GestureRecognitionPipeline.cpp

/*
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.
*/

#define GRT_DLL_EXPORTS
#include "GestureRecognitionPipeline.h"

GRT_BEGIN_NAMESPACE

GestureRecognitionPipeline::GestureRecognitionPipeline(void)
{
 init();

 debugLog.setProceedingText("[DEBUG GRP]");
 errorLog.setProceedingText("[ERROR GRP]");
 warningLog.setProceedingText("[WARNING GRP]");
 testingLog.setProceedingText("[TEST GRP]");
}

GestureRecognitionPipeline::GestureRecognitionPipeline(const GestureRecognitionPipeline &rhs){
 
 init();
 
 debugLog.setProceedingText("[DEBUG GRP]");
 errorLog.setProceedingText("[ERROR GRP]");
 warningLog.setProceedingText("[WARNING GRP]");
 testingLog.setProceedingText("[TEST GRP]");

 //Invoke the equals operator to copy the rhs data to this instance
 *this = rhs;
}


GestureRecognitionPipeline& GestureRecognitionPipeline::operator=(const GestureRecognitionPipeline &rhs){
 
 if( this != &rhs ){
 this->clear();
 
 //Copy the pipeline variables
 this->initialized = rhs.initialized;
 this->trained = rhs.trained;
 this->info = rhs.info;
 this->inputVectorDimensions = rhs.inputVectorDimensions;
 this->outputVectorDimensions = rhs.outputVectorDimensions;
 this->predictedClassLabel = rhs.predictedClassLabel;
 this->predictedClusterLabel = rhs.predictedClusterLabel;
 this->pipelineMode = rhs.pipelineMode;
 this->predictionModuleIndex = rhs.predictionModuleIndex;
 this->numTrainingSamples = rhs.numTrainingSamples;
 this->numTestSamples = rhs.numTestSamples;
 this->testAccuracy = rhs.testAccuracy;
 this->testRMSError = rhs.testRMSError;
 this->testSquaredError = rhs.testSquaredError;
 this->testTime = rhs.testTime;
 this->trainingTime = rhs.trainingTime;
 this->testFMeasure = rhs.testFMeasure;
 this->testPrecision = rhs.testPrecision;
 this->testRecall = rhs.testRecall;
 this->regressionData = rhs.regressionData;
 this->testRejectionPrecision = rhs.testRejectionPrecision;
 this->testRejectionRecall = rhs.testRejectionRecall;
 this->testConfusionMatrix = rhs.testConfusionMatrix;
 this->crossValidationResults = rhs.crossValidationResults;
 this->testResults = rhs.testResults;

 //Copy the GRT Base variables
 this->debugLog = rhs.debugLog;
 this->errorLog = rhs.errorLog;
 this->trainingLog = rhs.trainingLog;
 this->testingLog = rhs.testingLog;
 this->warningLog = rhs.warningLog;
 
 for(unsigned int i=0; i<rhs.preProcessingModules.size(); i++){
 this->addPreProcessingModule( *(rhs.preProcessingModules[i]) );
 }
 
 for(unsigned int i=0; i<rhs.featureExtractionModules.size(); i++){
 this->addFeatureExtractionModule( *(rhs.featureExtractionModules[i]) );
 }
 
 if( rhs.getIsPipelineInClassificationMode() ){
 setClassifier( *rhs.classifier );
 }
 
 if( rhs.getIsPipelineInRegressionMode() ){
 setRegressifier( *rhs.regressifier );
 }
 
 if( rhs.getIsClustererSet() ){
 setClusterer( *rhs.clusterer );
 }
 
 for(unsigned int i=0; i<rhs.postProcessingModules.size(); i++){
 this->addPostProcessingModule( *(rhs.postProcessingModules[i]) );
 }
 
 for(unsigned int k=0; k<NUM_CONTEXT_LEVELS; k++){
 for(unsigned int i=0; i<rhs.contextModules[k].size(); i++){
 this->addContextModule( *(rhs.contextModules[k][i]), k );
 }
 }

 //Adding the modules will automatically set trained to false, so make sure we get the correct state
 this->trained = rhs.trained;
 }
 
 return *this;
}

GestureRecognitionPipeline& GestureRecognitionPipeline::operator << (const PreProcessing &module) {
 this->addPreProcessingModule( module );
 return *this;
}

GestureRecognitionPipeline& GestureRecognitionPipeline::operator << (const FeatureExtraction &module) {
 this->addFeatureExtractionModule( module );
 return *this;
}

GestureRecognitionPipeline& GestureRecognitionPipeline::operator << (const Classifier &module) {
 this->setClassifier( module );
 return *this;
}

GestureRecognitionPipeline& GestureRecognitionPipeline::operator << (const Regressifier &module) {
 this->setRegressifier( module );
 return *this;
}

GestureRecognitionPipeline& GestureRecognitionPipeline::operator << (const Clusterer &module) {
 this->setClusterer( module );
 return *this;
}

GestureRecognitionPipeline& GestureRecognitionPipeline::operator << (const PostProcessing &module) {
 this->addPostProcessingModule( module );
 return *this;
}

GestureRecognitionPipeline::~GestureRecognitionPipeline(void)
{
 //Clean up the memory
 clear();
} 
 
bool GestureRecognitionPipeline::train(const ClassificationData &trainingData){
 
 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsClassifierSet() ){
 errorLog << "train(ClassificationData trainingData) - Failed To Train Classifier, the classifier has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(ClassificationData trainingData) - Failed To Train Classifier, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Set the input Vector dimension size
 inputVectorDimensions = trainingData.getNumDimensions();
 
 //Pass the training data through any pre-processing or feature extraction units
 UINT numDimensions = trainingData.getNumDimensions();
 
 //If there are any preprocessing or feature extraction modules, then get the size of the last module
 if( getIsPreProcessingSet() || getIsFeatureExtractionSet() ){
 if( getIsFeatureExtractionSet() ){
 numDimensions = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }else{
 numDimensions = preProcessingModules[ preProcessingModules.size()-1 ]->getNumOutputDimensions();
 }
 }
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 ClassificationData processedTrainingData( numDimensions );
 processedTrainingData.reserve( trainingData.getNumSamples() );
 UINT classLabel = 0;
 VectorFloat trainingSample;
 for(UINT i=0; i<trainingData.getNumSamples(); i++){
 bool okToAddProcessedData = true;
 classLabel = trainingData[i].getClassLabel();
 trainingSample = trainingData[i].getSample();
 
 //Perform any preprocessing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[moduleIndex]->process( trainingSample ) ){
 errorLog << "train(ClassificationData trainingData) - Failed to PreProcess Training Data. PreProcessingModuleIndex: ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 trainingSample = preProcessingModules[moduleIndex]->getProcessedData();
 }
 }
 
 //Compute any features
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[moduleIndex]->computeFeatures( trainingSample ) ){
 errorLog << "train(ClassificationData trainingData) - Failed to Compute Features from Training Data. FeatureExtractionModuleIndex ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 if( featureExtractionModules[moduleIndex]->getFeatureDataReady() ){
 trainingSample = featureExtractionModules[moduleIndex]->getFeatureVector();
 }else{
 okToAddProcessedData = false;
 break;
 }
 }
 }

 if( okToAddProcessedData ){
 //Add the training sample to the processed training data
 processedTrainingData.addSample(classLabel, trainingSample);
 }
 
 }
 
 if( processedTrainingData.getNumSamples() != trainingData.getNumSamples() ){
 warningLog << "train(ClassificationData trainingData) - Lost " << trainingData.getNumSamples()-processedTrainingData.getNumSamples() << " of " << trainingData.getNumSamples() << " training samples due to the processing stage!" << std::endl;
 }

 //Store the number of training samples
 numTrainingSamples = processedTrainingData.getNumSamples();
 
 //Train the classifier
 trained = classifier->train_( processedTrainingData );
 if( !trained ){
 errorLog << "train(ClassificationData trainingData) - Failed To Train Classifier: " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::train(const ClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling){
 
 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsClassifierSet() ){
 errorLog << "train(const ClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed To Train Classifier, the classifier has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(const ClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed To Train Classifier, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 //Get a copy of the data so we can spilt it
 ClassificationData data = trainingData;

 //Spilt the data into K folds
 bool spiltResult = data.spiltDataIntoKFolds(kFoldValue, useStratifiedSampling);
 
 if( !spiltResult ){
 return false;
 }
 
 //Run the k-fold training and testing
 Float crossValidationAccuracy = 0;
 ClassificationData foldTrainingData;
 ClassificationData foldTestData;
 Vector< TestResult > cvResults(kFoldValue);

 for(UINT k=0; k<kFoldValue; k++){
 foldTrainingData = data.getTrainingFoldData(k);
 
 if( !train( foldTrainingData ) ){
 return false;
 }
 
 //Test the classification system
 foldTestData = data.getTestFoldData(k);
 
 if( !test( foldTestData ) ){
 return false;
 }
 
 crossValidationAccuracy += getTestAccuracy();
 cvResults[k] = getTestResults();
 }

 //Flag that the model has been trained
 trained = true;
 
 //Set the accuracy of the classification system averaged over the kfolds
 testAccuracy = crossValidationAccuracy / Float(kFoldValue);
 crossValidationResults = cvResults;
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}

bool GestureRecognitionPipeline::train( const TimeSeriesClassificationData &trainingData ){
 
 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsClassifierSet() ){
 errorLog << "train(const TimeSeriesClassificationData &trainingData) - Failed To Train Classifier, the classifier has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To Train Classifier, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 //Set the input Vector dimension size of the pipeline
 inputVectorDimensions = trainingData.getNumDimensions();
 
 TimeSeriesClassificationData processedTrainingData( trainingData.getNumDimensions() );
 TimeSeriesClassificationData timeseriesClassificationData;
 ClassificationData classificationData;

 bool allowNullGestureClass = true;
 processedTrainingData.setAllowNullGestureClass( allowNullGestureClass );
 timeseriesClassificationData.setAllowNullGestureClass( allowNullGestureClass );
 classificationData.setAllowNullGestureClass( allowNullGestureClass );
 
 //Setup the data structure, if the classifier works with timeseries data then we use TimeSeriesClassificationData
 //otherwise we format the data as ClassificationData
 if( classifier->getTimeseriesCompatible() ){
 UINT trainingDataInputDimensionSize = trainingData.getNumDimensions();
 if( getIsPreProcessingSet() ){
 trainingDataInputDimensionSize = preProcessingModules[ preProcessingModules.size()-1 ]->getNumOutputDimensions();
 }
 if( getIsFeatureExtractionSet() ){
 trainingDataInputDimensionSize = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }
 timeseriesClassificationData.setNumDimensions( trainingDataInputDimensionSize );
 }else{
 UINT trainingDataInputDimensionSize = trainingData.getNumDimensions();
 if( getIsPreProcessingSet() ){
 trainingDataInputDimensionSize = preProcessingModules[ preProcessingModules.size()-1 ]->getNumOutputDimensions();
 }
 if( getIsFeatureExtractionSet() ){
 trainingDataInputDimensionSize = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }
 classificationData.setNumDimensions( trainingDataInputDimensionSize );
 }
 
