FirstOrderStochasticMinimizer.h

Go to the documentation of this file.

/*
 * Copyright (c) The Shogun Machine Learning Toolbox
 * Written (w) 2015 Wu Lin
 * 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 FIRSTORDERSTOCHASTICMINIMIZER_H
#define FIRSTORDERSTOCHASTICMINIMIZER_H
#include <shogun/optimization/FirstOrderMinimizer.h>
#include <shogun/optimization/FirstOrderStochasticCostFunction.h>
#include <shogun/optimization/DescendUpdater.h>
#include <shogun/optimization/LearningRate.h>
#include <shogun/optimization/SparsePenalty.h>
#include <shogun/optimization/ProximalPenalty.h>
namespace shogun
{

class FirstOrderStochasticMinimizer: public FirstOrderMinimizer
{
public:
    FirstOrderStochasticMinimizer()
        :FirstOrderMinimizer()
    {
        init();
    }

    FirstOrderStochasticMinimizer(FirstOrderStochasticCostFunction *fun)
        :FirstOrderMinimizer(fun)
    {
        init();
    }

    virtual ~FirstOrderStochasticMinimizer(){}

    virtual bool supports_batch_update() const {return false;}

    virtual void set_gradient_updater(DescendUpdater* gradient_updater)
    {
        REQUIRE(gradient_updater, "Gradient updater must set\n");
        m_gradient_updater=gradient_updater;
    }

    virtual float64_t minimize()=0;

    virtual void set_number_passes(int32_t num_passes)
    {
        REQUIRE(num_passes>0, "The number (%d) to go through data must be positive\n", num_passes);
        m_num_passes=num_passes;
    }

    virtual void load_from_context(CMinimizerContext* context)
    {
        REQUIRE(context,"Context must set\n");
        REQUIRE(m_gradient_updater,"Descend updater must set\n");
        FirstOrderMinimizer::load_from_context(context);
        m_gradient_updater->load_from_context(context);
        if(m_learning_rate)
            m_learning_rate->load_from_context(context);
        std::string key="FirstOrderStochasticMinimizer::m_iter_counter";
        m_iter_counter=context->get_data_int32(key);
    }

    virtual void set_learning_rate(LearningRate *learning_rate)
    {
        m_learning_rate=learning_rate;
    }

    virtual int32_t get_iteration_counter() {return m_iter_counter;}
protected:
    virtual void do_proximal_operation(SGVector<float64_t>variable_reference)
    {
        ProximalPenalty* proximal_penalty=dynamic_cast<ProximalPenalty*>(m_penalty_type);
        if(proximal_penalty)
        {
            float64_t proximal_weight=m_penalty_weight;
            SparsePenalty* sparse_penalty=dynamic_cast<SparsePenalty*>(m_penalty_type);
            if(sparse_penalty)
            {
                REQUIRE(m_learning_rate, "Learning rate must set when Sparse Penalty (eg, L1) is used\n");
                proximal_weight*=m_learning_rate->get_learning_rate(m_iter_counter);
            }
            proximal_penalty->update_variable_for_proximity(variable_reference,proximal_weight);
        }
    }

    virtual void update_context(CMinimizerContext* context)
    {
        REQUIRE(context,"Context must set\n");
        REQUIRE(m_gradient_updater,"Descend updater must set\n");
        FirstOrderMinimizer::update_context(context);
        m_gradient_updater->update_context(context);
        if(m_learning_rate)
            m_learning_rate->update_context(context);
        std::string key="FirstOrderStochasticMinimizer::m_iter_counter";
        context->save_data(key, m_iter_counter);
    }

    DescendUpdater* m_gradient_updater;

    virtual void init_minimization()
    {
        REQUIRE(m_fun,"Cost function must set\n");
        REQUIRE(m_gradient_updater,"Descend updater must set\n");
        REQUIRE(m_num_passes>0, "The number to go through data must set\n");
        m_cur_passes=0;
    }

    int32_t m_num_passes;

    int32_t m_cur_passes;

    int32_t m_iter_counter;

    LearningRate* m_learning_rate;
    
private:
    void init()
    {
        m_gradient_updater=NULL;
        m_learning_rate=NULL;
        m_num_passes=0;
        m_cur_passes=0;
        m_iter_counter=0;
    }
};

}
#endif /* FIRSTORDERSTOCHASTICMINIMIZER_H */