SingleLaplaceInferenceMethod.h

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/
 */

#ifndef CSINGLELAPLACEINFERENCEMETHOD_H_
#define CSINGLELAPLACEINFERENCEMETHOD_H_

#include <shogun/lib/config.h>
#include <shogun/machine/gp/LaplaceInference.h>
#include <shogun/optimization/Minimizer.h>

namespace shogun
{
class CSingleLaplaceInferenceMethod: public CLaplaceInference
{
friend class CSingleLaplaceNewtonOptimizer; 
friend class SingleLaplaceInferenceMethodCostFunction;
public:
    CSingleLaplaceInferenceMethod();

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

    virtual ~CSingleLaplaceInferenceMethod();

    virtual const char* get_name() const { return "SingleLaplaceInferenceMethod"; }

    virtual EInferenceType get_inference_type() const { return INF_LAPLACE_SINGLE; }

    static CSingleLaplaceInferenceMethod* obtain_from_generic(CInference* inference);

    virtual float64_t get_negative_log_marginal_likelihood();

    virtual bool supports_regression() const
    {
        check_members();
        return m_model->supports_regression();
    }

    virtual bool supports_binary() const
    {
        check_members();
        return m_model->supports_binary();
    }

    virtual SGVector<float64_t> get_diagonal_vector();

    virtual void update();

    virtual SGVector<float64_t> get_posterior_mean();

      
    virtual void register_minimizer(Minimizer* minimizer);
protected:

    virtual void update_init();

    virtual void update_alpha();

    virtual void update_chol();

    virtual void update_approx_cov();

    virtual void update_deriv();

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

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

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

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

    float64_t get_psi_wrt_alpha();

    void get_gradient_wrt_alpha(SGVector<float64_t> gradient);

private:
    void init();

protected:
    SGVector<float64_t> m_mean_f;

    SGVector<float64_t> m_sW;

    SGVector<float64_t> m_d2lp;

    SGVector<float64_t> m_d3lp;

    SGVector<float64_t> m_dfhat;

    SGMatrix<float64_t> m_Z;

    SGVector<float64_t> m_g;

    float64_t m_Psi;
};


class CSingleLaplaceNewtonOptimizer: public Minimizer
{
public:
    CSingleLaplaceNewtonOptimizer() :Minimizer() {  init(); }

    virtual const char* get_name() const { return "SingleLaplaceNewtonOptimizer"; }

    virtual ~CSingleLaplaceNewtonOptimizer() { SG_UNREF(m_obj); }

    void set_target(CSingleLaplaceInferenceMethod *obj);

    void unset_target(bool is_unref);

    virtual float64_t minimize();

    virtual void set_minimization_max(float64_t max) { m_opt_max=max; }

    virtual void set_minimization_tolerance(float64_t tol) { m_opt_tolerance=tol; }

    virtual void set_newton_iterations(int32_t iter) { m_iter=iter; }

    virtual void set_newton_tolerance(float64_t tol) { m_tolerance=tol; }

private:
    void init();

    CSingleLaplaceInferenceMethod *m_obj;

    float64_t m_tolerance;

    index_t m_iter;

    float64_t m_opt_tolerance;

    float64_t m_opt_max;
};

}
#endif /* CSINGLELAPLACEINFERENCEMETHOD_H_ */