init(CFeatures* features);
-
 			/** cleanup */
 			virtual void cleanup();
 

src/shogun/preprocessor/LogPlusOne.cpp

@@ -21,15 +21,6 @@ CLogPlusOne::~CLogPlusOne()
 {
 }
 
-/// initialize preprocessor from features
-bool CLogPlusOne::init(CFeatures* features)
-{
-	ASSERT(features->get_feature_class()==C_DENSE)
-	ASSERT(features->get_feature_type()==F_DREAL)
-
-	return true;
-}
-
 /// clean up allocated memory
 void CLogPlusOne::cleanup()
 {
@@ -56,8 +47,8 @@ bool CLogPlusOne::save(FILE* f)
 /// return pointer to feature_matrix, i.e. f->get_feature_matrix();
 SGMatrix<float64_t> CLogPlusOne::apply_to_feature_matrix(CFeatures* features)
 {
-	SGMatrix<float64_t> feature_matrix =
-			((CDenseFeatures<float64_t>*)features)->get_feature_matrix();
+	auto feature_matrix =
+	    features->as<CDenseFeatures<float64_t>>()->get_feature_matrix();
 
 	for (int32_t i=0; i<feature_matrix.num_cols; i++)
 	{

src/shogun/preprocessor/LogPlusOne.h

@@ -37,9 +37,6 @@ class CLogPlusOne : public CDensePreprocessor<float64_t>
 		/** destructor */
 		virtual ~CLogPlusOne();
 
-		/// initialize preprocessor from features
-		virtual bool init(CFeatures* features);
-
 		/// cleanup
 		virtual void cleanup();
 		/// initialize preprocessor from file

src/shogun/preprocessor/NormOne.cpp

@@ -22,15 +22,6 @@ CNormOne::~CNormOne()
 {
 }
 
-/// initialize preprocessor from features
-bool CNormOne::init(CFeatures* features)
-{
-	ASSERT(features->get_feature_class()==C_DENSE)
-	ASSERT(features->get_feature_type()==F_DREAL)
-
-	return true;
-}
-
 /// clean up allocated memory
 void CNormOne::cleanup()
 {
@@ -57,7 +48,8 @@ bool CNormOne::save(FILE* f)
 /// return pointer to feature_matrix, i.e. f->get_feature_matrix();
 SGMatrix<float64_t> CNormOne::apply_to_feature_matrix(CFeatures* features)
 {
-	SGMatrix<float64_t> feature_matrix=((CDenseFeatures<float64_t>*)features)->get_feature_matrix();
+	auto feature_matrix =
+	    features->as<CDenseFeatures<float64_t>>()->get_feature_matrix();
 
 	for (int32_t i=0; i<feature_matrix.num_cols; i++)
 	{

src/shogun/preprocessor/NormOne.h

@@ -37,8 +37,6 @@ class CNormOne : public CDensePreprocessor<float64_t>
 		/** destructor */
 		virtual ~CNormOne();
 
-		/// initialize preprocessor from features
-		virtual bool init(CFeatures* features);
 		/// cleanup
 		virtual void cleanup();
 		/// initialize preprocessor from file

src/shogun/preprocessor/PNorm.cpp

@@ -35,15 +35,6 @@ CPNorm::~CPNorm ()
 {
 }
 
-/// initialize preprocessor from features
-bool CPNorm::init (CFeatures* features)
-{
-	ASSERT(features->get_feature_class()==C_DENSE)
-	ASSERT(features->get_feature_type()==F_DREAL)
-
-	return true;
-}
-
 /// clean up allocated memory
 void CPNorm::cleanup ()
 {
@@ -70,7 +61,8 @@ bool CPNorm::save (FILE* f)
 /// return pointer to feature_matrix, i.e. f->get_feature_matrix();
 SGMatrix<float64_t> CPNorm::apply_to_feature_matrix (CFeatures* features)
 {
-	SGMatrix<float64_t> feature_matrix=((CDenseFeatures<float64_t>*)features)->get_feature_matrix();
+	auto feature_matrix =
+	    features->as<CDenseFeatures<float64_t>>()->get_feature_matrix();
 
 	for (int32_t i=0; i<feature_matrix.num_cols; i++)
 	{

src/shogun/preprocessor/PNorm.h

@@ -40,8 +40,6 @@ class CPNorm : public CDensePreprocessor<float64_t>
 		/** destructor */
 		virtual ~CPNorm ();
 
