GestureRecognitionToolkit  Version: 0.1.0
The Gesture Recognition Toolkit (GRT) is a cross-platform, open-source, c++ machine learning library for real-time gesture recognition.
HMM Class Reference

#include <HMM.h>

Inheritance diagram for HMM:
Classifier MLBase GRTBase Observer< TrainingResult > Observer< TestInstanceResult >

Public Member Functions

 HMM (const UINT hmmType=HMM_CONTINUOUS, const UINT modelType=HMM_LEFTRIGHT, const UINT delta=1, const bool useScaling=false, const bool useNullRejection=false)
 
 HMM (const HMM &rhs)
 
virtual ~HMM (void)
 
HMMoperator= (const HMM &rhs)
 
virtual bool deepCopyFrom (const Classifier *classifier)
 
virtual bool train (ClassificationData trainingData)
 
virtual bool train_ (TimeSeriesClassificationData &trainingData)
 
virtual bool predict_ (VectorFloat &inputVector)
 
virtual bool predict_ (MatrixFloat &timeseries)
 
virtual bool reset ()
 
virtual bool clear ()
 
virtual bool print () const
 
virtual bool saveModelToFile (std::fstream &file) const
 
virtual bool loadModelFromFile (std::fstream &file)
 
UINT getHMMType () const
 
UINT getModelType () const
 
UINT getDelta () const
 
UINT getNumStates () const
 
UINT getNumSymbols () const
 
UINT getNumRandomTrainingIterations () const
 
Vector< DiscreteHiddenMarkovModelgetDiscreteModels () const
 
Vector< ContinuousHiddenMarkovModelgetContinuousModels () const
 
bool setHMMType (const UINT hmmType)
 
bool setModelType (const UINT modelType)
 
bool setDelta (const UINT delta)
 
bool setDownsampleFactor (const UINT downsampleFactor)
 
bool setCommitteeSize (const UINT committeeSize)
 
bool setNumStates (const UINT numStates)
 
bool setNumSymbols (const UINT numStates)
 
bool setNumRandomTrainingIterations (const UINT numRandomTrainingIterations)
 
bool setSigma (const Float sigma)
 
bool setAutoEstimateSigma (const bool autoEstimateSigma)
 
- Public Member Functions inherited from Classifier
 Classifier (void)
 
virtual ~Classifier (void)
 
bool copyBaseVariables (const Classifier *classifier)
 
std::string getClassifierType () const
 
bool getSupportsNullRejection () const
 
bool getNullRejectionEnabled () const
 
Float getNullRejectionCoeff () const
 
Float getMaximumLikelihood () const
 
Float getBestDistance () const
 
Float getPhase () const
 
virtual UINT getNumClasses () const
 
UINT getClassLabelIndexValue (UINT classLabel) const
 
UINT getPredictedClassLabel () const
 
VectorFloat getClassLikelihoods () const
 
VectorFloat getClassDistances () const
 
VectorFloat getNullRejectionThresholds () const
 
Vector< UINT > getClassLabels () const
 
Vector< MinMaxgetRanges () const
 
bool enableNullRejection (bool useNullRejection)
 
virtual bool setNullRejectionCoeff (Float nullRejectionCoeff)
 
virtual bool setNullRejectionThresholds (VectorFloat newRejectionThresholds)
 
virtual bool recomputeNullRejectionThresholds ()
 
bool getTimeseriesCompatible () const
 
ClassifiercreateNewInstance () const
 
ClassifierdeepCopy () const
 
const ClassifiergetClassifierPointer () const
 
const ClassifiergetBaseClassifier () const
 
- Public Member Functions inherited from MLBase
 MLBase (void)
 
virtual ~MLBase (void)
 
bool copyMLBaseVariables (const MLBase *mlBase)
 
virtual bool train_ (ClassificationData &trainingData)
 
virtual bool train (RegressionData trainingData)
 
virtual bool train_ (RegressionData &trainingData)
 
virtual bool train (TimeSeriesClassificationData trainingData)
 