 //Pass the timeseries data through any pre-processing modules and add it to the processedTrainingData structure
 for(UINT i=0; i<trainingData.getNumSamples(); i++){
 UINT classLabel = trainingData[i].getClassLabel();
 MatrixFloat trainingSample = trainingData[i].getData();
 
 if( getIsPreProcessingSet() ){
 
 //Try to process the matrix data row-by-row
 bool resetPreprocessingModule = true;
 for(UINT r=0; r<trainingSample.getNumRows(); r++){
 VectorFloat sample = trainingSample.getRow( r );
 
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 
 if( resetPreprocessingModule ){
 preProcessingModules[moduleIndex]->reset();
 }
 
 //Validate the input and output dimensions match!
 if( preProcessingModules[moduleIndex]->getNumInputDimensions() != preProcessingModules[moduleIndex]->getNumOutputDimensions() ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To PreProcess Training Data. The number of inputDimensions (";
 errorLog << preProcessingModules[moduleIndex]->getNumInputDimensions();
 errorLog << ") in PreProcessingModule ";
 errorLog << moduleIndex;
 errorLog << " do not match the number of outputDimensions (";
 errorLog << preProcessingModules[moduleIndex]->getNumOutputDimensions();
 errorLog << std::endl;
 return false;
 }
 
 if( !preProcessingModules[moduleIndex]->process( sample ) ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To PreProcess Training Data. PreProcessingModuleIndex: ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 sample = preProcessingModules[moduleIndex]->getProcessedData();
 }
 
 //The preprocessing modules should only be reset when r==0
 resetPreprocessingModule = false;
 
 //Overwrite the original training sample with the preProcessed sample
 for(UINT c=0; c<sample.size(); c++){
 trainingSample[r][c] = sample[c];
 }
 }
 
 }
 
 //Add the training sample to the processed training data
 processedTrainingData.addSample(classLabel,trainingSample);
 }
 
 //Loop over the processed training data, perfrom any feature extraction if needed
 //Add the data to either the timeseries or classification data structures
 for(UINT i=0; i<processedTrainingData.getNumSamples(); i++){
 UINT classLabel = processedTrainingData[i].getClassLabel();
 MatrixFloat trainingSample = processedTrainingData[i].getData();
 bool featureDataReady = false;
 bool resetFeatureExtractionModules = true;
 
 VectorFloat inputVector;
 MatrixFloat featureData;
 //Try to process the matrix data row-by-row
 for(UINT r=0; r<trainingSample.getNumRows(); r++){
 inputVector = trainingSample.getRow( r );
 featureDataReady = true;
 
 //Pass the processed training data through the feature extraction
 if( getIsFeatureExtractionSet() ){
 
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 
 if( resetFeatureExtractionModules ){
 featureExtractionModules[moduleIndex]->reset();
 }
 
 if( !featureExtractionModules[moduleIndex]->computeFeatures( inputVector ) ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To Compute Features For Training Data. FeatureExtractionModuleIndex: ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 
 //Overwrite the input Vector with the features so this can either be input to the next feature module 
 //or converted to the LabelledClassificationData format
 inputVector = featureExtractionModules[moduleIndex]->getFeatureVector();
 featureDataReady = featureExtractionModules[moduleIndex]->getFeatureDataReady();
 
 //If feature data is not ready at any stage, stop the rest of the pipeline.
 if( !featureDataReady ) { break; }
 }
 
 //The feature extraction modules should only be reset on r == 0
 resetFeatureExtractionModules = false;
 
 if( featureDataReady ){
 
 if( classifier->getTimeseriesCompatible() ){
 if( !featureData.push_back( inputVector ) ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To add feature Vector to feature data matrix! FeatureExtractionModuleIndex: " << std::endl;
 return false;
 }
 }else classificationData.addSample(classLabel, inputVector);
 }
 
 }else{
 if( classifier->getTimeseriesCompatible() ){
 if( !featureData.push_back( inputVector ) ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To add feature Vector to feature data matrix! FeatureExtractionModuleIndex: " << std::endl;
 return false;
 }
 }
 else classificationData.addSample(classLabel, inputVector);
 }
 }
 
 if( classifier->getTimeseriesCompatible() ) timeseriesClassificationData.addSample(classLabel, featureData);
 
 }
 
 //Train the classification system
 if( classifier->getTimeseriesCompatible() ){
 numTrainingSamples = timeseriesClassificationData.getNumSamples();
 trained = classifier->train( timeseriesClassificationData );
 }else{
 numTrainingSamples = classificationData.getNumSamples();
 trained = classifier->train( classificationData );
 }

 if( !trained ){
 errorLog << "train(TimeSeriesClassificationData trainingData) - Failed To Train Classifier" << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::train(const TimeSeriesClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling){
 
 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsClassifierSet() ){
 errorLog << "train(const TimeSeriesClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed To Train Classifier, the classifier has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(const TimeSeriesClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed To Train Classifier, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Start the training timer
 Timer timer;
 timer.start();

 //Get a copy of the data so we can split it
 TimeSeriesClassificationData data = trainingData;
 
 //Spilt the data into K folds
 if( !data.spiltDataIntoKFolds(kFoldValue, useStratifiedSampling) ){
 errorLog << "train(const TimeSeriesClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed To Spilt Dataset into KFolds!" << std::endl;
 return false;
 }
 
 //Run the k-fold training and testing
 Float crossValidationAccuracy = 0;
 TimeSeriesClassificationData foldTrainingData;
 TimeSeriesClassificationData foldTestData;
 
 for(UINT k=0; k<kFoldValue; k++){
 foldTrainingData = data.getTrainingFoldData(k);
 
 if( !train( foldTrainingData ) ){
 errorLog << "train(const TimeSeriesClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed to train pipeline for fold " << k << "." << std::endl;
 return false;
 }
 
 //Test the classification system
 foldTestData = data.getTestFoldData(k);
 
 if( !test( foldTestData ) ){
 errorLog << "train(const TimeSeriesClassificationData &trainingData,const UINT kFoldValue,const bool useStratifiedSampling) - Failed to test pipeline for fold " << k << "." << std::endl;
 return false;
 }
 
 crossValidationAccuracy += getTestAccuracy();
 }

 //Flag that the model has been trained
 trained = true;
 
 //Set the accuracy of the classification system averaged over the kfolds
 testAccuracy = crossValidationAccuracy / Float(kFoldValue);
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}

bool GestureRecognitionPipeline::train(const ClassificationDataStream &trainingData){

 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsClassifierSet() ){
 errorLog << "train(ClassificationDataStream trainingData) - Failed To train Classifier, the classifier has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(ClassificationDataStream trainingData) - Failed To train Classifier, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Set the input Vector dimension size
 inputVectorDimensions = trainingData.getNumDimensions();
 
 //Pass the training data through any pre-processing or feature extraction units
 UINT numDimensions = inputVectorDimensions;
 
 //If there are any preprocessing or feature extraction modules, then get the size of the last module
 if( getIsPreProcessingSet() || getIsFeatureExtractionSet() ){
 if( getIsFeatureExtractionSet() ){
 numDimensions = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }else{
 numDimensions = preProcessingModules[ preProcessingModules.size()-1 ]->getNumOutputDimensions();
 }
 }
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 ClassificationData processedTrainingData( numDimensions );
 processedTrainingData.reserve( trainingData.getNumSamples() );
 UINT classLabel = 0;
 VectorFloat trainingSample;
 for(UINT i=0; i<trainingData.getNumSamples(); i++){
 bool okToAddProcessedData = true;
 classLabel = trainingData[i].getClassLabel();
 trainingSample = trainingData[i].getSample();
 
 //Perform any preprocessing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[moduleIndex]->process( trainingSample ) ){
 errorLog << "train(ClassificationDataStream trainingData) - Failed to PreProcess training Data. PreProcessingModuleIndex: ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 trainingSample = preProcessingModules[moduleIndex]->getProcessedData();
 }
 }
 
 //Compute any features
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[moduleIndex]->computeFeatures( trainingSample ) ){
 errorLog << "train(ClassificationDataStream trainingData) - Failed to Compute Features from training Data. FeatureExtractionModuleIndex ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 if( featureExtractionModules[moduleIndex]->getFeatureDataReady() ){
 trainingSample = featureExtractionModules[moduleIndex]->getFeatureVector();
 }else{
 okToAddProcessedData = false;
 break;
 }
 }
 }

 if( okToAddProcessedData ){
 //Add the training sample to the processed training data
 processedTrainingData.addSample(classLabel, trainingSample);
 }
 
 }
 
 if( processedTrainingData.getNumSamples() != trainingData.getNumSamples() ){
 warningLog << "train(ClassificationDataStream trainingData) - Lost " << trainingData.getNumSamples()-processedTrainingData.getNumSamples() << " of " << trainingData.getNumSamples() << " training samples due to the processing stage!" << std::endl;
 }

 //Store the number of training samples
 numTrainingSamples = processedTrainingData.getNumSamples();
 
 //Train the classifier
 trained = classifier->train_( processedTrainingData );
 if( !trained ){
 errorLog << "train(ClassificationDataStream trainingData) - Failed To Train Classifier: " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();

 return true;
}
 
bool GestureRecognitionPipeline::train(const RegressionData &trainingData){
 
 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 //Reset all the modules
 reset();
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 //Set the input Vector dimension size
 inputVectorDimensions = trainingData.getNumInputDimensions();
 
 //Pass the training data through any pre-processing or feature extraction units
 RegressionData processedTrainingData;
 
 //Set the dimensionality of the data
 UINT numInputs = 0;
 UINT numTargets = trainingData.getNumTargetDimensions();
 if( !getIsPreProcessingSet() && !getIsFeatureExtractionSet() ){
 numInputs = trainingData.getNumInputDimensions();
 }else{
 
 if( getIsPreProcessingSet() && !getIsFeatureExtractionSet() ){
 numInputs = preProcessingModules[ preProcessingModules.size()-1 ]->getNumOutputDimensions();
 }
 
 if( !getIsPreProcessingSet() && getIsFeatureExtractionSet() ){
 numInputs = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }
 
 if( getIsPreProcessingSet() && getIsFeatureExtractionSet() ){
 numInputs = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }
 }
 
 processedTrainingData.setInputAndTargetDimensions(numInputs, numTargets);
 
 for(UINT i=0; i<trainingData.getNumSamples(); i++){
 VectorFloat inputVector = trainingData[i].getInputVector();
 VectorFloat targetVector = trainingData[i].getTargetVector();
 
