GaussianProcessRegression.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) 2013 Roman Votyakov
 * Copyright (C) 2012 Jacob Walker
 * Copyright (C) 2013 Roman Votyakov
 *
 * Code adapted from Gaussian Process Machine Learning Toolbox
 * http://www.gaussianprocess.org/gpml/code/matlab/doc/
 */

#include <shogun/regression/GaussianProcessRegression.h>

#ifdef HAVE_EIGEN3

#include <shogun/io/SGIO.h>
#include <shogun/machine/gp/FITCInferenceMethod.h>

using namespace shogun;

CGaussianProcessRegression::CGaussianProcessRegression()
        : CGaussianProcessMachine()
{
}

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

CGaussianProcessRegression::~CGaussianProcessRegression()
{
}

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

    CRegressionLabels* result;

    // if regression data equals to NULL, then apply regression on training
    // features
    if (!data)
    {
        CFeatures* feat;

        // use inducing features for FITC inference method
        if (m_method->get_inference_type()==INF_FITC_REGRESSION)
        {
            CFITCInferenceMethod* fitc_method=
                CFITCInferenceMethod::obtain_from_generic(m_method);
            feat=fitc_method->get_inducing_features();
            SG_UNREF(fitc_method);
        }
        else
            feat=m_method->get_features();

        result=new CRegressionLabels(get_mean_vector(feat));

        SG_UNREF(feat);
    }
    else
    {
        result=new CRegressionLabels(get_mean_vector(data));
    }

    return result;
}

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

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

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

    return mu;
}

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

#endif