virtual bool train (ClassificationDataStream trainingData)
 
virtual bool train_ (ClassificationDataStream &trainingData)
 
virtual bool train (UnlabelledData trainingData)
 
virtual bool train_ (UnlabelledData &trainingData)
 
virtual bool train (MatrixFloat data)
 
virtual bool train_ (MatrixFloat &data)
 
virtual bool predict (VectorFloat inputVector)
 
virtual bool predict (MatrixFloat inputMatrix)
 
virtual bool map (VectorFloat inputVector)
 
virtual bool map_ (VectorFloat &inputVector)
 
virtual bool save (const std::string filename) const
 
virtual bool load (const std::string filename)
 
virtual bool saveModelToFile (std::string filename) const
 
virtual bool loadModelFromFile (std::string filename)
 
virtual bool getModel (std::ostream &stream) const
 
Float scale (const Float &x, const Float &minSource, const Float &maxSource, const Float &minTarget, const Float &maxTarget, const bool constrain=false)
 
virtual std::string getModelAsString () const
 
DataType getInputType () const
 
DataType getOutputType () const
 
UINT getBaseType () const
 
UINT getNumInputFeatures () const
 
UINT getNumInputDimensions () const
 
UINT getNumOutputDimensions () const
 
UINT getMinNumEpochs () const
 
UINT getMaxNumEpochs () const
 
UINT getValidationSetSize () const
 
UINT getNumTrainingIterationsToConverge () const
 
Float getMinChange () const
 
Float getLearningRate () const
 
Float getRootMeanSquaredTrainingError () const
 
Float getTotalSquaredTrainingError () const
 
Float getValidationSetAccuracy () const
 
VectorFloat getValidationSetPrecision () const
 
VectorFloat getValidationSetRecall () const
 
bool getUseValidationSet () const
 
bool getRandomiseTrainingOrder () const
 
bool getTrained () const
 
bool getModelTrained () const
 
bool getScalingEnabled () const
 
bool getIsBaseTypeClassifier () const
 
bool getIsBaseTypeRegressifier () const
 
bool getIsBaseTypeClusterer () const
 
bool enableScaling (const bool useScaling)
 
bool setMaxNumEpochs (const UINT maxNumEpochs)
 
bool setMinNumEpochs (const UINT minNumEpochs)
 
bool setMinChange (const Float minChange)
 
bool setLearningRate (const Float learningRate)
 
bool setUseValidationSet (const bool useValidationSet)
 
bool setValidationSetSize (const UINT validationSetSize)
 
bool setRandomiseTrainingOrder (const bool randomiseTrainingOrder)
 
bool setTrainingLoggingEnabled (const bool loggingEnabled)
 
bool registerTrainingResultsObserver (Observer< TrainingResult > &observer)
 
bool registerTestResultsObserver (Observer< TestInstanceResult > &observer)
 
bool removeTrainingResultsObserver (const Observer< TrainingResult > &observer)
 
bool removeTestResultsObserver (const Observer< TestInstanceResult > &observer)
 
bool removeAllTrainingObservers ()
 
bool removeAllTestObservers ()
 
bool notifyTrainingResultsObservers (const TrainingResult &data)
 
bool notifyTestResultsObservers (const TestInstanceResult &data)
 
MLBasegetMLBasePointer ()
 
const MLBasegetMLBasePointer () const
 
Vector< TrainingResult > getTrainingResults () const
 
- Public Member Functions inherited from GRTBase
 GRTBase (void)
 
virtual ~GRTBase (void)
 
bool copyGRTBaseVariables (const GRTBase *GRTBase)
 
std::string getClassType () const
 
std::string getLastWarningMessage () const
 
std::string getLastErrorMessage () const
 
std::string getLastInfoMessage () const
 
bool setInfoLoggingEnabled (const bool loggingEnabled)
 
bool setWarningLoggingEnabled (const bool loggingEnabled)
 
bool setErrorLoggingEnabled (const bool loggingEnabled)
 
