Softmax.h

Go to the documentation of this file.

//===========================================================================
/*!
 * 
 *
 * \brief       Soft-max transformation.
 * 
 * 
 *
 * \author      O. Krause, T. Glasmachers
 * \date        2010-2011
 *
 *
 * \par Copyright 1995-2017 Shark Development Team
 * 
 * <BR><HR>
 * This file is part of Shark.
 * <http://shark-ml.org/>
 * 
 * Shark is free software: you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published 
 * by the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 * 
 * Shark is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU Lesser General Public License for more details.
 * 
 * You should have received a copy of the GNU Lesser General Public License
 * along with Shark.  If not, see <http://www.gnu.org/licenses/>.
 *
 */
//===========================================================================


#ifndef SHARK_MODELS_SOFTMAX_H
#define SHARK_MODELS_SOFTMAX_H

#include <shark/Core/DLLSupport.h>
#include <shark/Models/AbstractModel.h>
namespace shark {


///
/// \brief Softmax function
///
/// \par
/// Squash an n-dimensional real vector space
/// to the (n-1)-dimensional probability simplex:
/// \f[
///      f_i(x) = \frac{\exp(x_i)}{\sum_j \exp(x_j)}
/// \f]
/// This also corresponds to the exponential norm of the input.
///
/// in the case of n=1, the output is
/// \f[
///      f_i(x) = \frac{\exp((2i-1)x)}{\exp(x_j)+\exp(-x_j)}
/// \f]
/// and the output dimension is 2. 
///
/// This convention ensures that all models that are trained via CrossEntropy
/// can be used as input to this model and the output will be the probability
/// of the labels.
    
class Softmax : public AbstractModel<RealVector,RealVector>
{
private:
    struct InternalState : public State{
        RealMatrix results;

        void resize(std::size_t numPatterns,std::size_t inputs){
            results.resize(numPatterns,inputs);
        }
    };

public:
    /// Constructor
    SHARK_EXPORT_SYMBOL Softmax(size_t inputs);
    /// Constructor
    SHARK_EXPORT_SYMBOL Softmax();

    /// \brief From INameable: return the class name.
    std::string name() const
    { return "Softmax"; }

    RealVector parameterVector()const{
        return RealVector();
    }
    void setParameterVector(RealVector const& newParameters){
        SIZE_CHECK(newParameters.size()==0);
    }

    size_t inputSize()const{
        return m_inputSize;
    }
    size_t outputSize()const{
        return m_inputSize==1?2:m_inputSize;
    }
    size_t numberOfParameters()const{
        return 0;
    }

    boost::shared_ptr<State> createState()const{
        return boost::shared_ptr<State>(new InternalState());
    }

    SHARK_EXPORT_SYMBOL void eval(BatchInputType const& patterns,BatchOutputType& output)const;
    SHARK_EXPORT_SYMBOL void eval(BatchInputType const& patterns,BatchOutputType& output, State & state)const;
    using AbstractModel<RealVector,RealVector>::eval;
    
    SHARK_EXPORT_SYMBOL void weightedParameterDerivative(
        BatchInputType const& patterns, BatchOutputType const& coefficients,  State const& state, RealVector& gradient
    )const;
    SHARK_EXPORT_SYMBOL void weightedInputDerivative(
        BatchInputType const& patterns, RealMatrix const& coefficients,  State const& state, BatchOutputType& gradient
    )const;

    void setStructure(std::size_t inputSize){
        m_inputSize = inputSize;
    }
    
    /// From ISerializable, reads a model from an archive
    SHARK_EXPORT_SYMBOL void read( InArchive & archive );

    /// From ISerializable, writes a model to an archive
    SHARK_EXPORT_SYMBOL void write( OutArchive & archive ) const;
    
private:
    std::size_t m_inputSize;
};

}
#endif