NearestNeighborModel.h

Go to the documentation of this file.

//===========================================================================
/*!
 * 
 *
 * \brief      NEarest neighbor model for classification and regression
 * 
 * 
 *
 * \author      T. Glasmachers, C. Igel, O.Krause
 * \date        2012-2017
 *
 *
 * \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_NEARESTNEIGHBOR_H
#define SHARK_MODELS_NEARESTNEIGHBOR_H


#include <shark/Models/AbstractModel.h>
#include <shark/Models/Classifier.h>
#include <shark/Algorithms/NearestNeighbors/AbstractNearestNeighbors.h>

namespace shark {
    
namespace detail{
template <class InputType, class LabelType>
class BaseNearestNeighbor : public AbstractModel<InputType, RealVector>
{
public:
    typedef AbstractNearestNeighbors<InputType,LabelType> NearestNeighbors;
    typedef AbstractModel<InputType, RealVector> base_type;
    typedef typename base_type::BatchInputType BatchInputType;
    typedef typename base_type::BatchOutputType BatchOutputType;

    /// \brief Constructor
    ///
    /// \param algorithm the used algorithm for nearest neighbor search
    /// \param neighbors number of neighbors
    BaseNearestNeighbor(NearestNeighbors const* algorithm, std::size_t outputDimensions, unsigned int neighbors = 3)
    : m_algorithm(algorithm)
    , m_outputDimensions(outputDimensions)
    , m_neighbors(neighbors)
    , m_uniform(true)
    { }

    /// \brief From INameable: return the class name.
    std::string name() const
    { return "Internal"; }
    
    Shape inputShape() const{
        return m_algorithm->inputShape();
    }
    Shape outputShape() const{
        return Shape(m_outputDimensions);
    }

    /// return the number of neighbors
    unsigned int neighbors() const{
        return m_neighbors;
    }

    /// set the number of neighbors
    void setNeighbors(unsigned int neighbors){
        m_neighbors=neighbors;
    }
    
    bool uniformWeights() const{
        return m_uniform;
    }
    bool& uniformWeights(){
        return m_uniform;
    }

    /// get internal parameters of the model
    virtual RealVector parameterVector() const{
        RealVector parameters(1);
        parameters(0) = m_neighbors;
        return parameters;
    }

    /// set internal parameters of the model
    virtual void setParameterVector(RealVector const& newParameters){
        SHARK_RUNTIME_CHECK(newParameters.size() == 1,"Invalid number of parameters");
        m_neighbors = (unsigned int)newParameters(0);
    }

    /// return the size of the parameter vector
    virtual std::size_t numberOfParameters() const{
        return 1;
    }
    
    boost::shared_ptr<State> createState()const{
        return boost::shared_ptr<State>(new EmptyState());
    }

    /// soft k-nearest-neighbor prediction
    void eval(BatchInputType const& patterns, BatchOutputType& outputs) const {
        std::size_t numPatterns = batchSize(patterns);
        std::vector<typename NearestNeighbors::DistancePair> neighbors = m_algorithm->getNeighbors(patterns, m_neighbors);

        outputs.resize(numPatterns, m_outputDimensions);
        outputs.clear();

        for(std::size_t p = 0; p != numPatterns;++p)
        {
            double wsum = 0.0;
            for ( std::size_t k = 0; k != m_neighbors; ++k)
            {
                double w = 1.0;
                if (!m_uniform){
                    double d = neighbors[p*m_neighbors+k].key;
                    if (d < 1e-100) w = 1e100;
                    else w = 1.0 / d;
                }
                updatePrediction(outputs, p, w, neighbors[p*m_neighbors+k].value);
                wsum += w;
            }
            row(outputs, p) /= wsum;
        }
    }
    
    void eval(BatchInputType const& patterns, BatchOutputType& outputs, State&) const {
        eval(patterns,outputs);
    }
    using base_type::eval;

    /// from ISerializable, reads a model from an archive
    void read(InArchive& archive){
        archive & m_neighbors;
        archive & m_outputDimensions;
        archive & m_uniform;
    }