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[ moduleIndex ]->process( inputVector ) ){
 errorLog << "train(const RegressionData &trainingData) - Failed To Compute Features For Training Data. PreProcessingModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 inputVector = preProcessingModules[ moduleIndex ]->getProcessedData();
 }
 }
 
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[ moduleIndex ]->computeFeatures( inputVector ) ){
 errorLog << "train(const RegressionData &trainingData) - Failed To Compute Features For Training Data. FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 inputVector = featureExtractionModules[ moduleIndex ]->getFeatureVector();
 }
 }
 
 //Add the training sample to the processed training data
 if( !processedTrainingData.addSample(inputVector,targetVector) ){
 errorLog << "train(const RegressionData &trainingData) - Failed to add processed training sample to training data" << std::endl;
 return false;
 }
 }
 
 //Store the number of training samples
 numTrainingSamples = processedTrainingData.getNumSamples();
 
 //Train the classification system
 if( getIsRegressifierSet() ){
 trained = regressifier->train( processedTrainingData );
 if( !trained ){
 errorLog << "train(const RegressionData &trainingData) - Failed To Train Regressifier: " << regressifier->getLastErrorMessage() << std::endl;
 return false;
 }
 }else{
 errorLog << "train(const RegressionData &trainingData) - Classifier is not set" << std::endl;
 return false;
 }
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::train(const RegressionData &trainingData,const UINT kFoldValue){

 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsRegressifierSet() ){
 errorLog << "train(const RegressionData &trainingData,const UINT kFoldValue) - Failed To Train Regressifier, the regressifier has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(const RegressionData &trainingData,const UINT kFoldValue) - Failed To Train Regressifier, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 //Get a copy of the training data so we can split it
 RegressionData data = trainingData;

 //Spilt the data into K folds
 bool spiltResult = data.spiltDataIntoKFolds(kFoldValue);
 
 if( !spiltResult ){
 return false;
 }
 
 //Run the k-fold training and testing
 Float crossValidationAccuracy = 0;
 RegressionData foldTrainingData;
 RegressionData foldTestData;
 for(UINT k=0; k<kFoldValue; k++){
 foldTrainingData = data.getTrainingFoldData(k);
 
 if( !train( foldTrainingData ) ){
 return false;
 }
 
 //Test the classification system
 foldTestData = data.getTestFoldData(k);
 
 if( !test( foldTestData ) ){
 return false;
 }

 crossValidationAccuracy += getTestRMSError();
 
 }

 //Flag that the model has been trained
 trained = true;

 testAccuracy = crossValidationAccuracy / Float(kFoldValue);
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::train(const UnlabelledData &trainingData){
 
 trained = false;
 trainingTime = 0;
 clearTestResults();
 
 if( !getIsClustererSet() ){
 errorLog << "train(const UnlabelledData &trainingData) - Failed To Train Clusterer, the clusterer has not been set!" << std::endl;
 return false;
 }
 
 if( trainingData.getNumSamples() == 0 ){
 errorLog << "train(const UnlabelledData &trainingData) - Failed To Train Clusterer, there is no training data!" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Set the input Vector dimension size
 inputVectorDimensions = trainingData.getNumDimensions();
 
 //Pass the training data through any pre-processing or feature extraction units
 UINT numDimensions = trainingData.getNumDimensions();
 
 //If there are any preprocessing or feature extraction modules, then get the size of the last module
 if( getIsPreProcessingSet() || getIsFeatureExtractionSet() ){
 if( getIsFeatureExtractionSet() ){
 numDimensions = featureExtractionModules[ featureExtractionModules.size()-1 ]->getNumOutputDimensions();
 }else{
 numDimensions = preProcessingModules[ preProcessingModules.size()-1 ]->getNumOutputDimensions();
 }
 }
 
 //Start the training timer
 Timer timer;
 timer.start();
 
 UnlabelledData processedTrainingData( numDimensions );
 
 for(UINT i=0; i<trainingData.getNumSamples(); i++){
 bool okToAddProcessedData = true;
 VectorFloat trainingSample = trainingData[i];
 
 //Perform any preprocessing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[moduleIndex]->process( trainingSample ) ){
 errorLog << "train(const UnlabelledData &trainingData) - Failed to PreProcess Training Data. PreProcessingModuleIndex: ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 trainingSample = preProcessingModules[moduleIndex]->getProcessedData();
 }
 }
 
 //Compute any features
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[moduleIndex]->computeFeatures( trainingSample ) ){
 errorLog << "train(const UnlabelledData &trainingData) - Failed to Compute Features from Training Data. FeatureExtractionModuleIndex ";
 errorLog << moduleIndex;
 errorLog << std::endl;
 return false;
 }
 if( featureExtractionModules[moduleIndex]->getFeatureDataReady() ){
 trainingSample = featureExtractionModules[moduleIndex]->getFeatureVector();
 }else{
 okToAddProcessedData = false;
 break;
 }
 }
 }
 
 if( okToAddProcessedData ){
 //Add the training sample to the processed training data
 processedTrainingData.addSample(trainingSample);
 }
 
 }
 
 if( processedTrainingData.getNumSamples() != trainingData.getNumSamples() ){
 
 warningLog << "train(const ClassificationData &trainingData) - Lost " << trainingData.getNumSamples()-processedTrainingData.getNumSamples() << " of " << trainingData.getNumSamples() << " training samples due to the processing stage!" << std::endl;
 }
 
 //Store the number of training samples
 numTrainingSamples = processedTrainingData.getNumSamples();
 
 //Train the cluster model
 trained = clusterer->train_( processedTrainingData );
 if( !trained ){
 errorLog << "train(const UnlabelledData &trainingData) - Failed To Train Clusterer: " << clusterer->getLastErrorMessage() << std::endl;
 return false;
 }
 
 //Store the training time
 trainingTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::test(const ClassificationData &testData){
 
 //Clear any previous test results
 clearTestResults();
 
 //Make sure the classification model has been trained
 if( !trained ){
 errorLog << "test(const ClassificationData &testData) - Classifier is not trained" << std::endl;
 return false;
 }
 
 //Make sure the dimensionality of the test data matches the input Vector's dimensions
 if( testData.getNumDimensions() != inputVectorDimensions ){
 errorLog << "test(const ClassificationData &testData) - The dimensionality of the test data (" + Util::toString(testData.getNumDimensions()) + ") does not match that of the input Vector dimensions of the pipeline (" << inputVectorDimensions << ")" << std::endl;
 return false;
 }
 
 if( !getIsClassifierSet() ){
 errorLog << "test(const ClassificationData &testData) - The classifier has not been set" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();

 //Validate that the class labels in the test data match the class labels in the model
 bool classLabelValidationPassed = true;
 for(UINT i=0; i<testData.getNumClasses(); i++){
 bool labelFound = false;
 for(UINT k=0; k<classifier->getNumClasses(); k++){
 if( testData.getClassTracker()[i].classLabel == classifier->getClassLabels()[k] ){
 labelFound = true;
 break;
 }
 }

 if( !labelFound ){
 classLabelValidationPassed = false;
 errorLog << "test(const ClassificationData &testData) - The test dataset contains a class label (" << testData.getClassTracker()[i].classLabel << ") that is not in the model!" << std::endl;
 }
 }

 if( !classLabelValidationPassed ){
 errorLog << "test(const ClassificationData &testData) - Model Class Labels: ";
 for(UINT k=0; k<classifier->getNumClasses(); k++){
 errorLog << classifier->getClassLabels()[k] << "\t";
 }
 errorLog << std::endl;
 return false;
 }

 Float rejectionPrecisionCounter = 0;
 Float rejectionRecallCounter = 0;
 unsigned int confusionMatrixSize = classifier->getNullRejectionEnabled() ? classifier->getNumClasses()+1 : classifier->getNumClasses();
 VectorFloat precisionCounter(classifier->getNumClasses(), 0);
 VectorFloat recallCounter(classifier->getNumClasses(), 0);
 VectorFloat confusionMatrixCounter(confusionMatrixSize,0);
 
 //Resize the test matrix
 testConfusionMatrix.resize(confusionMatrixSize, confusionMatrixSize);
 testConfusionMatrix.setAllValues(0);
 
 //Resize the precision and recall Vectors
 testPrecision.clear();
 testRecall.clear();
 testFMeasure.clear();
 testPrecision.resize(getNumClassesInModel(), 0);
 testRecall.resize(getNumClassesInModel(), 0);
 testFMeasure.resize(getNumClassesInModel(), 0);
 testResults.resize(testData.getNumSamples());
 numTestSamples = testData.getNumSamples();
 
 //Start the test timer
 Timer timer;
 timer.start();

 //Run the test
 for(UINT i=0; i<numTestSamples; i++){
 UINT classLabel = testData[i].getClassLabel();
 VectorFloat testSample = testData[i].getSample();
 
 //Pass the test sample through the pipeline
 if( !predict( testSample ) ){
 errorLog << "test(const ClassificationData &testData) - Prediction failed for test sample at index: " << i << std::endl;
 return false;
 }
 
 //Update the test metrics
 UINT predictedClassLabel = getPredictedClassLabel();
 
 if( !updateTestMetrics(classLabel,predictedClassLabel,precisionCounter,recallCounter,rejectionPrecisionCounter,rejectionRecallCounter, confusionMatrixCounter) ){
 errorLog << "test(const ClassificationData &testData) - Failed to update test metrics at test sample index: " << i << std::endl;
 return false;
 }
 //cout << "i: " << i << " class label: " << classLabel << " predictedClassLabel: " << predictedClassLabel << std::endl;
 
 //Keep track of the classification results encase the user needs them later
 testResults[i].setClassificationResult(i, classLabel, predictedClassLabel, getUnProcessedPredictedClassLabel(),getClassLikelihoods(), getClassDistances());

 //Update any observers
 classifier->notifyTestResultsObservers( testResults[i] );
 }
 
 if( !computeTestMetrics(precisionCounter,recallCounter,rejectionPrecisionCounter,rejectionRecallCounter, confusionMatrixCounter, numTestSamples) ){
 errorLog <<"test(const ClassificationData &testData) - Failed to compute test metrics!" << std::endl;
 return false;
 }
 
 testTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::test(const TimeSeriesClassificationData &testData){

 //Clear any previous test results
 clearTestResults();
 
 //Make sure the classification model has been trained
 if( !trained ){
 errorLog << "test(const TimeSeriesClassificationData &testData) - The classifier has not been trained" << std::endl;
 return false;
 }
 
