xmmHmmSingleClass.hpp

Go to the documentation of this file.

/*
 * xmmHmmSingleClass.hpp
 *
 * Hidden Markov Model Definition for a Single Class
 *
 * Contact:
 * - Jules Francoise <jules.francoise@ircam.fr>
 *
 * This code has been initially authored by Jules Francoise
 * <http://julesfrancoise.com> during his PhD thesis, supervised by Frederic
 * Bevilacqua <href="http://frederic-bevilacqua.net>, in the Sound Music
 * Movement Interaction team <http://ismm.ircam.fr> of the
 * STMS Lab - IRCAM, CNRS, UPMC (2011-2015).
 *
 * Copyright (C) 2015 UPMC, Ircam-Centre Pompidou.
 *
 * This File is part of XMM.
 *
 * XMM is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * XMM 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 General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with XMM.  If not, see <http://www.gnu.org/licenses/>.
 */

#ifndef xmmHmmSingleClass_hpp
#define xmmHmmSingleClass_hpp

#include "../gmm/xmmGmmSingleClass.hpp"
#include "xmmHmmParameters.hpp"
#include "xmmHmmResults.hpp"

namespace xmm {
class SingleClassHMM : public SingleClassProbabilisticModel {
    template <typename SingleClassModel, typename ModelType>
    friend class Model;
    friend class HierarchicalHMM;

  public:
    static const float DEFAULT_EXITPROBABILITY_LAST_STATE() { return 0.1; }
    static const float TRANSITION_REGULARIZATION() { return 1.0e-5; }

    SingleClassHMM(std::shared_ptr<SharedParameters> p = NULL);

    SingleClassHMM(SingleClassHMM const& src);

    explicit SingleClassHMM(std::shared_ptr<SharedParameters> p,
                            Json::Value const& root);

    SingleClassHMM& operator=(SingleClassHMM const& src);


    void addExitPoint(int stateIndex, float proba);



    void reset();

    double filter(std::vector<float> const& observation);



    virtual Json::Value toJson() const;

    virtual void fromJson(Json::Value const& root);


    //        /**
    //         @brief Convert to bimodal HMM in place
    //         @param dimension_input dimension of the input modality
    //         @throws runtime_error if the model is already bimodal
    //         @throws out_of_range if the requested input dimension is too
    //         large
    //         */
    //        void makeBimodal(unsigned int dimension_input);
    //
    //        /**
    //         @brief Convert to unimodal HMM in place
    //         @throws runtime_error if the model is already unimodal
    //         */
    //        void makeUnimodal();
    //
    //        /**
    //         @brief extract a submodel with the given columns
    //         @param columns columns indices in the target order
    //         @throws runtime_error if the model is training
    //         @throws out_of_range if the number or indices of the requested
    //         columns exceeds the current dimension
    //         @return a HMM from the current model considering only the target
    //         columns
    //         */
    //        HMM extractSubmodel(std::vector<unsigned int>& columns) const;
    //
    //        /**
    //         @brief extract the submodel of the input modality
    //         @throws runtime_error if the model is training or if it is not
    //         bimodal
    //         @return a unimodal HMM of the input modality from the current
    //         bimodal model
    //         */
    //        HMM extractSubmodel_input() const;
    //
    //        /**
    //         @brief extract the submodel of the output modality
    //         @throws runtime_error if the model is training or if it is not
    //         bimodal
    //         @return a unimodal HMM of the output modality from the current
    //         bimodal model
    //         */
    //        HMM extractSubmodel_output() const;
    //
    //        /**
    //         @brief extract the model with reversed input and output
    //         modalities
    //         @throws runtime_error if the model is training or if it is not
    //         bimodal
    //         @return a bimodal HMM  that swaps the input and output modalities
    //         */
    //        HMM extract_inverse_model() const;

    ClassParameters<HMM> parameters;

    ClassResults<HMM> results;

    std::vector<double> alpha;

    std::vector<double> alpha_h[3];

    std::vector<SingleClassGMM> states;

    std::vector<float> prior;

    std::vector<float> transition;

  protected:
    void allocate();

    void initParametersToDefault(std::vector<float> const& dataStddev);

    void initMeansWithAllPhrases(TrainingSet* trainingSet);

    void initCovariances_fullyObserved(TrainingSet* trainingSet);

    void initMeansCovariancesWithGMMEM(TrainingSet* trainingSet);

    void setErgodic();

    void setLeftRight();

    void normalizeTransitions();

    double forward_init(const float* observation,
                        const float* observation_output = NULL);

    double forward_update(const float* observation,
                          const float* observation_output = NULL);

    void backward_init(double ct);

    void backward_update(double ct, const float* observation,
                         const float* observation_output = NULL);

    void emAlgorithmInit(TrainingSet* trainingSet);

    void emAlgorithmTerminate();

    virtual double emAlgorithmUpdate(TrainingSet* trainingSet);

    double baumWelch_forwardBackward(std::shared_ptr<Phrase> currentPhrase,
                                     int phraseIndex);

    double baumWelch_forward_update(
        std::vector<double>::iterator observation_likelihoods);

    void baumWelch_backward_update(
        double ct, std::vector<double>::iterator observation_likelihoods);

    void baumWelch_gammaSum(TrainingSet* trainingSet);

    void baumWelch_estimateMixtureCoefficients(TrainingSet* trainingSet);

    void baumWelch_estimateMeans(TrainingSet* trainingSet);

    void baumWelch_estimateCovariances(TrainingSet* trainingSet);

    void baumWelch_estimatePrior(TrainingSet* trainingSet);

    void baumWelch_estimateTransitions(TrainingSet* trainingSet);

    void addCyclicTransition(double proba);

    void updateAlphaWindow();

    void regression(std::vector<float> const& observation_input);

    void updateResults();

    void updateExitProbabilities(float* _exitProbabilities = NULL);

  protected:
    bool forward_initialized_;

    std::vector<double> previous_alpha_;

    std::vector<double> beta_;

    std::vector<double> previous_beta_;

    std::vector<std::vector<double> > gamma_sequence_;

    std::vector<std::vector<double> > epsilon_sequence_;

    std::vector<std::vector<std::vector<double> > > gamma_sequence_per_mixture_;

    std::vector<double> alpha_seq_;

    std::vector<double> beta_seq_;

    std::vector<double> gamma_sum_;

    std::vector<double> gamma_sum_per_mixture_;

    bool is_hierarchical_;

    std::vector<float> exit_probabilities_;

    int window_minindex_;

    int window_maxindex_;

    double window_normalization_constant_;
};
}

#endif