    /// from ISerializable, writes a model to an archive
    void write(OutArchive& archive) const{
        archive & m_neighbors;
        archive & m_outputDimensions;
        archive & m_uniform;
    }
    
private:
    void updatePrediction(RealMatrix& outputs, std::size_t p, double w, unsigned int const label) const{
        outputs(p, label) += w;
    }
    template<class T>
    void updatePrediction(RealMatrix& outputs, std::size_t p, double w, blas::vector<T> const& label)const{
        noalias(row(outputs,p)) += w * label;
    }
    NearestNeighbors const* m_algorithm;

    /// number of classes
    std::size_t m_outputDimensions;

    /// number of neighbors to be taken into account
    unsigned int m_neighbors;

    /// type of distance-based weights computation
    bool m_uniform;
};
}

/// \brief NearestNeighbor model for classification and regression
///
/// The classification, the model predicts a class label
/// according to a local majority decision among its k
/// nearest neighbors. It is not specified how ties are
/// broken.
///
/// For Regression, the (weighted) mean of the k nearest
/// neighbours is computed.
template <class InputType, class LabelType>
class NearestNeighborModel: public detail::BaseNearestNeighbor<InputType,LabelType>
{
public:
    typedef AbstractNearestNeighbors<InputType,LabelType> NearestNeighbors;
    typedef detail::BaseNearestNeighbor<InputType,LabelType> base_type;

    /// \brief Type of distance-based weights.
    enum DistanceWeights{
        UNIFORM,                ///< uniform (= no) distance-based weights
        ONE_OVER_DISTANCE,      ///< weight each neighbor's label with 1/distance
    };

    /// \brief Constructor
    ///
    /// \param algorithm the used algorithm for nearest neighbor search
    /// \param neighbors number of neighbors
    NearestNeighborModel(NearestNeighbors const* algorithm, unsigned int neighbors = 3)
    : base_type(algorithm, labelDimension(algorithm->dataset()), neighbors)
    { }

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

    /// query the way distances enter as weights
    DistanceWeights getDistanceWeightType() const{
        return this->decisionFunction().uniformWeights() ? UNIFORM : ONE_OVER_DISTANCE;
    }

    /// set the way distances enter as weights
    void setDistanceWeightType(DistanceWeights dw){
        this->decisionFunction().uniformWeights() = (dw == UNIFORM);
    }
};


template <class InputType>
class NearestNeighborModel<InputType, unsigned int>: public Classifier<detail::BaseNearestNeighbor<InputType,unsigned int> >
{
public:
    typedef AbstractNearestNeighbors<InputType,unsigned int> NearestNeighbors;
    typedef Classifier<detail::BaseNearestNeighbor<InputType,unsigned int> > base_type;

    /// \brief Type of distance-based weights.
    enum DistanceWeights{
        UNIFORM,                ///< uniform (= no) distance-based weights
        ONE_OVER_DISTANCE,      ///< weight each neighbor's label with 1/distance
    };

    /// \brief Constructor
    ///
    /// \param algorithm the used algorithm for nearest neighbor search
    /// \param neighbors number of neighbors
    NearestNeighborModel(NearestNeighbors const* algorithm, unsigned int neighbors = 3)
    : base_type(detail::BaseNearestNeighbor<InputType,unsigned int>(algorithm, numberOfClasses(algorithm->dataset()), neighbors))
    { }

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

    /// return the number of neighbors
    unsigned int neighbors() const{
        return this->decisionFunction().neighbors();
    }

    /// set the number of neighbors
    void setNeighbors(unsigned int neighbors){
        this->decisionFunction().setNeighbors(neighbors);
    }

    /// query the way distances enter as weights
    DistanceWeights getDistanceWeightType() const{
        return this->decisionFunction().uniformWeights() ? UNIFORM : ONE_OVER_DISTANCE;
    }

    /// set the way distances enter as weights
    void setDistanceWeightType(DistanceWeights dw){
        this->decisionFunction().uniformWeights() = (dw == UNIFORM);
    }
};


}
#endif