Go to the documentation of this file.00001
00002
00003
00004
00005
00006
00007
00008
00009
00010 #include <shogun/transfer/multitask/MultitaskClusteredLogisticRegression.h>
00011 #include <shogun/lib/malsar/malsar_clustered.h>
00012 #include <shogun/lib/malsar/malsar_options.h>
00013 #include <shogun/lib/SGVector.h>
00014
00015 namespace shogun
00016 {
00017
00018 CMultitaskClusteredLogisticRegression::CMultitaskClusteredLogisticRegression() :
00019 CMultitaskLogisticRegression(), m_rho1(0.0), m_rho2(0.0)
00020 {
00021 }
00022
00023 CMultitaskClusteredLogisticRegression::CMultitaskClusteredLogisticRegression(
00024 float64_t rho1, float64_t rho2, CDotFeatures* train_features,
00025 CBinaryLabels* train_labels, CTaskGroup* task_group, int32_t n_clusters) :
00026 CMultitaskLogisticRegression(0.0,train_features,train_labels,(CTaskRelation*)task_group)
00027 {
00028 set_rho1(rho1);
00029 set_rho2(rho2);
00030 set_num_clusters(n_clusters);
00031 }
00032
00033 int32_t CMultitaskClusteredLogisticRegression::get_rho1() const
00034 {
00035 return m_rho1;
00036 }
00037
00038 int32_t CMultitaskClusteredLogisticRegression::get_rho2() const
00039 {
00040 return m_rho2;
00041 }
00042
00043 void CMultitaskClusteredLogisticRegression::set_rho1(float64_t rho1)
00044 {
00045 m_rho1 = rho1;
00046 }
00047
00048 void CMultitaskClusteredLogisticRegression::set_rho2(float64_t rho2)
00049 {
00050 m_rho2 = rho2;
00051 }
00052
00053 int32_t CMultitaskClusteredLogisticRegression::get_num_clusters() const
00054 {
00055 return m_num_clusters;
00056 }
00057
00058 void CMultitaskClusteredLogisticRegression::set_num_clusters(int32_t num_clusters)
00059 {
00060 m_num_clusters = num_clusters;
00061 }
00062
00063 CMultitaskClusteredLogisticRegression::~CMultitaskClusteredLogisticRegression()
00064 {
00065 }
00066
00067 bool CMultitaskClusteredLogisticRegression::train_locked_implementation(SGVector<index_t>* tasks)
00068 {
00069 SGVector<float64_t> y(m_labels->get_num_labels());
00070 for (int32_t i=0; i<y.vlen; i++)
00071 y[i] = ((CBinaryLabels*)m_labels)->get_label(i);
00072
00073 malsar_options options = malsar_options::default_options();
00074 options.termination = m_termination;
00075 options.tolerance = m_tolerance;
00076 options.max_iter = m_max_iter;
00077 options.n_tasks = ((CTaskGroup*)m_task_relation)->get_num_tasks();
00078 options.tasks_indices = tasks;
00079 options.n_clusters = m_num_clusters;
00080
00081 #ifdef HAVE_EIGEN3
00082 malsar_result_t model = malsar_clustered(
00083 features, y.vector, m_rho1, m_rho2, options);
00084
00085 m_tasks_w = model.w;
00086 m_tasks_c = model.c;
00087 #else
00088 SG_WARNING("Please install Eigen3 to use MultitaskClusteredLogisticRegression\n");
00089 m_tasks_w = SGMatrix<float64_t>(((CDotFeatures*)features)->get_dim_feature_space(), options.n_tasks);
00090 m_tasks_c = SGVector<float64_t>(options.n_tasks);
00091 #endif
00092 return true;
00093 }
00094
00095 bool CMultitaskClusteredLogisticRegression::train_machine(CFeatures* data)
00096 {
00097 if (data && (CDotFeatures*)data)
00098 set_features((CDotFeatures*)data);
00099
00100 ASSERT(features);
00101 ASSERT(m_labels);
00102 ASSERT(m_task_relation);
00103
00104 SGVector<float64_t> y(m_labels->get_num_labels());
00105 for (int32_t i=0; i<y.vlen; i++)
00106 y[i] = ((CBinaryLabels*)m_labels)->get_label(i);
00107
00108 malsar_options options = malsar_options::default_options();
00109 options.termination = m_termination;
00110 options.tolerance = m_tolerance;
00111 options.max_iter = m_max_iter;
00112 options.n_tasks = ((CTaskGroup*)m_task_relation)->get_num_tasks();
00113 options.tasks_indices = ((CTaskGroup*)m_task_relation)->get_tasks_indices();
00114 options.n_clusters = m_num_clusters;
00115
00116 #ifdef HAVE_EIGEN3
00117 malsar_result_t model = malsar_clustered(
00118 features, y.vector, m_rho1, m_rho2, options);
00119
00120 m_tasks_w = model.w;
00121 m_tasks_c = model.c;
00122 #else
00123 SG_WARNING("Please install Eigen3 to use MultitaskClusteredLogisticRegression\n");
00124 m_tasks_w = SGMatrix<float64_t>(((CDotFeatures*)features)->get_dim_feature_space(), options.n_tasks);
00125 m_tasks_c = SGVector<float64_t>(options.n_tasks);
00126 #endif
00127
00128 for (int32_t i=0; i<options.n_tasks; i++)
00129 options.tasks_indices[i].~SGVector<index_t>();
00130 SG_FREE(options.tasks_indices);
00131
00132 return true;
00133 }
00134
00135 }