Made DimensionReductionPreprocessor class templated, adapted existing…

… preprocessors to the new structure
shogun-toolbox · Oct 4, 2011 · 5fab1f0 · 5fab1f0
1 parent c0c4231
commit 5fab1f0
Show file tree

Hide file tree

Showing 15 changed files with 61 additions and 62 deletions.
diff --git a/src/interfaces/modular/Preprocessor.i b/src/interfaces/modular/Preprocessor.i
@@ -16,7 +16,6 @@
 %rename(PruneVarSubMean) CPruneVarSubMean;
 %rename(RandomFourierGaussPreproc) CRandomFourierGaussPreproc;
 
-%rename(DimensionReductionPreprocessor) CDimensionReductionPreprocessor;
 %rename(PCA) CPCA;
 %rename(KernelPCA) CKernelPCA;
 %rename(MultidimensionalScaling) CMultidimensionalScaling;
@@ -51,6 +50,13 @@ namespace shogun
     %template(SimpleCharPreprocessor) CSimplePreprocessor<char>;
 }
 
+/* Templates Class DimensionReductionPreprocessor */
+%include <shogun/preprocessor/DimensionReductionPreprocessor.h>
+namespace shogun
+{
+    %template(DimensionReductionPreprocessor) CDimensionReductionPreprocessor<float64_t>;
+}
+
 /* Templates Class StringPreprocessor*/
 %include <shogun/preprocessor/StringPreprocessor.h>
 namespace shogun
@@ -76,7 +82,6 @@ namespace shogun
 %include <shogun/preprocessor/PruneVarSubMean.h>
 %include <shogun/preprocessor/RandomFourierGaussPreproc.h>
 
-%include <shogun/preprocessor/DimensionReductionPreprocessor.h>
 %include <shogun/preprocessor/PCA.h>
 %include <shogun/preprocessor/KernelPCA.h>
 %include <shogun/preprocessor/MultidimensionalScaling.h>

diff --git a/src/shogun/preprocessor/DiffusionMaps.cpp b/src/shogun/preprocessor/DiffusionMaps.cpp
@@ -27,7 +27,7 @@
 using namespace shogun;
 
 CDiffusionMaps::CDiffusionMaps() :
-		CDimensionReductionPreprocessor()
+		CDimensionReductionPreprocessor<float64_t>()
 {
 	m_t = 10;
 

diff --git a/src/shogun/preprocessor/DiffusionMaps.h b/src/shogun/preprocessor/DiffusionMaps.h
@@ -23,7 +23,7 @@ class CFeatures;
 class CKernel;
 
 /** @brief */
-class CDiffusionMaps: public CDimensionReductionPreprocessor
+class CDiffusionMaps: public CDimensionReductionPreprocessor<float64_t>
 {
 public:
 

diff --git a/src/shogun/preprocessor/DimensionReductionPreprocessor.cpp b/src/shogun/preprocessor/DimensionReductionPreprocessor.cpp
@@ -4,8 +4,11 @@
 
 using namespace shogun;
 
-CDimensionReductionPreprocessor::CDimensionReductionPreprocessor()
-: CSimplePreprocessor<float64_t>()
+namespace shogun 
+{
+template<class ST>
+CDimensionReductionPreprocessor<ST>::CDimensionReductionPreprocessor()
+: CSimplePreprocessor<ST>()
 {
 	m_target_dim = 1;
 	m_distance = new CEuclidianDistance();
@@ -14,45 +17,42 @@ CDimensionReductionPreprocessor::CDimensionReductionPreprocessor()
 	init();
 }
 
-CDimensionReductionPreprocessor::~CDimensionReductionPreprocessor() 
+template<class ST>
+CDimensionReductionPreprocessor<ST>::~CDimensionReductionPreprocessor() 
 {
 	SG_UNREF(m_distance);
 	SG_UNREF(m_kernel);
 }
 
-bool CDimensionReductionPreprocessor::init(CFeatures* data)
+template <class ST>
+bool CDimensionReductionPreprocessor<ST>::init(CFeatures* data)
 {
 	return true;
 }
 