GRTBasegetGRTBasePointer ()
 
const GRTBasegetGRTBasePointer () const
 
- Public Member Functions inherited from Observer< TrainingResult >
virtual void notify (const TrainingResult &data)
 
- Public Member Functions inherited from Observer< TestInstanceResult >
virtual void notify (const TestInstanceResult &data)
 

Protected Member Functions

bool train_discrete (TimeSeriesClassificationData &trainingData)
 
bool train_continuous (TimeSeriesClassificationData &trainingData)
 
bool predict_discrete (VectorFloat &inputVector)
 
bool predict_continuous (VectorFloat &inputVector)
 
bool predict_discrete (MatrixFloat &timeseries)
 
bool predict_continuous (MatrixFloat &timeseries)
 
bool convertDataToObservationSequence (TimeSeriesClassificationData &classData, Vector< Vector< UINT > > &observationSequences)
 
bool loadLegacyModelFromFile (std::fstream &file)
 
- Protected Member Functions inherited from Classifier
bool saveBaseSettingsToFile (std::fstream &file) const
 
bool loadBaseSettingsFromFile (std::fstream &file)
 
- Protected Member Functions inherited from MLBase
bool saveBaseSettingsToFile (std::fstream &file) const
 
bool loadBaseSettingsFromFile (std::fstream &file)
 
- Protected Member Functions inherited from GRTBase
Float SQR (const Float &x) const
 

Protected Attributes

UINT hmmType
 Controls if this is a HMM_DISCRETE or a HMM_CONTINUOUS.
 
UINT modelType
 
UINT delta
 
UINT numStates
 
UINT numSymbols
 
UINT numRandomTrainingIterations
 
Vector< DiscreteHiddenMarkovModeldiscreteModels
 
UINT downsampleFactor
 
UINT committeeSize
 
Float sigma
 
bool autoEstimateSigma
 
Vector< ContinuousHiddenMarkovModelcontinuousModels
 
- Protected Attributes inherited from Classifier
std::string classifierType
 
bool supportsNullRejection
 
bool useNullRejection
 
UINT numClasses
 
UINT predictedClassLabel
 
UINT classifierMode
 
Float nullRejectionCoeff
 
Float maxLikelihood
 
Float bestDistance
 
Float phase
 
VectorFloat classLikelihoods
 
VectorFloat classDistances
 
VectorFloat nullRejectionThresholds
 
Vector< UINT > classLabels
 
Vector< MinMaxranges
 
- Protected Attributes inherited from MLBase
bool trained
 
bool useScaling
 
DataType inputType
 
DataType outputType
 
UINT baseType
 
UINT numInputDimensions
 
UINT numOutputDimensions
 
UINT numTrainingIterationsToConverge
 
UINT minNumEpochs
 
UINT maxNumEpochs
 
UINT validationSetSize
 
Float learningRate
 
Float minChange
 
Float rootMeanSquaredTrainingError
 
Float totalSquaredTrainingError
 
Float validationSetAccuracy
 
bool useValidationSet
 
bool randomiseTrainingOrder
 
VectorFloat validationSetPrecision
 
VectorFloat validationSetRecall
 
Random random
 
std::vector< TrainingResult > trainingResults
 
TrainingResultsObserverManager trainingResultsObserverManager
 
TestResultsObserverManager testResultsObserverManager
 
- Protected Attributes inherited from GRTBase
std::string classType
 
DebugLog debugLog
 
ErrorLog errorLog
 
InfoLog infoLog
 
TrainingLog trainingLog
 
TestingLog testingLog
 
WarningLog warningLog
 

Static Protected Attributes

static RegisterClassifierModule< HMMregisterModule
 

Additional Inherited Members

- Public Types inherited from Classifier
typedef std::map< std::string, Classifier *(*)() > StringClassifierMap
 
- Public Types inherited from MLBase
enum  BaseTypes { BASE_TYPE_NOT_SET =0, CLASSIFIER, REGRESSIFIER, CLUSTERER }
 