 //Make sure the dimensionality of the test data matches the input Vector's dimensions
 if( testData.getNumDimensions() != inputVectorDimensions ){
 errorLog << "test(const TimeSeriesClassificationData &testData) - The dimensionality of the test data (" << testData.getNumDimensions() << ") does not match that of the input Vector dimensions of the pipeline (" << inputVectorDimensions << ")" << std::endl;
 return false;
 }
 
 if( !getIsClassifierSet() ){
 errorLog << "test(const TimeSeriesClassificationData &testData) - The classifier has not been set" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 Float rejectionPrecisionCounter = 0;
 Float rejectionRecallCounter = 0;
 const UINT K = classifier->getNumClasses();
 UINT confusionMatrixSize = classifier->getNullRejectionEnabled() ? K+1 : K;
 VectorFloat precisionCounter(K, 0);
 VectorFloat recallCounter(K, 0);
 VectorFloat confusionMatrixCounter(confusionMatrixSize,0);
 
 //Resize the test matrix
 testConfusionMatrix.resize(confusionMatrixSize,confusionMatrixSize);
 testConfusionMatrix.setAllValues(0);
 
 //Resize the precision and recall Vectors
 testPrecision.resize(K, 0);
 testRecall.resize(K, 0);
 testFMeasure.resize(K, 0);
 numTestSamples = testData.getNumSamples();
 
 //Start the test timer
 Timer timer;
 timer.start();
 
 //Run the test
 const UINT M = testData.getNumSamples();
 for(UINT i=0; i<M; i++){
 UINT classLabel = testData[i].getClassLabel();
 MatrixFloat timeseries = testData[i].getData();
 
 //Pass the test timeseries through the pipeline
 if( !predict( timeseries ) ){
 errorLog << "test(const TimeSeriesClassificationData &testData) - Failed to run prediction for test sample index: " << i << std::endl;
 return false;
 }
 
 //Update the test metrics
 UINT predictedClassLabel = getPredictedClassLabel();
 
 if( !updateTestMetrics(classLabel,predictedClassLabel,precisionCounter,recallCounter,rejectionPrecisionCounter,rejectionRecallCounter, confusionMatrixCounter) ){
 errorLog << "test(const TimeSeriesClassificationData &testData) - Failed to update test metrics at test sample index: " << i << std::endl;
 return false;
 }
 
 }
 
 if( !computeTestMetrics(precisionCounter,recallCounter,rejectionPrecisionCounter,rejectionRecallCounter, confusionMatrixCounter, M) ){
 errorLog << "test(const TimeSeriesClassificationData &testData) - Failed to compute test metrics!" << std::endl;
 return false;
 }
 
 testTime = timer.getMilliSeconds();
 
 return true;
}
 
bool GestureRecognitionPipeline::test(const ClassificationDataStream &testData){
 
 //Clear any previous test results
 clearTestResults();
 
 //Make sure the classification model has been trained
 if( !trained ){
 errorLog << "test(const ClassificationDataStream &testData) - The classifier has not been trained" << std::endl;
 return false;
 }
 
 //Make sure the dimensionality of the test data matches the input Vector's dimensions
 if( testData.getNumDimensions() != inputVectorDimensions ){
 errorLog << "test(const ClassificationDataStream &testData) - The dimensionality of the test data (" + Util::toString(testData.getNumDimensions()) + ") does not match that of the input Vector dimensions of the pipeline (" << inputVectorDimensions << ")" << std::endl;
 return false;
 }
 
 if( !getIsClassifierSet() ){
 errorLog << "test(const ClassificationDataStream &testData) - The classifier has not been set" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 //Float rejectionPrecisionCounter = 0;
 //Float rejectionRecallCounter = 0;
 UINT confusionMatrixSize = classifier->getNullRejectionEnabled() ? classifier->getNumClasses()+1 : classifier->getNumClasses();
 VectorFloat precisionCounter(getNumClassesInModel(), 0);
 VectorFloat recallCounter(getNumClassesInModel(), 0);
 VectorFloat confusionMatrixCounter(confusionMatrixSize,0);
 
 //Resize the test matrix
 testConfusionMatrix.resize(confusionMatrixSize,confusionMatrixSize);
 testConfusionMatrix.setAllValues(0);
 
 //Resize the precision and recall Vectors
 testPrecision.resize(getNumClassesInModel(), 0);
 testRecall.resize(getNumClassesInModel(), 0);
 testFMeasure.resize(getNumClassesInModel(), 0);
 
 //Resize the classification results Vector
 testResults.resize(testData.getNumSamples());
 numTestSamples = testData.getNumSamples();
 
 //Start the test timer
 Timer timer;
 timer.start();

 //Get a copy of the data so we can modify it
 ClassificationDataStream data = testData;
 
 //Run the test
 data.resetPlaybackIndex(0); //Make sure that the test data start at 0
 for(UINT i=0; i<data.getNumSamples(); i++){
 ClassificationSample sample = data.getNextSample();
 UINT classLabel = sample.getClassLabel();
 VectorFloat testSample = sample.getSample();
 
 //Pass the test sample through the pipeline
 if( !predict( testSample ) ){
 errorLog << "test(const ClassificationDataStream &testData) - Prediction Failed! " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 
 //Update the test metrics
 UINT predictedClassLabel = getPredictedClassLabel();
 
 /* //TODO - Need to update this!
 if( !updateTestMetrics(classLabel,predictedClassLabel,precisionCounter,recallCounter,rejectionPrecisionCounter,rejectionRecallCounter, confusionMatrixCounter) ){
 errorLog << "test(LabelledContinuousTimeSeriesClassificationData &testData) - Failed to update test metrics at test sample index: " << i << std::endl;
 return false;
 }
 */
 
 if( classLabel == predictedClassLabel ) testAccuracy++;
 
 //Store the test results so they can be used later
 testResults[i].setClassificationResult(i, classLabel, predictedClassLabel, getUnProcessedPredictedClassLabel(), getClassLikelihoods(), getClassDistances());
 
 //Notify all observers of the test result
 classifier->notifyTestResultsObservers( testResults[i] );
 
 testingLog << "test iteration: " << i;
 testingLog << "\tClassLabel: " << classLabel;
 testingLog << "\tPredictedClassLabel: " << predictedClassLabel;
 testingLog << "\tLikelihood: " << getMaximumLikelihood() << std::endl;
 }
 
 /* //TODO - Need to update this!
 if( !computeTestMetrics(precisionCounter,recallCounter,rejectionPrecisionCounter,rejectionRecallCounter, confusionMatrixCounter, testData.getNumSamples()) ){
 errorLog << "test(LabelledContinuousTimeSeriesClassificationData &testData) - Failed to compute test metrics !" << std::endl;
 return false;
 }
 */
 
 testTime = timer.getMilliSeconds();
 testAccuracy = testAccuracy / Float( testData.getNumSamples() ) * 100.0;
 
 testingLog << "Test complete. Total testing time: " << testTime << std::endl;
 
 return true;
}
 
bool GestureRecognitionPipeline::test(const RegressionData &testData){
 
 //Clear any previous test results
 clearTestResults();
 
 //Make sure the classification model has been trained
 if( !trained ){
 errorLog << "test(const RegressionData &testData) - Regressifier is not trained" << std::endl;
 return false;
 }
 
 //Make sure the dimensionality of the test data matches the input Vector's dimensions
 if( testData.getNumInputDimensions() != inputVectorDimensions ){
 errorLog << "test(const RegressionData &testData) - The dimensionality of the test data (" << testData.getNumInputDimensions() << ") does not match that of the input Vector dimensions of the pipeline (" << inputVectorDimensions << ")" << std::endl;
 return false;
 }
 
 if( !getIsRegressifierSet() ){
 errorLog << "test(const RegressionData &testData) - The regressifier has not been set" << std::endl;
 return false;
 }
 
 if( regressifier->getNumOutputDimensions() != testData.getNumTargetDimensions() ){
 errorLog << "test(const RegressionData &testData) - The size of the output of the regressifier (" << regressifier->getNumOutputDimensions() << ") does not match that of the size of the number of target dimensions (" << testData.getNumTargetDimensions() << ")" << std::endl;
 return false;
 }
 
 //Reset all the modules
 reset();
 
 numTestSamples = testData.getNumSamples();
 testResults.resize( numTestSamples );
 
 //Start the test timer
 Timer timer;
 timer.start();
 
 //Run the test
 testSquaredError = 0;
 testRMSError = 0;
 for(UINT i=0; i<numTestSamples; i++){
 VectorFloat inputVector = testData[i].getInputVector();
 VectorFloat targetVector = testData[i].getTargetVector();
 
 //Pass the test sample through the pipeline
 if( !map( inputVector ) ){
 errorLog << "test(const RegressionData &testData) - Failed to map input Vector!" << std::endl;
 return false;
 }
 
 //Update the RMS error
 Float sum = 0;
 VectorFloat regressionData = regressifier->getRegressionData();
 for(UINT j=0; j<targetVector.size(); j++){
 sum += grt_sqr( regressionData[j]-targetVector[j] );
 }

 testSquaredError += sum;

 //Keep track of the regression results encase the user needs them later
 testResults[i].setRegressionResult(i,regressionData,targetVector);

 //Update any observers
 regressifier->notifyTestResultsObservers( testResults[i] );
 }
 
 //Compute the test metrics
 testRMSError = sqrt( testSquaredError / Float( testData.getNumSamples() ) );
 
 testTime = timer.getMilliSeconds();
 
 return true;
}

bool GestureRecognitionPipeline::predict(const VectorFloat &inputVector){
 
 //Make sure the classification model has been trained
 if( !trained ){
 errorLog << "predict(const VectorFloat &inputVector) - The classifier has not been trained" << std::endl;
 return false;
 }
 
 //Make sure the dimensionality of the input Vector matches the inputVectorDimensions
 if( inputVector.size() != inputVectorDimensions ){
 errorLog << "predict(const VectorFloat &inputVector) - The dimensionality of the input Vector (" << int(inputVector.size()) << ") does not match that of the input Vector dimensions of the pipeline (" << inputVectorDimensions << ")" << std::endl;
 return false;
 }

 if( getIsClassifierSet() ){
 return predict_classifier( inputVector );
 }

 if( getIsRegressifierSet() ){
 return predict_regressifier( inputVector );
 }
 
 if( getIsClustererSet() ){
 return predict_clusterer( inputVector );
 }

 errorLog << "predict(const VectorFloat &inputVector) - Neither a classifier, regressifer or clusterer is set" << std::endl;
 return false;
}

bool GestureRecognitionPipeline::predict(const MatrixFloat &input){
 
 //Make sure the classification model has been trained
 if( !trained ){
 errorLog << "predict(const MatrixFloat &inputMatrix) - The classifier has not been trained!" << std::endl;
 return false;
 }

 //Make sure the dimensionality of the input matrix matches the inputVectorDimensions
 if( input.getNumCols() != inputVectorDimensions ){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - The dimensionality of the input matrix (" << input.getNumCols() << ") does not match that of the input Vector dimensions of the pipeline (" << inputVectorDimensions << ")" << std::endl;
 return false;
 }

 if( !getIsClassifierSet() ){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - A classifier has not been set" << std::endl;
 return false;
 }

 //Get a copy of the input matrix so it can be processed
 MatrixFloat inputMatrix = input;

 //Get a pointer to the input matrix so we can pass it down the pipeline
 const void *data = static_cast< const void* >( &inputMatrix );
 DataType dataType = DATA_TYPE_MATRIX;

 //Setup a temporary matrix and vector
 //VectorFloat tmpVector;
 //MatrixFloat tmpMatrix;
 
 //Update the context module
 predictionModuleIndex = START_OF_PIPELINE;
 
 //Perform any pre-processing
 /*
 if( getIsPreProcessingSet() ){
 
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 MatrixFloat tmpMatrix( inputMatrix.getNumRows(), preProcessingModules[moduleIndex]->getNumOutputDimensions() );
 
 for(UINT i=0; i<inputMatrix.getNumRows(); i++){
 if( !preProcessingModules[moduleIndex]->process( inputMatrix.getRow(i) ) ){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - Failed to PreProcess Input Matrix. PreProcessingModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 tmpMatrix.setRowVector( preProcessingModules[moduleIndex]->getProcessedData(), i );
 }
 
 //Update the input matrix with the preprocessed data
 inputMatrix = tmpMatrix;
 }
 }
 */
 
 //Update the context module
 predictionModuleIndex = AFTER_PREPROCESSING;
 //Perform any feature extraction
 if( getIsFeatureExtractionSet() ){

 const void *feInput = data;
 const void *feOutput = NULL;
 const UINT numFeatureExtractionModules = featureExtractionModules.getSize();
 DataType inputType = DATA_TYPE_UNKNOWN;
 DataType outputType = DATA_TYPE_UNKNOWN;
 for(UINT moduleIndex=0; moduleIndex<numFeatureExtractionModules; moduleIndex++){

 inputType = featureExtractionModules[ moduleIndex ]->getInputType();
 outputType = featureExtractionModules[ moduleIndex ]->getOutputType();

 //Run the feature extraction algorithm
 switch( inputType ){
 case DATA_TYPE_VECTOR:
 if( !featureExtractionModules[ moduleIndex ]->computeFeatures( *static_cast< const VectorFloat* >( feInput ) ) ){
 errorLog << "predict(const MatrixFloat &inputMatrix) - Failed to PreProcess Input Matrix. FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 break;
 case DATA_TYPE_MATRIX:
 if( !featureExtractionModules[ moduleIndex ]->computeFeatures( *static_cast< const MatrixFloat* >( feInput ) ) ){
 errorLog << "predict(const MatrixFloat &inputMatrix) - Failed to PreProcess Input Matrix. FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 break;
 default:
 errorLog << "predict(const MatrixFloat &inputMatrix) - Failed to process data. Unknown output data type for FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 break;
 }
 
 //Get the results and store them in the feOutput pointer
 switch( outputType ){
 case DATA_TYPE_VECTOR:
 feOutput = static_cast< const void* >( &featureExtractionModules[ moduleIndex ]->getFeatureVector() );
 break;
 case DATA_TYPE_MATRIX:
 feOutput = static_cast< const void* >( &featureExtractionModules[ moduleIndex ]->getFeatureMatrix() );
 break;
 default:
 errorLog << "predict(const MatrixFloat &inputMatrix) - Failed to process data. Unknown output data type for FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 break;
 }

 //If there are more feature extraction modules to process, then set the feInput 
 if( moduleIndex+1 < numFeatureExtractionModules ){
 feInput = feOutput;
 }
 }

 //Update the data pointer to the data from the last output from the end feature extraction module
 data = feOutput;
 dataType = outputType;
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;

 //Perform the classification
 switch( dataType ){
 case DATA_TYPE_VECTOR:
 if( !classifier->predict_( *(VectorFloat*)data ) ){
 errorLog <<"predict_timeseries(const MatrixFloat &inputMatrix) - Prediction Failed! " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 break;
 case DATA_TYPE_MATRIX:
 if( !classifier->predict_( *(MatrixFloat*)data ) ){
 errorLog <<"predict_timeseries(const MatrixFloat &inputMatrix) - Prediction Failed! " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 break;
 default:
 errorLog << "predict(const MatrixFloat &inputMatrix) - Failed to run prediction. Unknown data type!" << std::endl;
 return false;
 break;
 }
 
 predictedClassLabel = classifier->getPredictedClassLabel();
 
 //Update the context module
 //Todo
 
 //Perform any post processing
 predictionModuleIndex = AFTER_CLASSIFIER;
/*
 if( getIsPostProcessingSet() ){
 
 if( pipelineMode != CLASSIFICATION_MODE){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - Pipeline Mode Is Not in CLASSIFICATION_MODE!" << std::endl;
 return false;
 }
 
 VectorFloat data;
 for(UINT moduleIndex=0; moduleIndex<postProcessingModules.size(); moduleIndex++){
 
 //Select which input we should give the postprocessing module
 if( postProcessingModules[moduleIndex]->getIsPostProcessingInputModePredictedClassLabel() ){
 //Set the input
 data.resize(1);
 data[0] = predictedClassLabel;
 
 //Verify that the input size is OK
 if( data.size() != postProcessingModules[moduleIndex]->getNumInputDimensions() ){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - The size of the data Vector (" << int(data.size()) << ") does not match that of the postProcessingModule (" << postProcessingModules[moduleIndex]->getNumInputDimensions() << ") at the moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 //Postprocess the data
 if( !postProcessingModules[moduleIndex]->process( data ) ){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - Failed to post process data. PostProcessing moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 //Select which output we should update
 data = postProcessingModules[moduleIndex]->getProcessedData(); 
 }
 
 //Select which output we should update
 if( postProcessingModules[moduleIndex]->getIsPostProcessingOutputModePredictedClassLabel() ){
 //Get the processed predicted class label
 data = postProcessingModules[moduleIndex]->getProcessedData(); 
 
 //Verify that the output size is OK
 if( data.size() != 1 ){
 errorLog << "predict_timeseries(const MatrixFloat &inputMatrix) - The size of the processed data Vector (" << int(data.size()) << ") from postProcessingModule at the moduleIndex: " << moduleIndex << " is not equal to 1 even though it is in OutputModePredictedClassLabel!" << std::endl;
 return false;
 }
 
 //Update the predicted class label
 predictedClassLabel = (UINT)data[0];
 }
 
 }
 } 
*/
 //Update the context module
 //TODO
 predictionModuleIndex = END_OF_PIPELINE;
 return true;
}

bool GestureRecognitionPipeline::map(const VectorFloat &inputVector){
 return predict_regressifier( inputVector );
}

bool GestureRecognitionPipeline::predict_classifier(const VectorFloat &input){
 
 predictedClassLabel = 0;
 VectorFloat inputVector = input;
 
 //Update the context module
 predictionModuleIndex = START_OF_PIPELINE;
 if( contextModules[ START_OF_PIPELINE ].size() > 0 ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ START_OF_PIPELINE ].size(); moduleIndex++){
 if( !contextModules[ START_OF_PIPELINE ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_classifier(const VectorFloat &inputVector) - Context Module Failed at START_OF_PIPELINE. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ START_OF_PIPELINE ][moduleIndex]->getOK() ){
 return true;
 }
 inputVector = contextModules[ START_OF_PIPELINE ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any pre-processing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[moduleIndex]->process( inputVector ) ){
 errorLog << "predict_classifier(const VectorFloat &inputVector) - Failed to PreProcess Input Vector. PreProcessingModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = preProcessingModules[moduleIndex]->getProcessedData();
 }
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_PREPROCESSING;
 if( contextModules[ AFTER_PREPROCESSING ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_PREPROCESSING ].size(); moduleIndex++){
 if( !contextModules[ AFTER_PREPROCESSING ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Context Module Failed at AFTER_PREPROCESSING. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_PREPROCESSING ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_PREPROCESSING;
 return false;
 }
 inputVector = contextModules[ AFTER_PREPROCESSING ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any feature extraction
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[moduleIndex]->computeFeatures( inputVector ) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Failed to compute features from data. FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = featureExtractionModules[moduleIndex]->getFeatureVector();
 }
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;
 if( contextModules[ AFTER_FEATURE_EXTRACTION ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_FEATURE_EXTRACTION ].size(); moduleIndex++){
 if( !contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Context Module Failed at AFTER_FEATURE_EXTRACTION. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;
 return false;
 }
 inputVector = contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform the classification
 if( !classifier->predict(inputVector) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Prediction Failed! " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 predictedClassLabel = classifier->getPredictedClassLabel();
 
 //Update the context module
 if( contextModules[ AFTER_CLASSIFIER ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_CLASSIFIER ].size(); moduleIndex++){
 if( !contextModules[ AFTER_CLASSIFIER ][moduleIndex]->process( VectorFloat(1,predictedClassLabel) ) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Context Module Failed at AFTER_CLASSIFIER. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_CLASSIFIER ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_CLASSIFIER;
 return false;
 }
 predictedClassLabel = (UINT)contextModules[ AFTER_CLASSIFIER ][moduleIndex]->getProcessedData()[0];
 }
 }
 
 //Perform any post processing
 predictionModuleIndex = AFTER_CLASSIFIER;
 if( getIsPostProcessingSet() ){
 
 if( pipelineMode != CLASSIFICATION_MODE){
 errorLog << "predict_classifier(VectorFloat inputVector) - Pipeline Mode Is Not in CLASSIFICATION_MODE!" << std::endl;
 return false;
 }
 
 VectorFloat data;
 for(UINT moduleIndex=0; moduleIndex<postProcessingModules.size(); moduleIndex++){
 
 //Select which input we should give the postprocessing module
 if( postProcessingModules[moduleIndex]->getIsPostProcessingInputModePredictedClassLabel() ){
 //Set the input
 data.resize(1);
 data[0] = predictedClassLabel;
 
 //Verify that the input size is OK
 if( data.size() != postProcessingModules[moduleIndex]->getNumInputDimensions() ){
 errorLog << "predict_classifier(VectorFloat inputVector) - The size of the data Vector (" << int(data.size()) << ") does not match that of the postProcessingModule (" << postProcessingModules[moduleIndex]->getNumInputDimensions() << ") at the moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 //Postprocess the data
 if( !postProcessingModules[moduleIndex]->process( data ) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Failed to post process data. PostProcessing moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 //Select which output we should update
 data = postProcessingModules[moduleIndex]->getProcessedData(); 
 }
 
 //Select which output we should update
 if( postProcessingModules[moduleIndex]->getIsPostProcessingOutputModePredictedClassLabel() ){
 //Get the processed predicted class label
 data = postProcessingModules[moduleIndex]->getProcessedData(); 
 
 //Verify that the output size is OK
 if( data.size() != 1 ){
 errorLog << "predict_classifier(VectorFloat inputVector) - The size of the processed data Vector (" << int(data.size()) << ") from postProcessingModule at the moduleIndex: " << moduleIndex << " is not equal to 1 even though it is in OutputModePredictedClassLabel!" << std::endl;
 return false;
 }
 
 //Update the predicted class label
 predictedClassLabel = (UINT)data[0];
 }
 
 }
 } 
 
 //Update the context module
 predictionModuleIndex = END_OF_PIPELINE;
 if( contextModules[ END_OF_PIPELINE ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ END_OF_PIPELINE ].size(); moduleIndex++){
 if( !contextModules[ END_OF_PIPELINE ][moduleIndex]->process( VectorFloat(1,predictedClassLabel) ) ){
 errorLog << "predict_classifier(VectorFloat inputVector) - Context Module Failed at END_OF_PIPELINE. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ END_OF_PIPELINE ][moduleIndex]->getOK() ){
 predictionModuleIndex = END_OF_PIPELINE;
 return false;
 }
 predictedClassLabel = (UINT)contextModules[ END_OF_PIPELINE ][moduleIndex]->getProcessedData()[0];
 }
 }
 
 return true;
}
 
bool GestureRecognitionPipeline::predict_regressifier(const VectorFloat &input){
 
 VectorFloat inputVector = input;

 //Update the context module
 predictionModuleIndex = START_OF_PIPELINE;
 if( contextModules[ START_OF_PIPELINE ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ START_OF_PIPELINE ].size(); moduleIndex++){
 if( !contextModules[ START_OF_PIPELINE ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Context Module Failed at START_OF_PIPELINE. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ START_OF_PIPELINE ][moduleIndex]->getOK() ){
 return true;
 }
 inputVector = contextModules[ START_OF_PIPELINE ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any pre-processing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[moduleIndex]->process( inputVector ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Failed to PreProcess Input Vector. PreProcessingModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = preProcessingModules[moduleIndex]->getProcessedData();
 }
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_PREPROCESSING;
 if( contextModules[ AFTER_PREPROCESSING ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_PREPROCESSING ].size(); moduleIndex++){
 if( !contextModules[ AFTER_PREPROCESSING ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Context Module Failed at AFTER_PREPROCESSING. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_PREPROCESSING ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_PREPROCESSING;
 return false;
 }
 inputVector = contextModules[ AFTER_PREPROCESSING ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any feature extraction
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[moduleIndex]->computeFeatures( inputVector ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Failed to compute features from data. FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = featureExtractionModules[moduleIndex]->getFeatureVector();
 }
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;
 if( contextModules[ AFTER_FEATURE_EXTRACTION ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_FEATURE_EXTRACTION ].size(); moduleIndex++){
 if( !contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Context Module Failed at AFTER_FEATURE_EXTRACTION. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;
 return false;
 }
 inputVector = contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform the regression
 if( !regressifier->predict(inputVector) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Prediction Failed! " << regressifier->getLastErrorMessage() << std::endl;
 return false;
 }
 regressionData = regressifier->getRegressionData();
 
 //Update the context module
 if( contextModules[ AFTER_CLASSIFIER ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_CLASSIFIER ].size(); moduleIndex++){
 if( !contextModules[ AFTER_CLASSIFIER ][moduleIndex]->process( regressionData ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Context Module Failed at AFTER_CLASSIFIER. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_CLASSIFIER ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_CLASSIFIER;
 return false;
 }
 regressionData = contextModules[ AFTER_CLASSIFIER ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any post processing
 predictionModuleIndex = AFTER_CLASSIFIER;
 if( getIsPostProcessingSet() ){
 
 if( pipelineMode != REGRESSION_MODE ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Pipeline Mode Is Not In RegressionMode!" << std::endl;
 return false;
 }
 
 for(UINT moduleIndex=0; moduleIndex<postProcessingModules.size(); moduleIndex++){
 if( regressionData.size() != postProcessingModules[moduleIndex]->getNumInputDimensions() ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - The size of the regression Vector (" << int(regressionData.size()) << ") does not match that of the postProcessingModule (" << postProcessingModules[moduleIndex]->getNumInputDimensions() << ") at the moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 if( !postProcessingModules[moduleIndex]->process( regressionData ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Failed to post process data. PostProcessing moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 regressionData = postProcessingModules[moduleIndex]->getProcessedData(); 
 }
 
 } 
 
 //Update the context module
 predictionModuleIndex = END_OF_PIPELINE;
 if( contextModules[ END_OF_PIPELINE ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ END_OF_PIPELINE ].size(); moduleIndex++){
 if( !contextModules[ END_OF_PIPELINE ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_regressifier(VectorFloat inputVector) - Context Module Failed at END_OF_PIPELINE. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ END_OF_PIPELINE ][moduleIndex]->getOK() ){
 predictionModuleIndex = END_OF_PIPELINE;
 return false;
 }
 regressionData = contextModules[ END_OF_PIPELINE ][moduleIndex]->getProcessedData();
 }
 }
 
 return true;
}
 
bool GestureRecognitionPipeline::predict_clusterer(const VectorFloat &input){
 
 VectorFloat inputVector = input;
 predictedClusterLabel = 0;
 
 //Update the context module
 predictionModuleIndex = START_OF_PIPELINE;
 if( contextModules[ START_OF_PIPELINE ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ START_OF_PIPELINE ].size(); moduleIndex++){
 if( !contextModules[ START_OF_PIPELINE ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Context Module Failed at START_OF_PIPELINE. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ START_OF_PIPELINE ][moduleIndex]->getOK() ){
 return true;
 }
 inputVector = contextModules[ START_OF_PIPELINE ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any pre-processing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[moduleIndex]->process( inputVector ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Failed to PreProcess Input Vector. PreProcessingModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = preProcessingModules[moduleIndex]->getProcessedData();
 }
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_PREPROCESSING;
 if( contextModules[ AFTER_PREPROCESSING ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_PREPROCESSING ].size(); moduleIndex++){
 if( !contextModules[ AFTER_PREPROCESSING ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Context Module Failed at AFTER_PREPROCESSING. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_PREPROCESSING ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_PREPROCESSING;
 return false;
 }
 inputVector = contextModules[ AFTER_PREPROCESSING ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform any feature extraction
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[moduleIndex]->computeFeatures( inputVector ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Failed to compute features from data. FeatureExtractionModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = featureExtractionModules[moduleIndex]->getFeatureVector();
 }
 }
 
 //Update the context module
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;
 if( contextModules[ AFTER_FEATURE_EXTRACTION ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_FEATURE_EXTRACTION ].size(); moduleIndex++){
 if( !contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->process( inputVector ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Context Module Failed at AFTER_FEATURE_EXTRACTION. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_FEATURE_EXTRACTION;
 return false;
 }
 inputVector = contextModules[ AFTER_FEATURE_EXTRACTION ][moduleIndex]->getProcessedData();
 }
 }
 
 //Perform the classification
 if( !clusterer->predict(inputVector) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Prediction Failed! " << clusterer->getLastErrorMessage() << std::endl;
 return false;
 }
 predictedClusterLabel = clusterer->getPredictedClusterLabel();
 
 //Update the context module
 if( contextModules[ AFTER_CLASSIFIER ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ AFTER_CLASSIFIER ].size(); moduleIndex++){
 if( !contextModules[ AFTER_CLASSIFIER ][moduleIndex]->process( VectorFloat(1,predictedClusterLabel) ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Context Module Failed at AFTER_CLASSIFIER. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ AFTER_CLASSIFIER ][moduleIndex]->getOK() ){
 predictionModuleIndex = AFTER_CLASSIFIER;
 return false;
 }
 predictedClusterLabel = (UINT)contextModules[ AFTER_CLASSIFIER ][moduleIndex]->getProcessedData()[0];
 }
 }
 
 //Perform any post processing
 predictionModuleIndex = AFTER_CLASSIFIER;
 if( getIsPostProcessingSet() ){
 
 if( pipelineMode != CLASSIFICATION_MODE){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Pipeline Mode Is Not in CLASSIFICATION_MODE!" << std::endl;
 return false;
 }
 
 VectorFloat data;
 for(UINT moduleIndex=0; moduleIndex<postProcessingModules.size(); moduleIndex++){
 
 //Select which input we should give the postprocessing module
 if( postProcessingModules[moduleIndex]->getIsPostProcessingInputModePredictedClassLabel() ){
 //Set the input
 data.resize(1);
 data[0] = predictedClusterLabel;
 
 //Verify that the input size is OK
 if( data.size() != postProcessingModules[moduleIndex]->getNumInputDimensions() ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - The size of the data Vector (" << int(data.size()) << ") does not match that of the postProcessingModule (" << postProcessingModules[moduleIndex]->getNumInputDimensions() << ") at the moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 //Postprocess the data
 if( !postProcessingModules[moduleIndex]->process( data ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Failed to post process data. PostProcessing moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 //Select which output we should update
 data = postProcessingModules[moduleIndex]->getProcessedData();
 }
 
 //Select which output we should update
 if( postProcessingModules[moduleIndex]->getIsPostProcessingOutputModePredictedClassLabel() ){
 //Get the processed predicted class label
 data = postProcessingModules[moduleIndex]->getProcessedData();
 
 //Verify that the output size is OK
 if( data.size() != 1 ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - The size of the processed data Vector (" << int(data.size()) << ") from postProcessingModule at the moduleIndex: " << moduleIndex << " is not equal to 1 even though it is in OutputModePredictedClassLabel!" << std::endl;
 return false;
 }
 
 //Update the predicted cluster label
 predictedClusterLabel = (UINT)data[0];
 }
 
 }
 }
 
 //Update the context module
 predictionModuleIndex = END_OF_PIPELINE;
 if( contextModules[ END_OF_PIPELINE ].size() ){
 for(UINT moduleIndex=0; moduleIndex<contextModules[ END_OF_PIPELINE ].size(); moduleIndex++){
 if( !contextModules[ END_OF_PIPELINE ][moduleIndex]->process( VectorFloat(1,predictedClassLabel) ) ){
 errorLog << "predict_clusterer(VectorFloat inputVector) - Context Module Failed at END_OF_PIPELINE. ModuleIndex: " << moduleIndex << std::endl;
 return false;
 }
 if( !contextModules[ END_OF_PIPELINE ][moduleIndex]->getOK() ){
 predictionModuleIndex = END_OF_PIPELINE;
 return false;
 }
 predictedClusterLabel = (UINT)contextModules[ END_OF_PIPELINE ][moduleIndex]->getProcessedData()[0];
 }
 }
 
 return true;
}
 
bool GestureRecognitionPipeline::reset(){
 
 //Reset any pre processing
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 if( !preProcessingModules[ moduleIndex ]->reset() ){
 errorLog << "Failed To Reset PreProcessingModule " << moduleIndex << std::endl;
 return false;
 }
 }
 }
 
 //Reset any feature extraction
 if( getIsFeatureExtractionSet() ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( !featureExtractionModules[ moduleIndex ]->reset() ){
 errorLog << "Failed To Reset FeatureExtractionModule " << moduleIndex << std::endl;
 return false;
 }
 }
 }
 
 //Reset the classifier
 if( getIsClassifierSet() ){
 if( !classifier->reset() ){
 errorLog << "Failed To Reset Classifier! " << classifier->getLastErrorMessage() << std::endl;
 return false;
 }
 }
 
 //Reset the regressiier
 if( getIsRegressifierSet() ){
 if( !regressifier->reset() ){
 errorLog << "Failed To Reset Regressifier! " << regressifier->getLastErrorMessage() << std::endl;
 return false;
 }
 }
 
 //Reset the clusterer
 if( getIsClustererSet() ){
 if( !clusterer->reset() ){
 errorLog << "Failed To Reset clusterer! " << clusterer->getLastErrorMessage() << std::endl;
 return false;
 }
 }
 
 //Reset any post processing
 if( getIsPostProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<postProcessingModules.size(); moduleIndex++){
 if( !postProcessingModules[ moduleIndex ]->reset() ){
 errorLog << "Failed To Reset PostProcessingModule " << moduleIndex << std::endl;
 return false;
 }
 }
 } 
 
 return true;
}

bool GestureRecognitionPipeline::clearAll(){ return clear(); }

bool GestureRecognitionPipeline::clear(){

 //Clear the entire pipeline
 clearTestResults();
 deleteAllPreProcessingModules();
 deleteAllFeatureExtractionModules();
 deleteClassifier();
 deleteRegressifier();
 deleteClusterer();
 deleteAllPostProcessingModules();
 deleteAllContextModules();

 //Reset the pipeline
 return init();
}

bool GestureRecognitionPipeline::clearModel(){
 
 //Clear any preprocessing module
 for(UINT i=0; i<getNumPreProcessingModules(); i++){
 preProcessingModules[i]->clear();
 }
 
 //Clear any feature extraction module
 for(UINT i=0; i<getNumFeatureExtractionModules(); i++){
 featureExtractionModules[i]->clear();
 }
 
 //Clear any ML model
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 if( getIsClassifierSet() ){
 classifier->clear();
 }
 break;
 case REGRESSION_MODE:
 if( getIsRegressifierSet() ){
 regressifier->clear();
 }
 break;
 case CLUSTER_MODE:
 if( getIsClustererSet() ){
 clusterer->clear();
 }
 break;
 default:
 break;
 }
 
 //Clear any post processing modules
 for(UINT i=0; i<getNumPostProcessingModules(); i++){
 postProcessingModules[i]->clear();
 }

 return true;
}
 
bool GestureRecognitionPipeline::save(const std::string &filename) const {
 return savePipelineToFile( filename );
}
 
bool GestureRecognitionPipeline::savePipelineToFile(const std::string &filename) const {
 
 if( !initialized ){
 errorLog << "Failed to write pipeline to file as the pipeline has not been initialized yet!" << std::endl;
 return false;
 }
 
 std::fstream file;
 
 file.open(filename.c_str(), std::iostream::out );
 
 if( !file.is_open() ){
 errorLog << "Failed to open file with filename: " << filename << std::endl;
 return false;
 }
 
 //Write the pipeline header info
 file << "GRT_PIPELINE_FILE_V3.0\n";
 file << "PipelineMode: " << getPipelineModeAsString() << std::endl;
 file << "NumPreprocessingModules: " << getNumPreProcessingModules() << std::endl;
 file << "NumFeatureExtractionModules: " << getNumFeatureExtractionModules() << std::endl;
 file << "NumPostprocessingModules: " << getNumPostProcessingModules() << std::endl;
 file << "Trained: " << getTrained() << std::endl;
 file << "Info: " << info << std::endl;
 
 //Write the module datatype names
 file << "PreProcessingModuleDatatypes:";
 for(UINT i=0; i<getNumPreProcessingModules(); i++){
 file << "\t" << preProcessingModules[i]->getPreProcessingType();
 }
 file << std::endl;
 
 file << "FeatureExtractionModuleDatatypes:";
 for(UINT i=0; i<getNumFeatureExtractionModules(); i++){
 file << "\t" << featureExtractionModules[i]->getFeatureExtractionType();
 }
 file << std::endl;
 
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 if( getIsClassifierSet() ) file << "ClassificationModuleDatatype:\t" << classifier->getClassifierType() << std::endl;
 else file << "ClassificationModuleDatatype:\tCLASSIFIER_NOT_SET" << std::endl;
 break;
 case REGRESSION_MODE:
 if( getIsRegressifierSet() ) file << "RegressionModuleDatatype:\t" << regressifier->getRegressifierType() << std::endl;
 else file << "RegressionModuleDatatype:\tREGRESSIFIER_NOT_SET" << std::endl;
 break;
 case CLUSTER_MODE:
 if( getIsClustererSet() ) file << "ClusterModuleDatatype:\t" << clusterer->getClustererType() << std::endl;
 else file << "ClusterModuleDatatype:\tCLUSTERER_NOT_SET" << std::endl;
 break;
 default:
 break;
 }
 
 file << "PostProcessingModuleDatatypes:";
 for(UINT i=0; i<getNumPostProcessingModules(); i++){
 file << "\t" << postProcessingModules[i]->getPostProcessingType();
 }
 file << std::endl;
 
 //Write the preprocessing module data to the file
 for(UINT i=0; i<getNumPreProcessingModules(); i++){
 file << "PreProcessingModule_" << Util::intToString(i+1) << std::endl;
 if( !preProcessingModules[i]->save( file ) ){
 errorLog << "Failed to write preprocessing module " << i << " settings to file!" << std::endl;
 file.close();
 return false;
 }
 }
 
 //Write the feature extraction module data to the file
 for(UINT i=0; i<getNumFeatureExtractionModules(); i++){
 file << "FeatureExtractionModule_" << Util::intToString(i+1) << std::endl;
 if( !featureExtractionModules[i]->save( file ) ){
 errorLog << "Failed to write feature extraction module " << i << " settings to file!" << std::endl;
 file.close();
 return false;
 }
 }
 
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 if( getIsClassifierSet() ){
 if( !classifier->save( file ) ){
 errorLog << "Failed to write classifier model to file!" << std::endl;
 file.close();
 return false;
 }
 }
 break;
 case REGRESSION_MODE:
 if( getIsRegressifierSet() ){
 if( !regressifier->save( file ) ){
 errorLog << "Failed to write regressifier model to file!" << std::endl;
 file.close();
 return false;
 }
 }
 break;
 case CLUSTER_MODE:
 if( getIsClustererSet() ){
 if( !clusterer->save( file ) ){
 errorLog << "Failed to write clusterer model to file!" << std::endl;
 file.close();
 return false;
 }
 }
 break;
 default:
 break;
 }
 
 //Write the post processing module data to the file
 for(UINT i=0; i<getNumPostProcessingModules(); i++){
 file << "PostProcessingModule_" << Util::intToString(i+1) << std::endl;
 if( !postProcessingModules[i]->save( file ) ){
 errorLog <<"Failed to write post processing module " << i << " settings to file!" << std::endl;
 file.close();
 return false;
 }
 }
 
 //Close the file
 file.close();
 
 return true;
}

bool GestureRecognitionPipeline::load(const std::string &filename){
 return loadPipelineFromFile( filename );
}
 
bool GestureRecognitionPipeline::loadPipelineFromFile(const std::string &filename){
 
 std::fstream file;

 //Clear any previous setup
 clear();
 
 file.open(filename.c_str(), std::iostream::in );
 
 if( !file.is_open() ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to open file with filename: " << filename << std::endl;
 return false;
 }

 std::string word;
 
 //Load the file header
 file >> word;
 if( word != "GRT_PIPELINE_FILE_V3.0" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read file header" << std::endl;
 file.close();
 return false;
 }
 
 //Load the pipeline mode
 file >> word;
 if( word != "PipelineMode:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read PipelineMode" << std::endl;
 file.close();
 return false;
 }
 file >> word;
 pipelineMode = getPipelineModeFromString(word);
 
 //Load the NumPreprocessingModules
 file >> word;
 if( word != "NumPreprocessingModules:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read NumPreprocessingModules header" << std::endl;
 file.close();
 return false;
 }
 unsigned int numPreprocessingModules;
 file >> numPreprocessingModules;
 
 //Load the NumFeatureExtractionModules
 file >> word;
 if( word != "NumFeatureExtractionModules:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read NumFeatureExtractionModules header" << std::endl;
 file.close();
 return false;
 }
 unsigned int numFeatureExtractionModules;
 file >> numFeatureExtractionModules;
 
 //Load the NumPostprocessingModules
 file >> word;
 if( word != "NumPostprocessingModules:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read NumPostprocessingModules header" << std::endl;
 file.close();
 return false;
 }
 unsigned int numPostprocessingModules;
 file >> numPostprocessingModules;
 
 //Load if the pipeline has been trained
 file >> word;
 if( word != "Trained:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read Trained header" << std::endl;
 file.close();
 return false;
 }
 file >> trained;
 
 //Load the info
 file >> word;
 if( word != "Info:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read Info header" << std::endl;
 file.close();
 return false;
 }
 info = "";
 //Read the info text
 file >> word;
 while( word != "PreProcessingModuleDatatypes:" ){
 info += word;
 file >> word;
 }
 
 //Resize the modules
 if( numPreprocessingModules > 0 ) preProcessingModules.resize(numPreprocessingModules,NULL);
 if( numFeatureExtractionModules > 0 ) featureExtractionModules.resize(numFeatureExtractionModules,NULL);
 if( numPostprocessingModules > 0 ) postProcessingModules.resize(numPostprocessingModules,NULL);
 
 //Load the preprocessing module datatypes and initialize the modules
 if( word != "PreProcessingModuleDatatypes:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read PreProcessingModuleDatatypes" << std::endl;
 file.close();
 return false;
 }
 for(UINT i=0; i<numPreprocessingModules; i++){
 file >> word;
 preProcessingModules[i] = PreProcessing::createInstanceFromString( word );
 if( preProcessingModules[i] == NULL ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to create preprocessing instance from string: " << word << std::endl;
 file.close();
 return false;
 }
 }