-		/// initialize preprocessor from features
-		virtual bool init (CFeatures* features);
 		/// cleanup
 		virtual void cleanup ();
 		/// initialize preprocessor from file

src/shogun/preprocessor/PruneVarSubMean.cpp

@@ -26,84 +26,78 @@ CPruneVarSubMean::~CPruneVarSubMean()
 	cleanup();
 }
 
-/// initialize preprocessor from features
-bool CPruneVarSubMean::init(CFeatures* features)
+void CPruneVarSubMean::fit(CFeatures* features)
 {
-	if (!m_initialized)
-	{
-		ASSERT(features->get_feature_class()==C_DENSE)
-		ASSERT(features->get_feature_type()==F_DREAL)
+	if (m_initialized)
+		cleanup();
 
-		CDenseFeatures<float64_t>* simple_features=(CDenseFeatures<float64_t>*) features;
-		int32_t num_examples = simple_features->get_num_vectors();
-		int32_t num_features = simple_features->get_num_features();
+	auto simple_features = features->as<CDenseFeatures<float64_t>>();
+	int32_t num_examples = simple_features->get_num_vectors();
+	int32_t num_features = simple_features->get_num_features();
 
-		m_mean = SGVector<float64_t>();
-		m_idx = SGVector<int32_t>();
-		m_std = SGVector<float64_t>();;
+	m_mean = SGVector<float64_t>();
+	m_idx = SGVector<int32_t>();
+	m_std = SGVector<float64_t>();
 
-		m_mean.resize_vector(num_features);
-		float64_t* var=SG_MALLOC(float64_t, num_features);
-		int32_t i,j;
+	m_mean.resize_vector(num_features);
+	float64_t* var = SG_MALLOC(float64_t, num_features);
+	int32_t i, j;
 
-		memset(var, 0, num_features*sizeof(float64_t));
-		m_mean.zero();
+	memset(var, 0, num_features * sizeof(float64_t));
+	m_mean.zero();
 
-		SGMatrix<float64_t> feature_matrix = simple_features->get_feature_matrix();
+	auto feature_matrix = simple_features->get_feature_matrix();
 
-		// compute mean
-		for (i=0; i<num_examples; i++)
-		{
-			for (j=0; j<num_features; j++)
-				m_mean[j]+=feature_matrix.matrix[i*num_features+j];
-		}
+	// compute mean
+	for (i = 0; i < num_examples; i++)
+	{
+		for (j = 0; j < num_features; j++)
+			m_mean[j] += feature_matrix.matrix[i * num_features + j];
+	}
 
-		for (j=0; j<num_features; j++)
-			m_mean[j]/=num_examples;
+	for (j = 0; j < num_features; j++)
+		m_mean[j] /= num_examples;
 
-		// compute var
-		for (i=0; i<num_examples; i++)
-		{
-			for (j=0; j<num_features; j++)
-				var[j]+=CMath::sq(m_mean[j]-feature_matrix.matrix[i*num_features+j]);
-		}
+	// compute var
+	for (i = 0; i < num_examples; i++)
+	{
+		for (j = 0; j < num_features; j++)
+			var[j] += CMath::sq(
+			    m_mean[j] - feature_matrix.matrix[i * num_features + j]);
+	}
 
-		int32_t num_ok=0;
-		int32_t* idx_ok=SG_MALLOC(int32_t, num_features);
+	int32_t num_ok = 0;
+	int32_t* idx_ok = SG_MALLOC(int32_t, num_features);
 
-		for (j=0; j<num_features; j++)
-		{
-			var[j]/=num_examples;
+	for (j = 0; j < num_features; j++)
+	{
+		var[j] /= num_examples;
 
-			if (var[j]>=1e-14)
-			{
-				idx_ok[num_ok]=j;
-				num_ok++ ;
-			}
+		if (var[j] >= 1e-14)
+		{
+			idx_ok[num_ok] = j;
+			num_ok++;
 		}
+	}
 