- Static Public Member Functions inherited from Classifier
static ClassifiercreateInstanceFromString (std::string const &classifierType)
 
static Vector< std::string > getRegisteredClassifiers ()
 
- Static Public Member Functions inherited from GRTBase
static std::string getGRTVersion (bool returnRevision=true)
 
static std::string getGRTRevison ()
 
- Protected Types inherited from Classifier
enum  ClassifierModes { STANDARD_CLASSIFIER_MODE =0, TIMESERIES_CLASSIFIER_MODE }
 
- Static Protected Member Functions inherited from Classifier
static StringClassifierMapgetMap ()
 

Detailed Description

GRT MIT License Copyright (c) <2012> <Nicholas Gillian, Media Lab, MIT>

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Definition at line 41 of file HMM.h.

Constructor & Destructor Documentation

HMM::HMM ( const UINT  hmmType = HMM_CONTINUOUS,
const UINT  modelType = HMM_LEFTRIGHT,
const UINT  delta = 1,
const bool  useScaling = false,
const bool  useNullRejection = false 
)

Default Constructor.

Sets up the HMM instance with default parameters.

Parameters
hmmTypesets the HMM type, this should be either HMM_DISCRETE or HMM_CONTINUOUS. Default: HMM_CONTINUOUS
modelTypesets the model type used by either the discrete or continuous hmm. This should be either HMM_ERGODIC or HMM_LEFTRIGHT. Default: HMM_LEFTRIGHT
deltasets the number of states a model can transistion to for a HMM_LEFTRIGHT model. Default: 1
useScalingsets if the training/input data should be scaled to the range [0 1]. Default: false
useNullRejectionsets if the algorithm should use null rejection (i.e., automatically gesture spotting). Default: false

Definition at line 28 of file HMM.cpp.

HMM::HMM ( const HMM rhs)

Default Constructor.

Copies the settings/models from the rhs HMM instance to this instance.

Parameters
rhsanother HMM instance from which the settings/models will be copied to this instance

Definition at line 57 of file HMM.cpp.

HMM::~HMM ( void  )
virtual

Default Destructor.

Definition at line 67 of file HMM.cpp.

Member Function Documentation

bool HMM::clear ( )
virtual

This overrides the clear function in the Classifier base class. It will completely clear the ML module, removing any trained model and setting all the base variables to their default values.

Returns
returns true if the module was cleared succesfully, false otherwise

Reimplemented from Classifier.

Definition at line 648 of file HMM.cpp.

bool HMM::deepCopyFrom ( const Classifier classifier)
virtual

This is required for the Gesture Recognition Pipeline for when the pipeline.setClassifier(...) method is called. It clones the data from the Base Class Classifier pointer (which should be pointing to an HMM instance) into this instance

Parameters
classifiera pointer to the Classifier Base Class, this should be pointing to another HMM instance
Returns
returns true if the clone was successfull, false otherwise

Reimplemented from Classifier.

Definition at line 91 of file HMM.cpp.

Vector< ContinuousHiddenMarkovModel > HMM::getContinuousModels ( ) const

This function gets returns a Vector of trained ContinuousHiddenMarkovModels. There will be one HiddenMarkovModel for each sample in the training data. This is only relevant if the HMM model type is HMM_CONTINUOUS.

Returns
returns the trained ContinuousHiddenMarkovModels

Definition at line 953 of file HMM.cpp.

UINT HMM::getDelta ( ) const

This function gets the delta value used for each HMM. The delta value controls how many states a model can transition to if the LEFTRIGHT model type is used.

Returns
returns the delta parameter for each HMM

Definition at line 933 of file HMM.cpp.

Vector< DiscreteHiddenMarkovModel > HMM::getDiscreteModels ( ) const

This function gets returns a Vector of trained DiscreteHiddenMarkovModels. There will be one HiddenMarkovModel for each class in the training data. This is only relevant if the HMM model type is HMM_DISCRETE.

Returns
returns the trained DiscreteHiddenMarkovModels

Definition at line 949 of file HMM.cpp.

