MultitaskLeastSquaresRegression.cpp

Go to the documentation of this file.

/*
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * Copyright (C) 2012 Sergey Lisitsyn
 */

#include <shogun/transfer/multitask/MultitaskLeastSquaresRegression.h>
#include <shogun/transfer/multitask/TaskGroup.h>
#include <shogun/transfer/multitask/TaskTree.h>
#include <shogun/lib/slep/slep_solver.h>
#include <shogun/lib/slep/slep_options.h>

namespace shogun
{

CMultitaskLeastSquaresRegression::CMultitaskLeastSquaresRegression() :
    CMultitaskLinearMachine()
{
    initialize_parameters();
    register_parameters();
}

CMultitaskLeastSquaresRegression::CMultitaskLeastSquaresRegression(
     float64_t z, CDotFeatures* train_features, 
     CRegressionLabels* train_labels, CTaskRelation* task_relation) :
    CMultitaskLinearMachine(train_features,(CLabels*)train_labels,task_relation)
{
    set_z(z);
    initialize_parameters();
    register_parameters();
}

CMultitaskLeastSquaresRegression::~CMultitaskLeastSquaresRegression()
{
}

void CMultitaskLeastSquaresRegression::register_parameters()
{
    SG_ADD(&m_z, "z", "regularization coefficient", MS_AVAILABLE);
    SG_ADD(&m_q, "q", "q of L1/Lq", MS_AVAILABLE);
    SG_ADD(&m_termination, "termination", "termination", MS_NOT_AVAILABLE);
    SG_ADD(&m_regularization, "regularization", "regularization", MS_NOT_AVAILABLE);
    SG_ADD(&m_tolerance, "tolerance", "tolerance", MS_NOT_AVAILABLE);
    SG_ADD(&m_max_iter, "max_iter", "maximum number of iterations", MS_NOT_AVAILABLE);
}

void CMultitaskLeastSquaresRegression::initialize_parameters()
{
    set_z(0.0);
    set_q(2.0);
    set_termination(0);
    set_regularization(0);
    set_tolerance(1e-3);
    set_max_iter(1000);
}

bool CMultitaskLeastSquaresRegression::train_locked_implementation(SGVector<index_t>* tasks)
{
    SG_NOTIMPLEMENTED;
    return false;
}

float64_t CMultitaskLeastSquaresRegression::apply_one(int32_t i)
{
    float64_t dot = features->dense_dot(i,m_tasks_w.get_column_vector(m_current_task),m_tasks_w.num_rows);
    return dot + m_tasks_c[m_current_task];
}

int32_t CMultitaskLeastSquaresRegression::get_max_iter() const
{
    return m_max_iter;
}
int32_t CMultitaskLeastSquaresRegression::get_regularization() const
{
    return m_regularization;
}
int32_t CMultitaskLeastSquaresRegression::get_termination() const
{
    return m_termination;
}
float64_t CMultitaskLeastSquaresRegression::get_tolerance() const
{
    return m_tolerance;
}
float64_t CMultitaskLeastSquaresRegression::get_z() const
{
    return m_z;
}
float64_t CMultitaskLeastSquaresRegression::get_q() const
{
    return m_q;
}

void CMultitaskLeastSquaresRegression::set_max_iter(int32_t max_iter)
{
    ASSERT(max_iter>=0);
    m_max_iter = max_iter;
}
void CMultitaskLeastSquaresRegression::set_regularization(int32_t regularization)
{
    ASSERT(regularization==0 || regularization==1);
    m_regularization = regularization;
}
void CMultitaskLeastSquaresRegression::set_termination(int32_t termination)
{
    ASSERT(termination>=0 && termination<=4);
    m_termination = termination;
}
void CMultitaskLeastSquaresRegression::set_tolerance(float64_t tolerance)
{
    ASSERT(tolerance>0.0);
    m_tolerance = tolerance;
}
void CMultitaskLeastSquaresRegression::set_z(float64_t z)
{
    m_z = z;
}
void CMultitaskLeastSquaresRegression::set_q(float64_t q)
{
    m_q = q;
}

bool CMultitaskLeastSquaresRegression::train_machine(CFeatures* data)
{
    if (data && (CDotFeatures*)data)
        set_features((CDotFeatures*)data);

    ASSERT(features);
    ASSERT(m_labels);

    SGVector<float64_t> y = ((CRegressionLabels*)m_labels)->get_labels();
    
    slep_options options = slep_options::default_options();
    options.n_tasks = m_task_relation->get_num_tasks();
    options.tasks_indices = m_task_relation->get_tasks_indices();
    options.q = m_q;
    options.regularization = m_regularization;
    options.termination = m_termination;
    options.tolerance = m_tolerance;
    options.max_iter = m_max_iter;

    ETaskRelationType relation_type = m_task_relation->get_relation_type();
    switch (relation_type)
    {
        case TASK_GROUP:
        {
            //CTaskGroup* task_group = (CTaskGroup*)m_task_relation;
            options.mode = MULTITASK_GROUP;
            options.loss = LEAST_SQUARES;
            m_tasks_w = slep_solver(features, y.vector, m_z, options).w;
            m_tasks_c = SGVector<float64_t>(options.n_tasks);
            m_tasks_c.zero();
        }
        break;
        case TASK_TREE: 
        {
            CTaskTree* task_tree = (CTaskTree*)m_task_relation;
            SGVector<float64_t> ind_t = task_tree->get_SLEP_ind_t();
            options.ind_t = ind_t.vector;
            options.n_nodes = ind_t.vlen/3;
            options.mode = MULTITASK_TREE;
            options.loss = LEAST_SQUARES;
            m_tasks_w = slep_solver(features, y.vector, m_z, options).w;
            m_tasks_c = SGVector<float64_t>(options.n_tasks);
            m_tasks_c.zero();
        }
        break;
        default: 
            SG_ERROR("Not supported task relation type\n");
    }

    for (int32_t i=0; i<options.n_tasks; i++)
        options.tasks_indices[i].~SGVector<index_t>();
    SG_FREE(options.tasks_indices);

    return true;
}

}