-void CDimensionReductionPreprocessor::cleanup()
+template <class ST>
+void CDimensionReductionPreprocessor<ST>::cleanup()
 {
 
 }
 
-SGMatrix<float64_t> CDimensionReductionPreprocessor::apply_to_feature_matrix(CFeatures* features)
-{
-	return ((CSimpleFeatures<float64_t>*)features)->get_feature_matrix();
-};
-
-SGVector<float64_t> CDimensionReductionPreprocessor::apply_to_feature_vector(SGVector<float64_t> vector)
-{
-	return vector;
-};
-
-EPreprocessorType CDimensionReductionPreprocessor::get_type() const { return P_DIMENSIONREDUCTIONPREPROCESSOR; };
+template<class ST>
+EPreprocessorType CDimensionReductionPreprocessor<ST>::get_type() const { return P_DIMENSIONREDUCTIONPREPROCESSOR; };
 
-void CDimensionReductionPreprocessor::set_target_dim(int32_t dim)
+template<class ST>
+void CDimensionReductionPreprocessor<ST>::set_target_dim(int32_t dim)
 {
 	m_target_dim = dim;
 }
 
-int32_t CDimensionReductionPreprocessor::get_target_dim() const
+template<class ST>
+int32_t CDimensionReductionPreprocessor<ST>::get_target_dim() const
 {
 	return m_target_dim;
 }
 
-int32_t CDimensionReductionPreprocessor::calculate_effective_target_dim(int32_t dim)
+template<class ST>
+int32_t CDimensionReductionPreprocessor<ST>::calculate_effective_target_dim(int32_t dim)
 {
 	if (m_target_dim<0)
 	{
@@ -67,44 +67,47 @@ int32_t CDimensionReductionPreprocessor::calculate_effective_target_dim(int32_t
 		return m_target_dim;
 }
 
-void CDimensionReductionPreprocessor::set_distance(CDistance* distance)
+template<class ST>
+void CDimensionReductionPreprocessor<ST>::set_distance(CDistance* distance)
 {
 	SG_UNREF(m_distance);
 	SG_REF(distance);
 	m_distance = distance;
 }
 
-CDistance* CDimensionReductionPreprocessor::get_distance() const
+template<class ST>
+CDistance* CDimensionReductionPreprocessor<ST>::get_distance() const
 {
 	SG_REF(m_distance);
 	return m_distance;
 }
 
-void CDimensionReductionPreprocessor::set_kernel(CKernel* kernel)
+template<class ST>
+void CDimensionReductionPreprocessor<ST>::set_kernel(CKernel* kernel)
 {
 	SG_UNREF(m_kernel);
 	SG_REF(kernel);
 	m_kernel = kernel;
 }
 
-CKernel* CDimensionReductionPreprocessor::get_kernel() const
+template<class ST>
+CKernel* CDimensionReductionPreprocessor<ST>::get_kernel() const
 {
 	SG_REF(m_kernel);
 	return m_kernel;
 }
 
-int32_t CDimensionReductionPreprocessor::detect_dim(SGMatrix<float64_t> distance_matrix)
+template<class ST>
+void CDimensionReductionPreprocessor<ST>::init()
 {
-	SG_NOTIMPLEMENTED;
-	return 0;
-}
-
-void CDimensionReductionPreprocessor::init()
-{
-	m_parameters->add(&m_target_dim, "target_dim",
+	this->m_parameters->add(&m_target_dim, "target_dim",
 					  "target dimensionality of preprocessor");
-	m_parameters->add((CSGObject**)&m_distance, "distance",
+	this->m_parameters->add((CSGObject**)&m_distance, "distance",
 					  "distance to be used for embedding");
-	m_parameters->add((CSGObject**)&m_kernel, "kernel",
+	this->m_parameters->add((CSGObject**)&m_kernel, "kernel",
 					  "kernel to be used for embedding");
 }
+
+template class CDimensionReductionPreprocessor<float32_t>;
+template class CDimensionReductionPreprocessor<float64_t>;
+}
diff --git a/src/shogun/preprocessor/DimensionReductionPreprocessor.h b/src/shogun/preprocessor/DimensionReductionPreprocessor.h
@@ -26,7 +26,8 @@ class CKernel;
  * class for preprocessors used to lower the dimensionality of given 
  * simple features (dense matrices). 
  */
-class CDimensionReductionPreprocessor: public CSimplePreprocessor<float64_t>
+template <class ST>
+class CDimensionReductionPreprocessor: public CSimplePreprocessor<ST>
 {
 public:
 
