Inference.h

Go to the documentation of this file.

/*
 * Copyright (c) The Shogun Machine Learning Toolbox
 * Written (W) 2015 Wu Lin
 * Written (W) 2013-2014 Heiko Strathmann
 * Written (W) 2013 Roman Votyakov
 * Written (W) 2012 Jacob Walker
 * 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.
 *
 */
#ifndef CINFERENCE_H_
#define CINFERENCE_H_

#include <shogun/lib/config.h>


#include <shogun/base/SGObject.h>
#include <shogun/kernel/Kernel.h>
#include <shogun/features/Features.h>
#include <shogun/labels/Labels.h>
#include <shogun/machine/gp/LikelihoodModel.h>
#include <shogun/machine/gp/MeanFunction.h>
#include <shogun/evaluation/DifferentiableFunction.h>
#include <shogun/optimization/Minimizer.h>

namespace shogun
{

enum EInferenceType
{
    INF_NONE=0,
    INF_EXACT=10,
    INF_SPARSE=20,
    INF_FITC_REGRESSION=21,
    INF_FITC_LAPLACE_SINGLE=22,
    INF_LAPLACE=30,
    INF_LAPLACE_SINGLE=31,
    INF_LAPLACE_MULTIPLE=32,
    INF_EP=40,
    INF_KL=50,
    INF_KL_DIAGONAL=51,
    INF_KL_CHOLESKY=52,
    INF_KL_COVARIANCE=53,
    INF_KL_DUAL=54,
    INF_KL_SPARSE_REGRESSION=55
};

class CInference : public CDifferentiableFunction
{
public:
    CInference();

    CInference(CKernel* kernel, CFeatures* features,
            CMeanFunction* mean, CLabels* labels, CLikelihoodModel* model);

    virtual ~CInference();

    virtual EInferenceType get_inference_type() const { return INF_NONE; }

    virtual float64_t get_negative_log_marginal_likelihood()=0;

    float64_t get_marginal_likelihood_estimate(int32_t num_importance_samples=1,
            float64_t ridge_size=1e-15);

    virtual CMap<TParameter*, SGVector<float64_t> >*
    get_negative_log_marginal_likelihood_derivatives(CMap<TParameter*,
            CSGObject*>* parameters);

    virtual SGVector<float64_t> get_alpha()=0;

    virtual SGMatrix<float64_t> get_cholesky()=0;

    virtual SGVector<float64_t> get_diagonal_vector()=0;

    virtual SGVector<float64_t> get_posterior_mean()=0;

    virtual SGMatrix<float64_t> get_posterior_covariance()=0;

    virtual CMap<TParameter*, SGVector<float64_t> >* get_gradient(
            CMap<TParameter*, CSGObject*>* parameters)
    {
        return get_negative_log_marginal_likelihood_derivatives(parameters);
    }

    virtual SGVector<float64_t> get_value()
    {
        SGVector<float64_t> result(1);
        result[0]=get_negative_log_marginal_likelihood();
        return result;
    }

    virtual CFeatures* get_features() { SG_REF(m_features); return m_features; }

    virtual void set_features(CFeatures* feat)
    {
        SG_REF(feat);
        SG_UNREF(m_features);
        m_features=feat;
    }

    virtual CKernel* get_kernel() { SG_REF(m_kernel); return m_kernel; }

    virtual void set_kernel(CKernel* kern)
    {
        SG_REF(kern);
        SG_UNREF(m_kernel);
        m_kernel=kern;
    }

    virtual CMeanFunction* get_mean() { SG_REF(m_mean); return m_mean; }

    virtual void set_mean(CMeanFunction* m)
    {
        SG_REF(m);
        SG_UNREF(m_mean);
        m_mean=m;
    }

    virtual CLabels* get_labels() { SG_REF(m_labels); return m_labels; }

    virtual void set_labels(CLabels* lab)
    {
        SG_REF(lab);
        SG_UNREF(m_labels);
        m_labels=lab;
    }

    CLikelihoodModel* get_model() { SG_REF(m_model); return m_model; }

    virtual void set_model(CLikelihoodModel* mod)
    {
        SG_REF(mod);
        SG_UNREF(m_model);
        m_model=mod;
    }

    virtual float64_t get_scale() const;

    virtual void set_scale(float64_t scale);

    virtual bool supports_regression() const { return false; }

    virtual bool supports_binary() const { return false; }

    virtual bool supports_multiclass() const { return false; }

    virtual void update();

    virtual SGMatrix<float64_t> get_multiclass_E();

    virtual void register_minimizer(Minimizer* minimizer);
protected:
    virtual void check_members() const;

    virtual void update_alpha()=0;

    virtual void update_chol()=0;

    virtual void update_deriv()=0;

    virtual void update_train_kernel();

    virtual SGVector<float64_t> get_derivative_wrt_inference_method(
            const TParameter* param)=0;

    virtual SGVector<float64_t> get_derivative_wrt_likelihood_model(
            const TParameter* param)=0;

    virtual SGVector<float64_t> get_derivative_wrt_kernel(
            const TParameter* param)=0;

    virtual SGVector<float64_t> get_derivative_wrt_mean(
            const TParameter* param)=0;

    static void* get_derivative_helper(void* p);

    virtual void compute_gradient();

private:
    void init();

protected:
    Minimizer* m_minimizer;

    CKernel* m_kernel;

    CMeanFunction* m_mean;

    CLikelihoodModel* m_model;

    CFeatures* m_features;

    CLabels* m_labels;

    SGVector<float64_t> m_alpha;

    SGMatrix<float64_t> m_L;

    float64_t m_log_scale;

    SGMatrix<float64_t> m_ktrtr;

    SGMatrix<float64_t> m_E;

    bool m_gradient_update;
};
}
#endif /* CINFERENCE_H_ */