GaussianProcessClassification.cpp

Go to the documentation of this file.

/*
 * Copyright (c) The Shogun Machine Learning Toolbox
 * Written (w) 2014 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.
 *
 * Code adapted from
 * Gaussian Process Machine Learning Toolbox
 * http://www.gaussianprocess.org/gpml/code/matlab/doc/
 * and
 * https://gist.github.com/yorkerlin/8a36e8f9b298aa0246a4
 */

#include <shogun/lib/config.h>

#ifdef HAVE_EIGEN3

#include <shogun/classifier/GaussianProcessClassification.h>
#include <shogun/mathematics/Math.h>
#include <shogun/machine/gp/SingleFITCLaplacianInferenceMethod.h>

using namespace shogun;

CGaussianProcessClassification::CGaussianProcessClassification()
    : CGaussianProcessMachine()
{
}

CGaussianProcessClassification::CGaussianProcessClassification(
        CInferenceMethod* method) : CGaussianProcessMachine(method)
{
    // set labels
    m_labels=method->get_labels();
}

CGaussianProcessClassification::~CGaussianProcessClassification()
{
}

CMulticlassLabels* CGaussianProcessClassification::apply_multiclass(CFeatures* data)
{
    // check whether given combination of inference method and likelihood
    // function supports classification
    REQUIRE(m_method, "Inference method should not be NULL\n")
    CLikelihoodModel* lik=m_method->get_model();
    REQUIRE(m_method->supports_multiclass(), "%s with %s doesn't support "
            "multi classification\n", m_method->get_name(), lik->get_name())
    SG_UNREF(lik);

    // if regression data equals to NULL, then apply classification on training
    // features
    if (!data)
    {
        if (m_method->get_inference_type()==INF_SPARSE)
        {
            SG_NOTIMPLEMENTED
        }
        else
            data=m_method->get_features();
    }
    else
        SG_REF(data);

    const index_t n=data->get_num_vectors();
    SGVector<float64_t> mean=get_mean_vector(data);
    const index_t C=mean.vlen/n;
    SGVector<index_t> lab(n);
    for (index_t idx=0; idx<n; idx++)
    {
        int32_t cate=CMath::arg_max(mean.vector+idx*C, 1, C);
        lab[idx]=cate;
    }
    CMulticlassLabels *result=new CMulticlassLabels();
    result->set_int_labels(lab);

    SG_UNREF(data);

    return result;
}

CBinaryLabels* CGaussianProcessClassification::apply_binary(
        CFeatures* data)
{
    // check whether given combination of inference method and likelihood
    // function supports classification
    REQUIRE(m_method, "Inference method should not be NULL\n")
    CLikelihoodModel* lik=m_method->get_model();
    REQUIRE(m_method->supports_binary(), "%s with %s doesn't support "
            "binary classification\n", m_method->get_name(), lik->get_name())
    SG_UNREF(lik);

    // if regression data equals to NULL, then apply classification on training
    // features
    if (!data)
    {
        if (m_method->get_inference_type()==INF_FITC_LAPLACIAN_SINGLE)
        {
            CSingleFITCLaplacianInferenceMethod* fitc_method=
                CSingleFITCLaplacianInferenceMethod::obtain_from_generic(m_method);
            data=fitc_method->get_inducing_features();
            SG_UNREF(fitc_method);
        }
        else
            data=m_method->get_features();
    }
    else
        SG_REF(data);

    CBinaryLabels* result=new CBinaryLabels(get_mean_vector(data));
    SG_UNREF(data);

    return result;
}

bool CGaussianProcessClassification::train_machine(CFeatures* data)
{
    // check whether given combination of inference method and likelihood
    // function supports classification
    REQUIRE(m_method, "Inference method should not be NULL\n")
    CLikelihoodModel* lik=m_method->get_model();
    REQUIRE(m_method->supports_binary() || m_method->supports_multiclass(), "%s with %s doesn't support "
            "classification\n", m_method->get_name(), lik->get_name())
    SG_UNREF(lik);

    if (data)
    {
        // set inducing features for FITC inference method
        if (m_method->get_inference_type()==INF_FITC_LAPLACIAN_SINGLE)
        {
            CSingleFITCLaplacianInferenceMethod* fitc_method=
                CSingleFITCLaplacianInferenceMethod::obtain_from_generic(m_method);
            fitc_method->set_inducing_features(data);
            SG_UNREF(fitc_method);
        }
        else
            m_method->set_features(data);
    }

    // perform inference
    m_method->update();

    return true;
}

SGVector<float64_t> CGaussianProcessClassification::get_mean_vector(
        CFeatures* data)
{
    // check whether given combination of inference method and likelihood
    // function supports classification
    REQUIRE(m_method, "Inference method should not be NULL\n")
    CLikelihoodModel* lik=m_method->get_model();
    REQUIRE(m_method->supports_binary() || m_method->supports_multiclass(),
        "%s with %s doesn't support classification\n", m_method->get_name(), lik->get_name())

    SG_REF(data);
    SGVector<float64_t> mu=get_posterior_means(data);
    SGVector<float64_t> s2=get_posterior_variances(data);
    SG_UNREF(data);

    // evaluate mean
    mu=lik->get_predictive_means(mu, s2);
    SG_UNREF(lik);

    return mu;
}

SGVector<float64_t> CGaussianProcessClassification::get_variance_vector(
        CFeatures* data)
{
    // check whether given combination of inference method and
    // likelihood function supports classification
    REQUIRE(m_method, "Inference method should not be NULL\n")
    CLikelihoodModel* lik=m_method->get_model();
    REQUIRE(m_method->supports_binary() || m_method->supports_multiclass(),
        "%s with %s doesn't support classification\n", m_method->get_name(), lik->get_name())

    SG_REF(data);
    SGVector<float64_t> mu=get_posterior_means(data);
    SGVector<float64_t> s2=get_posterior_variances(data);
    SG_UNREF(data);

    // evaluate variance
    s2=lik->get_predictive_variances(mu, s2);
    SG_UNREF(lik);

    return s2;
}

SGVector<float64_t> CGaussianProcessClassification::get_probabilities(
        CFeatures* data)
{
    // check whether given combination of inference method and likelihood
    // function supports classification
    REQUIRE(m_method, "Inference method should not be NULL\n")
    CLikelihoodModel* lik=m_method->get_model();
    REQUIRE(m_method->supports_binary() || m_method->supports_multiclass(),
        "%s with %s doesn't support classification\n", m_method->get_name(), lik->get_name())

    SG_REF(data);
    SGVector<float64_t> mu=get_posterior_means(data);
    SGVector<float64_t> s2=get_posterior_variances(data);
    SG_UNREF(data);

    // evaluate log probabilities
    SGVector<float64_t> p=lik->get_predictive_log_probabilities(mu, s2);
    SG_UNREF(lik);

    // evaluate probabilities
    for (index_t idx=0; idx<p.vlen; idx++)
        p[idx]=CMath::exp(p[idx]);

    return p;
}

#endif /* HAVE_EIGEN3 */