SparseInference.cpp

Go to the documentation of this file.

/*
 * Copyright (c) The Shogun Machine Learning Toolbox
 * Written (W) 2015 Wu Lin
 * Written (W) 2013 Roman Votyakov
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright notice, this
 *    list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright notice,
 *    this list of conditions and the following disclaimer in the documentation
 *    and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * The views and conclusions contained in the software and documentation are those
 * of the authors and should not be interpreted as representing official policies,
 * either expressed or implied, of the Shogun Development Team.
 *
 */
#include <shogun/machine/gp/SparseInference.h>


#include <shogun/machine/gp/GaussianLikelihood.h>
#include <shogun/mathematics/Math.h>
#include <shogun/labels/RegressionLabels.h>
#include <shogun/mathematics/eigen3.h>

using namespace shogun;
using namespace Eigen;

CSparseInference::CSparseInference() : CInference()
{
    init();
}

void CSparseInference::check_features()
{
    REQUIRE(m_features, "Input features not set\n")
}

void CSparseInference::convert_features()
{
    CDotFeatures *feat_type=dynamic_cast<CDotFeatures *>(m_features);

    SGMatrix<float64_t>lat_m(m_inducing_features.matrix,
        m_inducing_features.num_rows,m_inducing_features.num_cols,false);
    CDotFeatures *lat_type=new CDenseFeatures<float64_t>(lat_m);

    REQUIRE(feat_type, "Input features (%s) must be DotFeatures"
        " or one of its subclasses\n", m_features->get_name())
    REQUIRE(feat_type->get_dim_feature_space()==lat_type->get_dim_feature_space(),
        "The dim of feature spaces between"
        " input features (%d) and inducing features (%d) must be same\n",
        feat_type->get_dim_feature_space(),
        lat_type->get_dim_feature_space())
    if((m_features->get_feature_class()!=lat_type->get_feature_class())||
        (m_features->get_feature_type()!=lat_type->get_feature_type()))
    {
        if(m_features->get_feature_class()!=lat_type->get_feature_class())
        {
            SG_WARNING("Input features (%s) and inducing features (%s) are"
                " difference classes\n", m_features->get_name(),
                lat_type->get_name());
        }
        if(m_features->get_feature_type()!=lat_type->get_feature_type())
        {
            SG_WARNING("Input features and inducing features are difference types\n");
        }
        SG_WARNING("Input features may be deleted\n");
        SGMatrix<float64_t> feat_m=feat_type->get_computed_dot_feature_matrix();
        SG_UNREF(m_features);
        m_features=new CDenseFeatures<float64_t>(feat_m);
        SG_REF(m_features);
    }
    SG_UNREF(lat_type);
}

CSparseInference::CSparseInference(CKernel* kern, CFeatures* feat,
        CMeanFunction* m, CLabels* lab, CLikelihoodModel* mod, CFeatures* lat)
        : CInference(kern, feat, m, lab, mod)
{
    init();
    set_inducing_features(lat);
}

void CSparseInference::init()
{
    SG_ADD(&m_inducing_features, "inducing_features", "inducing features",
            MS_AVAILABLE, GRADIENT_AVAILABLE);
    SG_ADD(&m_log_ind_noise, "log_inducing_noise", "noise about inducing potins in log domain",
        MS_AVAILABLE, GRADIENT_AVAILABLE);
    SG_ADD(&m_mu, "mu", "mean vector of the approximation to the posterior", MS_NOT_AVAILABLE);
    SG_ADD(&m_Sigma, "Sigma", "covariance matrix of the approximation to the posterior", MS_NOT_AVAILABLE);
    SG_ADD(&m_ktrtr_diag, "ktrtr_diag", "diagonal elements of kernel matrix m_ktrtr", MS_NOT_AVAILABLE);

    m_log_ind_noise=CMath::log(1e-10);
    m_inducing_features=SGMatrix<float64_t>();
}

void CSparseInference::set_inducing_noise(float64_t noise)
{
    REQUIRE(noise>0, "Noise (%f) for inducing points must be postive",noise);
    m_log_ind_noise=CMath::log(noise);
}

float64_t CSparseInference::get_inducing_noise()
{
    return CMath::exp(m_log_ind_noise);
}

CSparseInference::~CSparseInference()
{
}

void CSparseInference::check_members() const
{
    CInference::check_members();

    REQUIRE(m_inducing_features.num_rows, "Inducing features should not be empty\n")
    REQUIRE(m_inducing_features.num_cols, "Inducing features should not be empty\n")
}

SGVector<float64_t> CSparseInference::get_alpha()
{
    if (parameter_hash_changed())
        update();

    SGVector<float64_t> result(m_alpha);
    return result;
}

SGMatrix<float64_t> CSparseInference::get_cholesky()
{
    if (parameter_hash_changed())
        update();

    SGMatrix<float64_t> result(m_L);
    return result;
}

void CSparseInference::update_train_kernel()
{
    //time complexity can be O(m*n) if the TO DO is done
    check_features();
    convert_features();

    m_kernel->init(m_features, m_features);
    m_ktrtr_diag=m_kernel->get_kernel_diagonal();

    CFeatures* inducing_features=get_inducing_features();

    // create kernel matrix for inducing features
    m_kernel->init(inducing_features, inducing_features);
    m_kuu=m_kernel->get_kernel_matrix();

    // create kernel matrix for inducing and training features
    m_kernel->init(inducing_features, m_features);
    m_ktru=m_kernel->get_kernel_matrix();

    SG_UNREF(inducing_features);
}