-		SG_INFO("Reducing number of features from %i to %i\n", num_features, num_ok)
+	SG_INFO("Reducing number of features from %i to %i\n", num_features, num_ok)
 
-		m_idx.resize_vector(num_ok);
-		SGVector<float64_t> new_mean(num_ok);
-		m_std.resize_vector(num_ok);
+	m_idx.resize_vector(num_ok);
+	SGVector<float64_t> new_mean(num_ok);
+	m_std.resize_vector(num_ok);
 
-		for (j=0; j<num_ok; j++)
-		{
-			m_idx[j]=idx_ok[j] ;
-			new_mean[j]=m_mean[idx_ok[j]];
-			m_std[j] = std::sqrt(var[idx_ok[j]]);
-		}
-		m_num_idx = num_ok;
-		SG_FREE(idx_ok);
-		SG_FREE(var);
-		m_mean = new_mean;
-
-		m_initialized = true;
-		return true;
+	for (j = 0; j < num_ok; j++)
+	{
+		m_idx[j] = idx_ok[j];
+		new_mean[j] = m_mean[idx_ok[j]];
+		m_std[j] = std::sqrt(var[idx_ok[j]]);
 	}
-	else
-		return false;
+	m_num_idx = num_ok;
+	SG_FREE(idx_ok);
+	SG_FREE(var);
+	m_mean = new_mean;
+
+	m_initialized = true;
 }
 
 /// clean up allocated memory

src/shogun/preprocessor/PruneVarSubMean.h

@@ -35,8 +35,8 @@ class CPruneVarSubMean : public CDensePreprocessor<float64_t>
 		/** destructor */
 		virtual ~CPruneVarSubMean();
 
-		/// initialize preprocessor from features
-		virtual bool init(CFeatures* features);
+		/// Fit preprocessor into features
+		virtual void fit(CFeatures* features);
 
 		/// cleanup
 		virtual void cleanup();

src/shogun/preprocessor/RandomFourierGaussPreproc.cpp

@@ -329,36 +329,26 @@ void CRandomFourierGaussPreproc::set_randomcoefficients(
 
 }
 
-bool CRandomFourierGaussPreproc::init(CFeatures *f) {
-	if (f->get_feature_class() != get_feature_class()) {
-		throw ShogunException(
-				"CRandomFourierGaussPreproc::init (CFeatures *f) requires CDenseFeatures<float64_t> as features\n");
-	}
-	if (f->get_feature_type() != get_feature_type()) {
-		throw ShogunException(
-				"CRandomFourierGaussPreproc::init (CFeatures *f) requires CDenseFeatures<float64_t> as features\n");
-	}
+void CRandomFourierGaussPreproc::fit(CFeatures* f)
+{
 	if (dim_feature_space <= 0) {
 		throw ShogunException(
 				"CRandomFourierGaussPreproc::init (CFeatures *f): dim_feature_space<=0 is not allowed, use void set_dim_feature_space(const int32 dim) before!\n");
 	}
 
 	SG_INFO("calling CRandomFourierGaussPreproc::init(...)\n")
 	int32_t num_features =
-			((CDenseFeatures<float64_t>*) f)->get_num_features();
+	    f->as<CDenseFeatures<float64_t>>()->get_num_features();
 
 	if (!test_rfinited()) {
 		dim_input_space = num_features;
 		init_randomcoefficients();
 		ASSERT( test_rfinited())
-		return true;
 	} else {
 		dim_input_space = num_features;
 		// does not reinit if dimension is the same to avoid overriding a previous call of set_randomcoefficients(...)
-		bool inited = init_randomcoefficients();
-		return inited;
+		init_randomcoefficients();
 	}
-
 }
 
 SGVector<float64_t> CRandomFourierGaussPreproc::apply_to_feature_vector(SGVector<float64_t> vector)
@@ -381,8 +371,6 @@ SGVector<float64_t> CRandomFourierGaussPreproc::apply_to_feature_vector(SGVector
 
 SGMatrix<float64_t> CRandomFourierGaussPreproc::apply_to_feature_matrix(CFeatures* features)
 {
-	init(features);
-
 	// version for case dim_feature_space < dim_input space with direct transformation on feature matrix ??
 
 	int32_t num_vectors = 0;