UINT HMM::getHMMType ( ) const

This function gets the HMM type. This will be either HMM_DISCRETE or HMM_CONTINUOUS.

Returns
returns the current HMM type

Definition at line 925 of file HMM.cpp.

UINT HMM::getModelType ( ) const

This function gets the model type for each HMM, this will be one of the HMM enum ModelTypes.

Returns
returns the model type for each HMM

Definition at line 929 of file HMM.cpp.

UINT HMM::getNumRandomTrainingIterations ( ) const

This function gets the number of random training iterations used to train each discrete HMM.

The accuracy of the Baum Welch algorithm can be effected by the starting values of the A, B, and PI matrices. The HMM algorithm therefore tries several different starting values and then continues to train the algorithm with the best settings.

The number of random training iterations parameter therefore controls the number of different starting values.

Returns
returns the maximum number of iterations used to trained each HMM

Definition at line 945 of file HMM.cpp.

UINT HMM::getNumStates ( ) const

This function gets the number of states in each HMM. This is only relevant if the HMM model type is HMM_DISCRETE.

Returns
returns the number of states in each discrete HMM

Definition at line 937 of file HMM.cpp.

UINT HMM::getNumSymbols ( ) const

This function gets the number of symbols in each HMM. This is only relevant if the HMM model type is HMM_DISCRETE.

Returns
returns the number of symbols in each discrete HMM

Definition at line 941 of file HMM.cpp.

bool HMM::loadModelFromFile ( std::fstream &  file)
virtual

This loads a trained HMM model from a file. This overrides the loadModelFromFile function in the Classifier base class.

Parameters
filea reference to the file the HMM model will be loaded from
Returns
returns true if the model was loaded successfully, false otherwise

Reimplemented from MLBase.

Definition at line 757 of file HMM.cpp.

HMM & HMM::operator= ( const HMM rhs)

Custom Equals Operator.

Copies the settings/models from the rhs HMM instance to this instance.

Parameters
rhsanother HMM instance from which the settings/models will be copied to this instance
Returns
returns a reference to this instance

Definition at line 71 of file HMM.cpp.

bool HMM::predict_ ( VectorFloat inputVector)
virtual

This predicts the class of the inputVector. This overrides the predict function in the Classifier base class.

Parameters
inputVectorthe input Vector to classify
Returns
returns true if the prediction was performed, false otherwise

Reimplemented from MLBase.

Definition at line 282 of file HMM.cpp.

bool HMM::predict_ ( MatrixFloat timeseries)
virtual

This predicts the class of the timeseries. This overrides the predict function in the Classifier base class.

Parameters
timeSeriesthe input timeseries to classify
Returns
returns true if the prediction was performed, false otherwise

Reimplemented from MLBase.

Definition at line 454 of file HMM.cpp.

bool HMM::print ( ) const
virtual

This is the main print interface for all the GRT machine learning algorithms. This should be overwritten by the derived class. It will print the model and settings to the display log.

Returns
returns true if the model was printed succesfully, false otherwise (the base class always returns true)

Reimplemented from MLBase.

Definition at line 659 of file HMM.cpp.

bool HMM::reset ( )
virtual

This resets the HMM classifier.

Returns
returns true if the HMM model was successfully reset, false otherwise.

Reimplemented from Classifier.

Definition at line 627 of file HMM.cpp.

bool HMM::saveModelToFile ( std::fstream &  file) const
virtual

This saves the trained HMM model to a file. This overrides the saveModelToFile function in the Classifier base class.

Parameters
filea reference to the file the HMM model will be saved to
Returns
returns true if the model was saved successfully, false otherwise

Reimplemented from MLBase.

Definition at line 702 of file HMM.cpp.

bool HMM::setCommitteeSize ( const UINT  committeeSize)

This function sets the committeeSize used for a HMM_CONTINUOUS. The commitee size controls the number of votes used to make a prediction for a continuous HMM. For example, if the committeeSize is 5, then the top 5 estimations will be combined togethere

The parameter must be greater than zero. This will NOT clear any trained model.

