MulticlassLibLinear.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 Sergey Lisitsyn
 * Copyright (C) 2012 Sergey Lisitsyn
 */

#include <shogun/lib/config.h>
#include <shogun/multiclass/MulticlassLibLinear.h>
#include <shogun/multiclass/MulticlassOneVsRestStrategy.h>
#include <shogun/mathematics/Math.h>
#include <shogun/lib/v_array.h>
#include <shogun/lib/Signal.h>
#include <shogun/labels/MulticlassLabels.h>

using namespace shogun;

CMulticlassLibLinear::CMulticlassLibLinear() :
    CLinearMulticlassMachine()
{
    init_defaults();
}

CMulticlassLibLinear::CMulticlassLibLinear(float64_t C, CDotFeatures* features, CLabels* labs) :
    CLinearMulticlassMachine(new CMulticlassOneVsRestStrategy(),features,NULL,labs)
{
    init_defaults();
    set_C(C);
}

void CMulticlassLibLinear::init_defaults()
{
    set_C(1.0);
    set_epsilon(1e-2);
    set_max_iter(10000);
    set_use_bias(false);
    set_save_train_state(false);
    m_train_state = NULL;
}

void CMulticlassLibLinear::register_parameters()
{
    SG_ADD(&m_C, "m_C", "regularization constant",MS_AVAILABLE);
    SG_ADD(&m_epsilon, "m_epsilon", "tolerance epsilon",MS_NOT_AVAILABLE);
    SG_ADD(&m_max_iter, "m_max_iter", "max number of iterations",MS_NOT_AVAILABLE);
    SG_ADD(&m_use_bias, "m_use_bias", "indicates whether bias should be used",MS_NOT_AVAILABLE);
    SG_ADD(&m_save_train_state, "m_save_train_state", "indicates whether bias should be used",MS_NOT_AVAILABLE);
}

CMulticlassLibLinear::~CMulticlassLibLinear()
{
    reset_train_state();
}

SGVector<int32_t> CMulticlassLibLinear::get_support_vectors() const
{
    if (!m_train_state)
        SG_ERROR("Please enable save_train_state option and train machine.\n");

    ASSERT(m_labels && m_labels->get_label_type() == LT_MULTICLASS);

    int32_t num_vectors = m_features->get_num_vectors();
    int32_t num_classes = ((CMulticlassLabels*) m_labels)->get_num_classes();

    v_array<int32_t> nz_idxs;
    nz_idxs.reserve(num_vectors);

    for (int32_t i=0; i<num_vectors; i++)
    {
        for (int32_t y=0; y<num_classes; y++)
        {
            if (CMath::abs(m_train_state->alpha[i*num_classes+y])>1e-6)
            {
                nz_idxs.push(i);
                break;
            }
        }
    }
    int32_t num_nz = nz_idxs.index();
    nz_idxs.reserve(num_nz);
    return SGVector<int32_t>(nz_idxs.begin,num_nz);
}

SGMatrix<float64_t> CMulticlassLibLinear::obtain_regularizer_matrix() const
{
    return SGMatrix<float64_t>();
}

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

    ASSERT(m_features);
    ASSERT(m_labels && m_labels->get_label_type()==LT_MULTICLASS);
    ASSERT(m_multiclass_strategy);

    int32_t num_vectors = m_features->get_num_vectors();
    int32_t num_classes = ((CMulticlassLabels*) m_labels)->get_num_classes();
    int32_t bias_n = m_use_bias ? 1 : 0;

    problem mc_problem;
    mc_problem.l = num_vectors;
    mc_problem.n = m_features->get_dim_feature_space() + bias_n;
    mc_problem.y = SG_MALLOC(float64_t, mc_problem.l);
    for (int32_t i=0; i<num_vectors; i++)
        mc_problem.y[i] = ((CMulticlassLabels*) m_labels)->get_int_label(i);

    mc_problem.x = m_features;
    mc_problem.use_bias = m_use_bias;

    SGMatrix<float64_t> w0 = obtain_regularizer_matrix();

    if (!m_train_state)
        m_train_state = new mcsvm_state();

    float64_t* C = SG_MALLOC(float64_t, num_vectors);
    for (int32_t i=0; i<num_vectors; i++)
        C[i] = m_C;

    CSignal::clear_cancel();

    Solver_MCSVM_CS solver(&mc_problem,num_classes,C,w0.matrix,m_epsilon,
                           m_max_iter,m_max_train_time,m_train_state);
    solver.solve();

    m_machines->reset_array();
    for (int32_t i=0; i<num_classes; i++)
    {
        CLinearMachine* machine = new CLinearMachine();
        SGVector<float64_t> cw(mc_problem.n-bias_n);

        for (int32_t j=0; j<mc_problem.n-bias_n; j++)
            cw[j] = m_train_state->w[j*num_classes+i];

        machine->set_w(cw);

        if (m_use_bias)
            machine->set_bias(m_train_state->w[(mc_problem.n-bias_n)*num_classes+i]);

        m_machines->push_back(machine);
    }

    if (!m_save_train_state)
        reset_train_state();

    SG_FREE(C);
    SG_FREE(mc_problem.y);

    return true;
}