@@ -51,12 +52,12 @@ class CDimensionReductionPreprocessor: public CSimplePreprocessor<float64_t>
 	/** apply preproc to feature matrix
 	 * by default does nothing, returns given features' matrix
 	 */
-	virtual SGMatrix<float64_t> apply_to_feature_matrix(CFeatures* features);
+	virtual SGMatrix<ST> apply_to_feature_matrix(CFeatures* features) = 0;
 
 	/** apply preproc to feature vector
 	 * by default does nothing, returns given feature vector
 	 */
-	virtual SGVector<float64_t> apply_to_feature_vector(SGVector<float64_t> vector);
+	virtual SGVector<ST> apply_to_feature_vector(SGVector<ST> vector) = 0;
 
 	/** get name */
 	virtual const char* get_name() const { return "DimensionReductionPreprocessor"; };
@@ -94,15 +95,6 @@ class CDimensionReductionPreprocessor: public CSimplePreprocessor<float64_t>
 	 */
 	CKernel* get_kernel() const;
 
-protected:
-
-	/** detect dimensionality from distance matrix
-	 * NOT YET IMPLEMENTED
-	 * @param distance_matrix distance matrix to be used 
-	 * @return detected dimensionality 
-	 */
-	int32_t detect_dim(SGMatrix<float64_t> distance_matrix);
-
 protected:
 
 	/** calculates effective target dimensionality

diff --git a/src/shogun/preprocessor/KernelPCA.cpp b/src/shogun/preprocessor/KernelPCA.cpp
@@ -26,12 +26,12 @@
 
 using namespace shogun;
 
-CKernelPCA::CKernelPCA() : CDimensionReductionPreprocessor()
+CKernelPCA::CKernelPCA() : CDimensionReductionPreprocessor<float64_t>()
 {
 	init();
 }
 
-CKernelPCA::CKernelPCA(CKernel* k) : CDimensionReductionPreprocessor()
+CKernelPCA::CKernelPCA(CKernel* k) : CDimensionReductionPreprocessor<float64_t>()
 {
 	init();
 	SG_REF(k);

diff --git a/src/shogun/preprocessor/KernelPCA.h b/src/shogun/preprocessor/KernelPCA.h
@@ -32,7 +32,7 @@ class CKernel;
  * Retrieved from http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.32.8744
  *
  */
-class CKernelPCA: public CDimensionReductionPreprocessor
+class CKernelPCA: public CDimensionReductionPreprocessor<float64_t>
 {
 public:
 		/** default constructor

diff --git a/src/shogun/preprocessor/LaplacianEigenmaps.cpp b/src/shogun/preprocessor/LaplacianEigenmaps.cpp
@@ -22,7 +22,7 @@
 using namespace shogun;
 
 CLaplacianEigenmaps::CLaplacianEigenmaps() :
-		CDimensionReductionPreprocessor()
+		CDimensionReductionPreprocessor<float64_t>()
 {
 	m_k = 3;
 	m_tau = 1.0;

diff --git a/src/shogun/preprocessor/LaplacianEigenmaps.h b/src/shogun/preprocessor/LaplacianEigenmaps.h
@@ -20,7 +20,6 @@ namespace shogun
 {
 
 class CFeatures;
-
 class CDistance;
 
