LibSVM.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) 1999-2009 Soeren Sonnenburg
 * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society
 */

#include <shogun/classifier/svm/LibSVM.h>
#include <shogun/io/SGIO.h>
#include <shogun/labels/BinaryLabels.h>

using namespace shogun;

CLibSVM::CLibSVM(LIBSVM_SOLVER_TYPE st)
: CSVM(), model(NULL), solver_type(st)
{
}

CLibSVM::CLibSVM(float64_t C, CKernel* k, CLabels* lab)
: CSVM(C, k, lab), model(NULL), solver_type(LIBSVM_C_SVC)
{
    problem = svm_problem();
}

CLibSVM::~CLibSVM()
{
}


bool CLibSVM::train_machine(CFeatures* data)
{
    struct svm_node* x_space;

    ASSERT(m_labels && m_labels->get_num_labels());
    ASSERT(m_labels->get_label_type() == LT_BINARY);

    if (data)
    {
        if (m_labels->get_num_labels() != data->get_num_vectors())
        {
            SG_ERROR("%s::train_machine(): Number of training vectors (%d) does"
                    " not match number of labels (%d)\n", get_name(),
                    data->get_num_vectors(), m_labels->get_num_labels());
        }
        kernel->init(data, data);
    }

    problem.l=m_labels->get_num_labels();
    SG_INFO( "%d trainlabels\n", problem.l);

    // set linear term
    if (m_linear_term.vlen>0)
    {
        if (m_labels->get_num_labels()!=m_linear_term.vlen)
            SG_ERROR("Number of training vectors does not match length of linear term\n");

        // set with linear term from base class
        problem.pv = get_linear_term_array();
    }
    else
    {
        // fill with minus ones
        problem.pv = SG_MALLOC(float64_t, problem.l);

        for (int i=0; i!=problem.l; i++)
            problem.pv[i] = -1.0;
    }

    problem.y=SG_MALLOC(float64_t, problem.l);
    problem.x=SG_MALLOC(struct svm_node*, problem.l);
    problem.C=SG_MALLOC(float64_t, problem.l);

    x_space=SG_MALLOC(struct svm_node, 2*problem.l);

    for (int32_t i=0; i<problem.l; i++)
    {
        problem.y[i]=((CBinaryLabels*) m_labels)->get_label(i);
        problem.x[i]=&x_space[2*i];
        x_space[2*i].index=i;
        x_space[2*i+1].index=-1;
    }

    int32_t weights_label[2]={-1,+1};
    float64_t weights[2]={1.0,get_C2()/get_C1()};

    ASSERT(kernel && kernel->has_features());
    ASSERT(kernel->get_num_vec_lhs()==problem.l);

    param.svm_type=solver_type; // C SVM or NU_SVM
    param.kernel_type = LINEAR;
    param.degree = 3;
    param.gamma = 0;    // 1/k
    param.coef0 = 0;
    param.nu = get_nu();
    param.kernel=kernel;
    param.cache_size = kernel->get_cache_size();
    param.max_train_time = m_max_train_time;
    param.C = get_C1();
    param.eps = epsilon;
    param.p = 0.1;
    param.shrinking = 1;
    param.nr_weight = 2;
    param.weight_label = weights_label;
    param.weight = weights;
    param.use_bias = get_bias_enabled();

    const char* error_msg = svm_check_parameter(&problem, &param);

    if(error_msg)
        SG_ERROR("Error: %s\n",error_msg);

    model = svm_train(&problem, &param);

    if (model)
    {
        ASSERT(model->nr_class==2);
        ASSERT((model->l==0) || (model->l>0 && model->SV && model->sv_coef && model->sv_coef[0]));

        int32_t num_sv=model->l;

        create_new_model(num_sv);
        CSVM::set_objective(model->objective);

        float64_t sgn=model->label[0];

        set_bias(-sgn*model->rho[0]);

        for (int32_t i=0; i<num_sv; i++)
        {
            set_support_vector(i, (model->SV[i])->index);
            set_alpha(i, sgn*model->sv_coef[0][i]);
        }

        SG_FREE(problem.x);
        SG_FREE(problem.y);
        SG_FREE(problem.pv);
        SG_FREE(problem.C);


        SG_FREE(x_space);

        svm_destroy_model(model);
        model=NULL;
        return true;
    }
    else
        return false;
}