DiscreteHiddenMarkovModel.h

Go to the documentation of this file.

#ifndef GRT_DISCRETE_HIDDEN_MARKOV_MODEL_HEADER
#define GRT_DISCRETE_HIDDEN_MARKOV_MODEL_HEADER

#include "HMMEnums.h"
#include "../../Util/GRTCommon.h"
#include "../../CoreModules/MLBase.h"

GRT_BEGIN_NAMESPACE

//This class is used for the HMM batch training
class GRT_API HMMTrainingObject{
    public:
    HMMTrainingObject(){
        pk = 0.0;
    }
    ~HMMTrainingObject(){}
    MatrixFloat alpha;     //The forward estimate matrix
    MatrixFloat beta;      //The backward estimate matrix
    VectorFloat c;         //The scaling coefficient Vector
    Float pk;               //P( O | Model )
};

class GRT_API DiscreteHiddenMarkovModel : public MLBase {
    
public:
    DiscreteHiddenMarkovModel();
    
    DiscreteHiddenMarkovModel(const UINT numStates,const UINT numSymbols,const UINT modelType,const UINT delta);
    
    DiscreteHiddenMarkovModel(const MatrixFloat &a,const MatrixFloat &b,const VectorFloat &pi,const UINT modelType,const UINT delta);
    
    DiscreteHiddenMarkovModel(const DiscreteHiddenMarkovModel &rhs);
    
    virtual ~DiscreteHiddenMarkovModel();
    
    Float predict(const UINT newSample);
    Float predict(const Vector<UINT> &obs);
    
    bool resetModel(const UINT numStates,const UINT numSymbols,const UINT modelType,const UINT delta);
    bool train(const Vector< Vector<UINT> > &trainingData);
    
    virtual bool reset();
    
    virtual bool save( std::fstream &file ) const;
    
    virtual bool load( std::fstream &file );
    
    bool randomizeMatrices(const UINT numStates,const UINT numSymbols);
    Float predictLogLikelihood(const Vector<UINT> &obs);
    bool forwardBackward(HMMTrainingObject &trainingObject,const Vector<UINT> &obs);
    bool train_(const Vector< Vector<UINT> > &obs,const UINT maxIter, UINT &currentIter,Float &newLoglikelihood);
    virtual bool print() const;
    
    VectorFloat getTrainingIterationLog() const;
    
    using MLBase::save;
    using MLBase::load;
    using MLBase::train;
    using MLBase::train_;
    using MLBase::predict_;
    using MLBase::predict;
    
protected:
    UINT numStates;             //The number of states for this model
    UINT numSymbols;            //The number of symbols for this model
    MatrixFloat a;             //The transitions probability matrix
    MatrixFloat b;             //The emissions probability matrix
    VectorFloat pi;            //The state start probability Vector
    VectorFloat trainingIterationLog;   //Stores the loglikelihood at each iteration the BaumWelch algorithm
    
    UINT modelType;
    UINT delta;             //The number of states a model can move to in a LeftRight model
    UINT numRandomTrainingIterations;       //The number of training loops to find the best starting values
    Float logLikelihood;    //The log likelihood of an observation sequence given the modal, calculated by the forward method
    Float cThreshold;       //The classification threshold for this model
    CircularBuffer<UINT> observationSequence;
    Vector< UINT > estimatedStates;
};

GRT_END_NAMESPACE

#endif //GRT_HIDDEN_MARKOV_MODEL_HEADER