SVMLin.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) 2006-2009 Soeren Sonnenburg
 * Copyright (C) 2006-2009 Fraunhofer Institute FIRST and Max-Planck-Society
 */

#include <shogun/classifier/svm/SVMLin.h>
#include <shogun/labels/Labels.h>
#include <shogun/mathematics/Math.h>
#include <shogun/lib/external/ssl.h>
#include <shogun/machine/LinearMachine.h>
#include <shogun/features/DotFeatures.h>
#include <shogun/labels/Labels.h>
#include <shogun/labels/BinaryLabels.h>

using namespace shogun;

CSVMLin::CSVMLin()
: CLinearMachine(), C1(1), C2(1), epsilon(1e-5), use_bias(true)
{
}

CSVMLin::CSVMLin(
    float64_t C, CDotFeatures* traindat, CLabels* trainlab)
: CLinearMachine(), C1(C), C2(C), epsilon(1e-5), use_bias(true)
{
    set_features(traindat);
    set_labels(trainlab);
}


CSVMLin::~CSVMLin()
{
}

bool CSVMLin::train_machine(CFeatures* data)
{
    ASSERT(m_labels);

    if (data)
    {
        if (!data->has_property(FP_DOT))
            SG_ERROR("Specified features are not of type CDotFeatures\n");
        set_features((CDotFeatures*) data);
    }

    ASSERT(features);

    SGVector<float64_t> train_labels=((CBinaryLabels*) m_labels)->get_labels();
    int32_t num_feat=features->get_dim_feature_space();
    int32_t num_vec=features->get_num_vectors();

    ASSERT(num_vec==train_labels.vlen);

    struct options Options;
    struct data Data;
    struct vector_double Weights;
    struct vector_double Outputs;

    Data.l=num_vec;
    Data.m=num_vec;
    Data.u=0;
    Data.n=num_feat+1;
    Data.nz=num_feat+1;
    Data.Y=train_labels.vector;
    Data.features=features;
    Data.C = SG_MALLOC(float64_t, Data.l);

    Options.algo = SVM;
    Options.lambda=1/(2*get_C1());
    Options.lambda_u=1/(2*get_C1());
    Options.S=10000;
    Options.R=0.5;
    Options.epsilon = get_epsilon();
    Options.cgitermax=10000;
    Options.mfnitermax=50;
    Options.Cp = get_C2()/get_C1();
    Options.Cn = 1;

    if (use_bias)
        Options.bias=1.0;
    else
        Options.bias=0.0;

    for (int32_t i=0;i<num_vec;i++)
    {
        if(train_labels.vector[i]>0)
            Data.C[i]=Options.Cp;
        else
            Data.C[i]=Options.Cn;
    }
    ssl_train(&Data, &Options, &Weights, &Outputs);
    ASSERT(Weights.vec && Weights.d==num_feat+1);

    float64_t sgn=train_labels.vector[0];
    for (int32_t i=0; i<num_feat+1; i++)
        Weights.vec[i]*=sgn;

    set_w(SGVector<float64_t>(Weights.vec, num_feat));
    set_bias(Weights.vec[num_feat]);

    SG_FREE(Data.C);
    SG_FREE(Outputs.vec);
    return true;
}