 /** @brief the class LaplacianEigenmaps used to preprocess
@@ -45,7 +44,7 @@ class CDistance;
  * memory usage. 
  *
  */
-class CLaplacianEigenmaps: public CDimensionReductionPreprocessor
+class CLaplacianEigenmaps: public CDimensionReductionPreprocessor<float64_t>
 {
 public:
 

diff --git a/src/shogun/preprocessor/LocallyLinearEmbedding.cpp b/src/shogun/preprocessor/LocallyLinearEmbedding.cpp
@@ -97,7 +97,7 @@ struct SPARSEDOT_THREAD_PARAM
 #endif /* DOXYGEN_SHOULD_SKIP_THIS */
 
 CLocallyLinearEmbedding::CLocallyLinearEmbedding() :
-		CDimensionReductionPreprocessor()
+		CDimensionReductionPreprocessor<float64_t>()
 {
 	m_k = 3;
 
@@ -366,7 +366,7 @@ SGMatrix<float64_t> CLocallyLinearEmbedding::find_null_space(SGMatrix<float64_t>
 		// using ARPACK (faster)
 		eigenvalues_vector = SG_MALLOC(float64_t, dimension+1);
 		#ifdef HAVE_ARPACK
-		arpack_xsxupd<float64_t>(matrix.matrix,NULL,N,dimension+1,"LA",3,true,-1e-7,0.0,
+		arpack_xsxupd<float64_t>(matrix.matrix,NULL,N,dimension+1,"LA",3,true,0.0,0.0,
 		                         eigenvalues_vector,matrix.matrix,eigenproblem_status);
 		#endif
 	}

diff --git a/src/shogun/preprocessor/LocallyLinearEmbedding.h b/src/shogun/preprocessor/LocallyLinearEmbedding.h
@@ -46,7 +46,7 @@ class CDistance;
  * false using set_posdef.
  *
  */
-class CLocallyLinearEmbedding: public CDimensionReductionPreprocessor
+class CLocallyLinearEmbedding: public CDimensionReductionPreprocessor<float64_t>
 {
 public:
 

diff --git a/src/shogun/preprocessor/MultidimensionalScaling.cpp b/src/shogun/preprocessor/MultidimensionalScaling.cpp
@@ -57,7 +57,7 @@ struct TRIANGULATION_THREAD_PARAM
 };
 #endif
 
-CMultidimensionalScaling::CMultidimensionalScaling() : CDimensionReductionPreprocessor()
+CMultidimensionalScaling::CMultidimensionalScaling() : CDimensionReductionPreprocessor<float64_t>()
 {
 	m_eigenvalues = SGVector<float64_t>(NULL,0,false);
 	m_landmark_number = 3;

diff --git a/src/shogun/preprocessor/MultidimensionalScaling.h b/src/shogun/preprocessor/MultidimensionalScaling.h
@@ -57,7 +57,7 @@ class CDistance;
  * values (30%-50% of total examples number) is pretty good for the
  * most tasks.
  */
-class CMultidimensionalScaling: public CDimensionReductionPreprocessor
+class CMultidimensionalScaling: public CDimensionReductionPreprocessor<float64_t>
 {
 public:
 

diff --git a/src/shogun/preprocessor/PCA.cpp b/src/shogun/preprocessor/PCA.cpp
@@ -25,7 +25,7 @@
 using namespace shogun;
 
 CPCA::CPCA(bool do_whitening_, EPCAMode mode_, float64_t thresh_)
-: CDimensionReductionPreprocessor(), num_dim(0), m_initialized(false),
+: CDimensionReductionPreprocessor<float64_t>(), num_dim(0), m_initialized(false),
 	m_whitening(do_whitening_), m_mode(mode_), thresh(thresh_)
 {
 	init();

diff --git a/src/shogun/preprocessor/PCA.h b/src/shogun/preprocessor/PCA.h
@@ -45,7 +45,7 @@ enum EPCAMode
  * covariance matrix is of size num_feat*num_feat. Note that vectors don't have
  * to have zero mean as it is substracted.
  */
-class CPCA: public CDimensionReductionPreprocessor
+class CPCA: public CDimensionReductionPreprocessor<float64_t>
 {
 	public: