ShareBoost.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.
 *
 * Written (W) 2012 Chiyuan Zhang
 * Copyright (C) 2012 Chiyuan Zhang
 */

#include <algorithm>

#include <shogun/lib/Time.h>
#include <shogun/mathematics/Math.h>
#include <shogun/multiclass/ShareBoost.h>
#include <shogun/multiclass/ShareBoostOptimizer.h>
#include <shogun/features/DenseSubsetFeatures.h>
#include <shogun/labels/RegressionLabels.h>

using namespace shogun;

CShareBoost::CShareBoost()
    :CLinearMulticlassMachine(), m_nonzero_feas(0)
{
    init_sb_params();
}

CShareBoost::CShareBoost(CDenseFeatures<float64_t> *features, CMulticlassLabels *labs, int32_t num_nonzero_feas)
    :CLinearMulticlassMachine(new CMulticlassOneVsRestStrategy(), features, NULL, labs), m_nonzero_feas(num_nonzero_feas)
{
    init_sb_params();
}

void CShareBoost::init_sb_params()
{
    SG_ADD(&m_nonzero_feas, "m_nonzero_feas", "Number of non-zero features", MS_NOT_AVAILABLE);
}

SGVector<int32_t> CShareBoost::get_activeset()
{
    return m_activeset.clone();
}

bool CShareBoost::train_machine(CFeatures* data)
{
    if (data)
        set_features(data);

    if (m_features == NULL)
        SG_ERROR("No features given for training\n")
    if (m_labels == NULL)
        SG_ERROR("No labels given for training\n")

    if (m_nonzero_feas <= 0)
        SG_ERROR("Set a valid (> 0) number of non-zero features to seek before training\n")
    if (m_nonzero_feas >= dynamic_cast<CDenseFeatures<float64_t>*>(m_features)->get_num_features())
        SG_ERROR("It doesn't make sense to use ShareBoost with num non-zero features >= num features in the data\n")

    m_fea = dynamic_cast<CDenseFeatures<float64_t> *>(m_features)->get_feature_matrix();
    m_rho = SGMatrix<float64_t>(m_multiclass_strategy->get_num_classes(), m_fea.num_cols);
    m_rho_norm = SGVector<float64_t>(m_fea.num_cols);
    m_pred = SGMatrix<float64_t>(m_fea.num_cols, m_multiclass_strategy->get_num_classes());
    m_pred.zero();

    m_activeset = SGVector<int32_t>(m_fea.num_rows);
    m_activeset.vlen = 0;

    m_machines->reset_array();
    for (int32_t i=0; i < m_multiclass_strategy->get_num_classes(); ++i)
        m_machines->push_back(new CLinearMachine());

    CTime *timer = new CTime();

    float64_t t_compute_pred = 0; // t of 1st round is 0, since no pred to compute
    for (int32_t t=0; t < m_nonzero_feas; ++t)
    {
        timer->start();
        compute_rho();
        int32_t i_fea = choose_feature();
        m_activeset.vector[m_activeset.vlen] = i_fea;
        m_activeset.vlen += 1;
        float64_t t_choose_feature = timer->cur_time_diff();
        timer->start();
        optimize_coefficients();
        float64_t t_optimize = timer->cur_time_diff();

        SG_SDEBUG(" SB[round %03d]: (%8.4f + %8.4f) sec.\n", t,
                t_compute_pred + t_choose_feature, t_optimize);

        timer->start();
        compute_pred();
        t_compute_pred = timer->cur_time_diff();
    }

    SG_UNREF(timer);

    // release memory
    m_fea = SGMatrix<float64_t>();
    m_rho = SGMatrix<float64_t>();
    m_rho_norm = SGVector<float64_t>();
    m_pred = SGMatrix<float64_t>();

    return true;
}

void CShareBoost::compute_pred()
{
    CDenseFeatures<float64_t> *fea = dynamic_cast<CDenseFeatures<float64_t> *>(m_features);
    CDenseSubsetFeatures<float64_t> *subset_fea = new CDenseSubsetFeatures<float64_t>(fea, m_activeset);
    SG_REF(subset_fea);
    for (int32_t i=0; i < m_multiclass_strategy->get_num_classes(); ++i)
    {
        CLinearMachine *machine = dynamic_cast<CLinearMachine *>(m_machines->get_element(i));
        CRegressionLabels *lab = machine->apply_regression(subset_fea);
        SGVector<float64_t> lab_raw = lab->get_labels();
        std::copy(lab_raw.vector, lab_raw.vector + lab_raw.vlen, m_pred.get_column_vector(i));
        SG_UNREF(machine);
        SG_UNREF(lab);
    }
    SG_UNREF(subset_fea);
}

void CShareBoost::compute_pred(const float64_t *W)
{
    int32_t w_len = m_activeset.vlen;

    for (int32_t i=0; i < m_multiclass_strategy->get_num_classes(); ++i)
    {
        CLinearMachine *machine = dynamic_cast<CLinearMachine *>(m_machines->get_element(i));
        SGVector<float64_t> w(w_len);
        std::copy(W + i*w_len, W + (i+1)*w_len, w.vector);
        machine->set_w(w);
        SG_UNREF(machine);
    }
    compute_pred();
}

void CShareBoost::compute_rho()
{
    CMulticlassLabels *lab = dynamic_cast<CMulticlassLabels *>(m_labels);
    for (int32_t i=0; i < m_rho.num_rows; ++i)
    { // i loop classes
        for (int32_t j=0; j < m_rho.num_cols; ++j)
        { // j loop samples
            int32_t label = lab->get_int_label(j);

            m_rho(i,j) = CMath::exp((label == i) - m_pred(j, label) + m_pred(j, i));
        }
    }

    // normalize
    for (int32_t j=0; j < m_rho.num_cols; ++j)
    {
        m_rho_norm[j] = 0;
        for (int32_t i=0; i < m_rho.num_rows; ++i)
            m_rho_norm[j] += m_rho(i,j);
    }
}

int32_t CShareBoost::choose_feature()
{
    SGVector<float64_t> l1norm(m_fea.num_rows);
    for (int32_t j=0; j < m_fea.num_rows; ++j)
    {
        if (std::find(&m_activeset[0], &m_activeset[m_activeset.vlen], j) !=
                &m_activeset[m_activeset.vlen])
        {
            l1norm[j] = 0;
        }
        else
        {
            l1norm[j] = 0;
            CMulticlassLabels *lab = dynamic_cast<CMulticlassLabels *>(m_labels);
            for (int32_t k=0; k < m_multiclass_strategy->get_num_classes(); ++k)
            {
                float64_t abssum = 0;
                for (int32_t ii=0; ii < m_fea.num_cols; ++ii)
                {
                    abssum += m_fea(j, ii)*(m_rho(k, ii)/m_rho_norm[ii] -
                            (j == lab->get_int_label(ii)));
                }
                l1norm[j] += CMath::abs(abssum);
            }
            l1norm[j] /= m_fea.num_cols;
        }
    }

    return CMath::arg_max(l1norm.vector, 1, l1norm.vlen);
}

void CShareBoost::optimize_coefficients()
{
    ShareBoostOptimizer optimizer(this, false);
    optimizer.optimize();
}

void CShareBoost::set_features(CFeatures *f)
{
    CDenseFeatures<float64_t> *fea = dynamic_cast<CDenseFeatures<float64_t> *>(f);
    if (fea == NULL)
        SG_ERROR("Require DenseFeatures<float64_t>\n")
    CLinearMulticlassMachine::set_features(fea);
}