bool CMultitaskL12LogisticRegression::train_locked_implementation(SGVector<index_t>* tasks)
{
SGVector<float64_t> y(m_labels->get_num_labels());
for (int32_t i=0; i<y.vlen; i++)
y[i] = ((CBinaryLabels*)m_labels)->get_label(i);
malsar_options options = malsar_options::default_options();
options.termination = m_termination;
options.tolerance = m_tolerance;
options.max_iter = m_max_iter;
options.n_tasks = ((CTaskGroup*)m_task_relation)->get_num_tasks();
options.tasks_indices = tasks;
#ifdef HAVE_EIGEN3
malsar_result_t model = malsar_joint_feature_learning(
features, y.vector, m_rho1, m_rho2, options);
m_tasks_w = model.w;
m_tasks_c = model.c;
#else
SG_WARNING("Please install Eigen3 to use MultitaskL12LogisticRegression\n")
m_tasks_w = SGMatrix<float64_t>(((CDotFeatures*)features)->get_dim_feature_space(), options.n_tasks);
m_tasks_c = SGVector<float64_t>(options.n_tasks);
#endif
return true;
}
bool CMultitaskL12LogisticRegression::train_machine(CFeatures* data)
{
if (data && (CDotFeatures*)data)
set_features((CDotFeatures*)data);
ASSERT(features)
ASSERT(m_labels)
ASSERT(m_task_relation)
SGVector<float64_t> y(m_labels->get_num_labels());
for (int32_t i=0; i<y.vlen; i++)
y[i] = ((CBinaryLabels*)m_labels)->get_label(i);
malsar_options options = malsar_options::default_options();
options.termination = m_termination;
options.tolerance = m_tolerance;
options.max_iter = m_max_iter;
options.n_tasks = ((CTaskGroup*)m_task_relation)->get_num_tasks();
options.tasks_indices = ((CTaskGroup*)m_task_relation)->get_tasks_indices();
malsar_result_t model = malsar_joint_feature_learning(
features, y.vector, self->m_rho1, self->m_rho2, options);
m_tasks_w = model.w;
m_tasks_c = model.c;
SG_FREE(options.tasks_indices);
return true;
}
bool CMultitaskL1L2LogisticRegression::train_machine(CFeatures* data)
{
if (data && (CDotFeatures*)data)
set_features((CDotFeatures*)data);
ASSERT(features);
ASSERT(m_labels);
SGVector<float64_t> y(m_labels->get_num_labels());
for (int32_t i=0; i<y.vlen; i++)
y[i] = ((CBinaryLabels*)m_labels)->get_label(i);
slep_options options = slep_options::default_options();
options.termination = m_termination;
options.tolerance = m_tolerance;
options.max_iter = m_max_iter;
SGVector<index_t> ind = ((CIndexBlockGroup*)m_task_relation)->get_SLEP_ind();
options.ind = ind.vector;
options.n_tasks = ind.vlen-1;
#ifdef HAVE_EIGEN3
slep_result_t model = malsar_joint_feature_learning(
features, y.vector, m_rho1, m_rho2, options);
m_tasks_w = model.w;
m_tasks_c = model.c;
#endif
ASSERT(m_task_relation);
return true;
}
bool CMultitaskL12LogisticRegression::train_locked_implementation(SGVector<index_t>* tasks)
{
SGVector<float64_t> y(m_labels->get_num_labels());
for (int32_t i=0; i<y.vlen; i++)
y[i] = ((CBinaryLabels*)m_labels)->get_label(i);
malsar_options options = malsar_options::default_options();
options.termination = m_termination;
options.tolerance = m_tolerance;
options.max_iter = m_max_iter;
options.n_tasks = ((CTaskGroup*)m_task_relation)->get_num_tasks();
options.tasks_indices = tasks;
malsar_result_t model = malsar_joint_feature_learning(
features, y.vector, self->m_rho1, self->m_rho2, options);
m_tasks_w = model.w;
m_tasks_c = model.c;
return true;
}