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 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 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 saveModelToFile( std::fstream &file ) const;
 
 virtual bool loadModelFromFile( 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;
 
 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