Parameters
committeeSizethe committeeSize used for a continuous HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 1006 of file HMM.cpp.

bool HMM::setDelta ( const UINT  delta)

This function sets the delta parameter in each HMM.

The delta value controls how many states a model can transition to if the LEFTRIGHT model type is used.

The parameter must be greater than zero.

This will clear any trained model.

Parameters
deltathe delta parameter used for each HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 983 of file HMM.cpp.

bool HMM::setDownsampleFactor ( const UINT  downsampleFactor)

This function sets the downsample factor used for a HMM_CONTINUOUS. The downsample factor controls the resampling of each training timeseries for the continuous HMM. A downsample factor of 5, for instance, will result in each timeseries being resized (smaller) by a factor of 5. Increasing the downsample factor will significantly increase the realtime prediction time for each model, however, setting this value too high may reduce the overall accuracy of the model.

The parameter must be greater than zero. This will clear any trained model.

Parameters
downsampleFactorthe downsample factor used for a continuous HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 996 of file HMM.cpp.

bool HMM::setHMMType ( const UINT  hmmType)

This function sets the hmmType. This should be either a HMM_DISCRETE, or HMM_CONTINUOUS.

This will clear any trained model.

Parameters
hmmTypethe new hmmType
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 957 of file HMM.cpp.

bool HMM::setModelType ( const UINT  modelType)

This function sets the modelType used for each HMM. This should be one of the HMM modelType enums.

This will clear any trained model.

Parameters
modelTypethe modelType in each HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 970 of file HMM.cpp.

bool HMM::setNumRandomTrainingIterations ( const UINT  numRandomTrainingIterations)

This function sets the number of random training iterations used to train each HMM.

The accuracy of the Baum Welch algorithm can be effected by the starting values of the A, B, and PI matrices. The HMM algorithm therefore tries several different starting values and then continues to train the algorithm with the best settings. The number of random training iterations parameter therefore controls the number of different starting values.

This will clear any trained model.

Parameters
numRandomTrainingIterationsthe number of random training iterations used to train each HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 1042 of file HMM.cpp.

bool HMM::setNumStates ( const UINT  numStates)

This function sets the number of states for a HMM_DISCRETE.

The parameter must be greater than zero. This will clear any trained model.

Parameters
numStatesthe number of states in each HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 1016 of file HMM.cpp.

bool HMM::setNumSymbols ( const UINT  numStates)

This function sets the number of symbols for a HMM_DISCRETE. The parameter must be greater than zero. This will clear any trained model.

Parameters
symbolsthe number of symbols in each HMM
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 1029 of file HMM.cpp.

bool HMM::setSigma ( const Float  sigma)

This function sets the sigma parameter used for HMM_CONTINUOUS.

Sigma controls the width of the Gaussian distance function that estimates the similarity between the input sample and the model.

Modifing sigma will NOT clear the trained model, enabling the user to tune this parameter when needed.

Sigma must be greater than zero.

Parameters
sigmathe new sigma parameter
Returns
returns true if the parameter was set correctly, false otherwise

Definition at line 1055 of file HMM.cpp.

bool HMM::train ( ClassificationData  trainingData)
virtual

This overrides the train function in the Classifier base class. It simply prints a warning message stating that the bool train(LabelledTimeSeriesClassificationData trainingData) function should be used to train the HMM model.

Parameters
trainingDataa reference to the training data
Returns
returns true if the HMM model was trained, false otherwise

Reimplemented from MLBase.

Definition at line 119 of file HMM.cpp.

bool HMM::train_ ( TimeSeriesClassificationData trainingData)
virtual

This trains the HMM model, using the labelled timeseries classification data. This is the main training function for the HMM class. This overrides the train function in the Classifier base class.

Parameters
trainingDataa reference to the training data
Returns
returns true if the HMM model was trained, false otherwise

Reimplemented from MLBase.

Definition at line 125 of file HMM.cpp.


The documentation for this class was generated from the following files: