class HMM

This class implements a Hidden Markov Model distribution.

Inheritance:

Public Fields

int n_states

The number of states of the HMM.

real prior_transitions

a prior on the transition probabilities

Distribution** states

keep the emission distributions

Distribution** shared_states

if the states are in fact shared in some way or another, the original ones are in shared_states

real** transitions

the initial transitions between states are kept as a matrix

real** log_transitions

in fact, we keep the transitions in log

real** dlog_transitions

the derivative of the log transitions for gradient descent

real** transitions_acc

the accumulators of the transitions for EM

Sequence* log_alpha

accumulator used in the forward phase to compute log likelihood

Sequence* log_beta

accumulator used in the backward phase to compute log likelihood

Sequence* arg_viterbi

for each state, for each time step, keep the best predecessor

int last_arg_viterbi

arg_viterbi of the finishing state

Sequence* viterbi_sequence

for each time step, keep the best state

Sequence* log_probabilities_s

keep for each time step and each model its emission log probability

bool initialize

do we need to initialize the model?

Public Methods

virtual void printTransitions(bool real_values=false, bool transitions_only=false)

this method can be used for debugging purpose to see the transitions

virtual void logAlpha(Sequence* inputs)

computes the log_alpha during forward phase of EM

virtual void logBeta(Sequence* inputs)

computes the log_beta during backward phase of EM

virtual void logViterbi(Sequence* inputs)

computes the log_viterbi during forward phase of Viterbi

virtual void decode(Sequence* input)

this method returns the state sequence associated to the input

virtual void logProbabilities(Sequence* inputs)

computes for each state and each time step of the sequence inputs its associated emission probability

Inherited from Distribution:

Public Fields

real log_probability

Sequence* log_probabilities

Public Methods

virtual real logProbability(Sequence* inputs)

virtual real viterbiLogProbability(Sequence* inputs)

virtual real frameLogProbability(int t, real* f_inputs)

virtual real viterbiFrameLogProbability(int t, real* f_inputs)

virtual void eMIterInitialize()

virtual void iterInitialize()

virtual void eMSequenceInitialize(Sequence* inputs)

virtual void sequenceInitialize(Sequence* inputs)

virtual void eMAccPosteriors(Sequence* inputs, real log_posterior)

virtual void frameEMAccPosteriors(int t, real* f_inputs, real log_posterior)

virtual void viterbiAccPosteriors(Sequence* inputs, real log_posterior)

virtual void frameViterbiAccPosteriors(int t, real* f_inputs, real log_posterior)

virtual void eMUpdate()

virtual void update()

virtual void eMForward(Sequence* inputs)

virtual void viterbiForward(Sequence* inputs)

virtual void frameBackward(int t, real* f_inputs, real* beta_, real* f_outputs, real* alpha_)

virtual void viterbiBackward(Sequence* inputs, Sequence* alpha)

virtual void frameDecision(int t, real* decision)

Public Members

Returns the decision of the distribution

Inherited from GradientMachine:

Public Fields

int n_inputs

int n_outputs

Parameters* params

Parameters* der_params

Sequence* beta

Public Methods

virtual void forward(Sequence* inputs)

virtual void backward(Sequence* inputs, Sequence* alpha)

virtual void setPartialBackprop(bool flag=true)

virtual void frameForward(int t, real* f_inputs, real* f_outputs)

virtual void loadXFile(XFile* file)

virtual void saveXFile(XFile* file)

Inherited from Machine:

Public Fields

Sequence* outputs

Public Methods

virtual void reset()

virtual void setDataSet(DataSet* dataset_)

Inherited from Object:

Public Fields

Allocator* allocator

Public Methods

void addOption(const char* name, int size, void* ptr, const char* help="")

void addIOption(const char* name, int* ptr, int init_value, const char* help="")

void addROption(const char* name, real* ptr, real init_value, const char* help="")

void addBOption(const char* name, bool* ptr, bool init_value, const char* help="")

void addOOption(const char* name, Object** ptr, Object* init_value, const char* help="")

void setOption(const char* name, void* ptr)

void setIOption(const char* name, int option)

void setROption(const char* name, real option)

void setBOption(const char* name, bool option)

void setOOption(const char* name, Object* option)

void load(const char* filename)

void save(const char* filename)

void* operator new(size_t size, Allocator* allocator_=NULL)

void* operator new(size_t size, Allocator* allocator_, void* ptr_)

void operator delete(void* ptr)

Documentation

This class implements a Hidden Markov Model distribution. It can be trained either by EM, Viterbi, or Gradient Descent.

Note that this kind of HMM always contain one initial state and one final state. Both are non-emitting.

Note that the log_probabilities is the average over all frames of the log_probability of the example.

int n_states

The number of states of the HMM. the first model is the initial state, the last model is the final (absorbing) state, (neither of them are emitting). hence, n_states > 2

real prior_transitions

a prior on the transition probabilities

Distribution** states

keep the emission distributions

Distribution** shared_states

if the states are in fact shared in some way or another, the original ones are in shared_states

real** transitions

the initial transitions between states are kept as a matrix

real** log_transitions

in fact, we keep the transitions in log

real** dlog_transitions

the derivative of the log transitions for gradient descent

real** transitions_acc

the accumulators of the transitions for EM

Sequence* log_alpha

accumulator used in the forward phase to compute log likelihood

Sequence* log_beta

accumulator used in the backward phase to compute log likelihood

Sequence* arg_viterbi

for each state, for each time step, keep the best predecessor

int last_arg_viterbi

arg_viterbi of the finishing state

Sequence* viterbi_sequence

for each time step, keep the best state

Sequence* log_probabilities_s

keep for each time step and each model its emission log probability

bool initialize

do we need to initialize the model?

virtual void printTransitions(bool real_values=false, bool transitions_only=false)

this method can be used for debugging purpose to see the transitions

virtual void logAlpha(Sequence* inputs)

computes the log_alpha during forward phase of EM

virtual void logBeta(Sequence* inputs)

computes the log_beta during backward phase of EM

virtual void logViterbi(Sequence* inputs)

computes the log_viterbi during forward phase of Viterbi

virtual void decode(Sequence* input)

this method returns the state sequence associated to the input

virtual void logProbabilities(Sequence* inputs)

computes for each state and each time step of the sequence inputs its associated emission probability

Direct child classes:

SpeechHMM
MAPHMM

Author:

Samy Bengio (bengio@idiap.ch)

Alphabetic index HTML hierarchy of classes or Java