SquaredLoss.h

Go to the documentation of this file.

/*!
 * 
 *
 * \brief       Implements the Squared Error Loss function for regression.
 * 
 * 
 * 
 *
 * \author      Oswin Krause, Christian Igel
 * \date        2011
 *
 *
 * \par Copyright 1995-2017 Shark Development Team
 * 
 * <BR><HR>
 * This file is part of Shark.
 * <https://shark-ml.github.io/Shark/>
 * 
 * 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_OBJECTIVEFUNCTIONS_LOSS_SQUAREDLOSS_H
#define SHARK_OBJECTIVEFUNCTIONS_LOSS_SQUAREDLOSS_H
 
 
#include <shark/ObjectiveFunctions/Loss/AbstractLoss.h>
 
namespace shark{
/// \brief squared loss for regression and classification
///
/// The SquaredLoss computes the squared distance
/// between target and prediction. It is defined for both
/// vectorial as well as integral labels. In the case of integral labels,
/// the label c is interpreted as unit-vector having the c-th component activated.
/// \ingroup lossfunctions
template<class OutputType = RealVector, class LabelType = OutputType >
class SquaredLoss : public AbstractLoss<LabelType,OutputType>
{
public:
    typedef AbstractLoss<LabelType,OutputType> base_type;
    typedef typename base_type::BatchOutputType BatchOutputType;
    typedef typename base_type::BatchLabelType BatchLabelType;
 
    /// Constructor.
    SquaredLoss()
    {
        this->m_features|=base_type::HAS_FIRST_DERIVATIVE;
    }
 
 
    /// \brief From INameable: return the class name.
    std::string name() const
    { return "SquaredLoss"; }
 
    using base_type::eval;
 
    /// Evaluate the squared loss \f$ (label - prediction)^2 \f$.
    double eval(BatchLabelType const& labels, BatchOutputType const& predictions) const {
        SIZE_CHECK(labels.size1()==predictions.size1());
        SIZE_CHECK(labels.size2()==predictions.size2());
 
        double error = sum(sqr(labels - predictions));
        return 0.5 * error;
    }
 
    /// Evaluate the squared loss \f$ (label - prediction)^2 \f$
    /// and its deriative \f$ \frac{\partial}{\partial prediction} 1/2 (label - prediction)^2 = prediction - label \f$.
    double evalDerivative(BatchLabelType const& label, BatchOutputType const& prediction, BatchOutputType& gradient) const {
        gradient.resize(prediction.size1(),prediction.size2());
        noalias(gradient) = (prediction - label);
        return SquaredLoss::eval(label,prediction);
    }
};
 
//specialisation for classification case.
template<class OutputType>
class SquaredLoss<OutputType,unsigned int> : public AbstractLoss<unsigned int,OutputType>
{
public:
    typedef AbstractLoss<unsigned int,OutputType> base_type;
    typedef typename base_type::BatchOutputType BatchOutputType;
    typedef typename base_type::BatchLabelType BatchLabelType;
 
    /// Constructor.
    SquaredLoss()
    {
        this->m_features|=base_type::HAS_FIRST_DERIVATIVE;
    }
 
 
    /// \brief From INameable: return the class name.
    std::string name() const
    { return "SquaredLoss"; }
 
    using base_type::eval;
 
    /// Evaluate the squared loss \f$ (label - prediction)^2 \f$.
    double eval(BatchLabelType const& labels, BatchOutputType const& predictions) const {
        SIZE_CHECK(labels.size()==predictions.size1());
 
        double error = 0;
        for(std::size_t i = 0; i != labels.size(); ++i){
            unsigned int c = labels(i);
            SIZE_CHECK(c < predictions.size2());
            error+=norm_sqr(row(predictions,i))+1.0-2.0*predictions(i,c);
        }
        return 0.5 * error;
    }
 
    /// Evaluate the squared loss \f$ (label - prediction)^2 \f$
    /// and its deriative \f$ \frac{\partial}{\partial prediction} 1/2 (label - prediction)^2 = prediction - label \f$.
    double evalDerivative(BatchLabelType const& labels, BatchOutputType const& predictions, BatchOutputType& gradient) const {
        gradient.resize(predictions.size1(),predictions.size2());
        noalias(gradient) = predictions;
        for(std::size_t i = 0; i != labels.size(); ++i){
            unsigned int c = labels(i);
            SIZE_CHECK(c < predictions.size2());
            gradient(i,c)-=1.0;
        }
        return SquaredLoss::eval(labels,predictions);
    }
};
 
//spcialisation for sequence data
template<>
class SquaredLoss<Sequence,Sequence> : public AbstractLoss<Sequence,Sequence>
{
public:
    /// \brief Constructor.
    ///
    /// \param ignore Specifies how many elements of the sequence are to be ignored during evaluation
    ///              must be strictly smaller than the smalles sequnce to evaluate.
    SquaredLoss(std::size_t ignore=0)
    :m_ignore(ignore){
        this->m_features|=base_type::HAS_FIRST_DERIVATIVE;
    }
 
 
    /// \brief From INameable: return the class name.
    std::string name() const
    { return "SquaredLoss"; }
 
    using base_type::eval;
 
    /// \brief Evaluate the squared loss \f$ (label - prediction)^2 \f$.
    ///
    /// For Sequences this is:
    /// \f[ sum_{i=i_0} (label_i-prediction_i)^2\f]
    /// where \f$ i_0 \f$ is the first element to be evaluated. By default it is 0
    double eval(BatchLabelType const& labels, BatchOutputType const& predictions) const {
        SIZE_CHECK(labels.size()==predictions.size());
 
        double error = 0;
        for(std::size_t i = 0; i != labels.size(); ++i){
            SIZE_CHECK(labels[i].size()==predictions[i].size());
            SHARK_RUNTIME_CHECK(labels[i].size()  > m_ignore,"Number of sequence elements to ignore is too large");
 
            for(std::size_t j = m_ignore; j != labels[i].size(); ++j){
                error += distanceSqr(predictions[i][j],labels[i][j]);
            }
        }
        return 0.5 * error;
    }
 
    /// Evaluate the squared loss \f$ (label - prediction)^2 \f$
    /// and its deriative \f$ \frac{\partial}{\partial prediction} 1/2 (label - prediction)^2 = prediction - label \f$.
    double evalDerivative(BatchLabelType const& labels, BatchOutputType const& predictions, BatchOutputType& gradient) const {
        SIZE_CHECK(labels.size()==predictions.size());
        gradient.resize(labels.size());
        
        double error = 0;
        for(std::size_t i = 0; i != labels.size(); ++i){
            SIZE_CHECK(labels[i].size()==predictions[i].size());
            SHARK_RUNTIME_CHECK(labels[i].size()  > m_ignore,"Number of sequence elements to ignore is too large");
            for(std::size_t j = 0; j != m_ignore; ++j){
                gradient[i].push_back(RealVector(predictions[i][j].size(),0.0));
            }
            for(std::size_t j = m_ignore; j != labels[i].size(); ++j){
                error += 0.5 * distanceSqr(predictions[i][j],labels[i][j]);
                gradient[i].push_back(predictions[i][j] - labels[i][j]);
                
            }
        }
        return error;
    }
private:
    std::size_t m_ignore;
};
 
}
#endif