 //Load the feature extraction module datatypes and initialize the modules
 file >> word;
 if( word != "FeatureExtractionModuleDatatypes:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read FeatureExtractionModuleDatatypes" << std::endl;
 file.close();
 return false;
 }
 for(UINT i=0; i<numFeatureExtractionModules; i++){
 file >> word;
 featureExtractionModules[i] = FeatureExtraction::createInstanceFromString( word );
 if( featureExtractionModules[i] == NULL ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to create feature extraction instance from string: " << word << std::endl;
 file.close();
 return false;
 }
 }

 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 file >> word;
 if( word != "ClassificationModuleDatatype:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read ClassificationModuleDatatype" << std::endl;
 file.close();
 return false;
 }
 //Load the classifier type
 file >> word;
 
 //Initialize the classifier
 classifier = Classifier::createInstanceFromString( word );
 if( classifier == NULL ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to create classifier instance from string: " << word << std::endl;
 file.close();
 return false;
 }
 break;
 case REGRESSION_MODE:
 file >> word;
 if( word != "RegressionModuleDatatype:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read RegressionModuleDatatype" << std::endl;
 file.close();
 return false;
 }
 //Load the regressifier type
 file >> word;
 
 //Initialize the regressifier
 regressifier = Regressifier::createInstanceFromString( word );
 if( regressifier == NULL ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to create regressifier instance from string: " << word << std::endl;
 file.close();
 return false;
 }
 break;
 case CLUSTER_MODE:
 file >> word;
 if( word != "ClusterModuleDatatype:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read ClusterModuleDatatype" << std::endl;
 file.close();
 return false;
 }
 //Load the clusterer type
 file >> word;
 
 //Initialize the clusterer
 clusterer = Clusterer::createInstanceFromString( word );
 if( clusterer == NULL ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to create clusterer instance from string: " << word << std::endl;
 file.close();
 return false;
 }
 break;
 default:
 break;
 }

 //Load the post processing module datatypes and initialize the modules
 file >> word;
 if( word != "PostProcessingModuleDatatypes:" ){
 errorLog << "loadPipelineFromFile(string filename) - Failed to read PostProcessingModuleDatatypes" << std::endl;
 file.close();
 return false;
 }
 for(UINT i=0; i<numPostprocessingModules; i++){
 file >> word;
 postProcessingModules[i] = PostProcessing::createInstanceFromString( word );
 }
 
 //Load the preprocessing module data from the file
 for(UINT i=0; i<numPreprocessingModules; i++){
 //Load the preprocessing module header
 file >> word;
 if( !preProcessingModules[i]->load( file ) ){
 errorLog << "Failed to load preprocessing module " << i << " settings from file!" << std::endl;
 file.close();
 return false;
 }
 }

 //Load the feature extraction module data from the file
 for(UINT i=0; i<numFeatureExtractionModules; i++){
 //Load the feature extraction module header
 file >> word;
 if( !featureExtractionModules[i]->load( file ) ){
 errorLog << "Failed to load feature extraction module " << i << " settings from file!" << std::endl;
 file.close();
 return false;
 }
 }
 
 //Load the classifier or regressifer data
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 if( !classifier->load( file ) ){
 errorLog << "Failed to load classifier model from file!" << std::endl;
 file.close();
 return false;
 }
 break;
 case REGRESSION_MODE:
 if( !regressifier->load( file ) ){
 errorLog << "Failed to load regressifier model from file!" << std::endl;
 file.close();
 return false;
 }
 break;
 case CLUSTER_MODE:
 if( !clusterer->load( file ) ){
 errorLog << "Failed to load cluster model from file!" << std::endl;
 file.close();
 return false;
 }
 break;
 default:
 break;
 }

 //Load the post processing module data from the file
 for(UINT i=0; i<numPostprocessingModules; i++){
 //Load the post processing module header
 file >> word;
 if( !postProcessingModules[i]->load( file ) ){
 errorLog << "Failed to load post processing module " << i << " settings from file!" << std::endl;
 file.close();
 return false;
 }
 }
 
 //Close the file
 file.close();
 
 //Set the expected input Vector size
 inputVectorDimensions = 0;
 
 if( numPreprocessingModules > 0 ){
 inputVectorDimensions = preProcessingModules[0]->getNumInputDimensions();
 }else{
 if( numFeatureExtractionModules > 0 ){
 inputVectorDimensions = featureExtractionModules[0]->getNumInputDimensions();
 }else{
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 inputVectorDimensions = classifier->getNumInputDimensions();
 break;
 case REGRESSION_MODE:
 inputVectorDimensions = regressifier->getNumInputDimensions();
 break;
 case CLUSTER_MODE:
 inputVectorDimensions = clusterer->getNumInputDimensions();
 break;
 default:
 break;
 }
 }
 }

 //Flag that the pipeline is now initialized
 initialized = true;
 
 return true;
}
 
bool GestureRecognitionPipeline::preProcessData(VectorFloat inputVector,bool computeFeatures){
 
 if( getIsPreProcessingSet() ){
 for(UINT moduleIndex=0; moduleIndex<preProcessingModules.size(); moduleIndex++){
 
 if( inputVector.size() != preProcessingModules[ moduleIndex ]->getNumInputDimensions() ){
 errorLog << "preProcessData(VectorFloat inputVector) - The size of the input Vector (" << preProcessingModules[ moduleIndex ]->getNumInputDimensions() << ") does not match that of the PreProcessing Module at moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 if( !preProcessingModules[ moduleIndex ]->process( inputVector ) ){
 errorLog << "preProcessData(VectorFloat inputVector) - Failed To PreProcess Input Vector. PreProcessing moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = preProcessingModules[ moduleIndex ]->getProcessedData();
 }
 }
 
 //Perform any feature extraction
 if( getIsFeatureExtractionSet() && computeFeatures ){
 for(UINT moduleIndex=0; moduleIndex<featureExtractionModules.size(); moduleIndex++){
 if( inputVector.size() != featureExtractionModules[ moduleIndex ]->getNumInputDimensions() ){
 errorLog << "FeatureExtraction(VectorFloat inputVector) - The size of the input Vector (" << featureExtractionModules[ moduleIndex ]->getNumInputDimensions() << ") does not match that of the FeatureExtraction Module at moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 
 if( !featureExtractionModules[ moduleIndex ]->computeFeatures( inputVector ) ){
 errorLog << "FeatureExtraction(VectorFloat inputVector) - Failed To Compute Features from Input Vector. FeatureExtraction moduleIndex: " << moduleIndex << std::endl;
 return false;
 }
 inputVector = featureExtractionModules[ moduleIndex ]->getFeatureVector();
 }
 }
 
 return true;
}
 
bool GestureRecognitionPipeline::getIsInitialized() const{ 
 return initialized; 
}
 
bool GestureRecognitionPipeline::getTrained() const{ 
 return trained;
}
 
bool GestureRecognitionPipeline::getIsPreProcessingSet() const{ 
 return preProcessingModules.size() > 0; 
} 
 
bool GestureRecognitionPipeline::getIsFeatureExtractionSet() const{ 
 return featureExtractionModules.size() > 0; 
}
 
bool GestureRecognitionPipeline::getIsClassifierSet() const{ 
 return (classifier!=NULL); 
}
 
bool GestureRecognitionPipeline::getIsRegressifierSet() const{ 
 return (regressifier!=NULL); 
}
 
bool GestureRecognitionPipeline::getIsClustererSet() const{
 return (clusterer!=NULL);
}
 
bool GestureRecognitionPipeline::getIsPostProcessingSet() const{ 
 return postProcessingModules.size() > 0; 
}
 
bool GestureRecognitionPipeline::getIsContextSet() const{ 
 for(UINT i=0; i<NUM_CONTEXT_LEVELS; i++){
 if( contextModules[i].size() > 0 ) return true;
 }
 return false;
}
 
bool GestureRecognitionPipeline::getIsPipelineModeSet() const{ 
 return (pipelineMode!=PIPELINE_MODE_NOT_SET); 
}
 
bool GestureRecognitionPipeline::getIsPipelineInClassificationMode() const{ 
 return (pipelineMode==CLASSIFICATION_MODE); 
}
 
bool GestureRecognitionPipeline::getIsPipelineInRegressionMode() const{ 
 return pipelineMode==REGRESSION_MODE; 
}

UINT GestureRecognitionPipeline::getInputVectorDimensionsSize() const{ 
 
 if( getIsPreProcessingSet() ){
 return preProcessingModules[0]->getNumInputDimensions();
 }
 
 if( getIsFeatureExtractionSet() ){
 return featureExtractionModules[0]->getNumInputDimensions();
 }
 
 if( getIsPipelineInClassificationMode() && getIsClassifierSet() ){
 return classifier->getNumInputFeatures();
 }
 if( getIsPipelineInRegressionMode() && getIsRegressifierSet() ){
 return regressifier->getNumInputFeatures(); 
 }
 return 0; 
}
 
UINT GestureRecognitionPipeline::getOutputVectorDimensionsSize() const{ 
 if( getIsClassifierSet() ) return 1; //The output of the pipeline for classification will always be 1
 if( getIsRegressifierSet() ){
 return regressifier->getNumOutputDimensions(); 
 }
 return 0;
}
 
UINT GestureRecognitionPipeline::getNumClassesInModel() const{ 
 return getNumClasses();
}
 
UINT GestureRecognitionPipeline::getNumClasses() const{
 UINT numClasses = 0;
 if( getIsClassifierSet() ){
 numClasses = classifier->getNumClasses();
 }
 if( getIsClustererSet() ){
 numClasses = clusterer->getNumClusters();
 }
 return numClasses;
}
 
UINT GestureRecognitionPipeline::getNumPreProcessingModules() const{ 
 return preProcessingModules.getSize(); 
}
 
UINT GestureRecognitionPipeline::getNumFeatureExtractionModules() const{ 
 return featureExtractionModules.getSize(); 
}
 
UINT GestureRecognitionPipeline::getNumPostProcessingModules() const{ 
 return postProcessingModules.getSize(); 
}
 
UINT GestureRecognitionPipeline::getPredictionModuleIndexPosition() const{ 
 return predictionModuleIndex; 
}
 
UINT GestureRecognitionPipeline::getPredictedClassLabel() const{
 
 if( getIsClassifierSet() ){
 return predictedClassLabel;
 }
 if( getIsClustererSet() ){
 return predictedClusterLabel;
 }
 return 0;
}
 
UINT GestureRecognitionPipeline::getUnProcessedPredictedClassLabel() const{
 if( getIsClassifierSet() ){
 return classifier->getPredictedClassLabel();
 }
 if( getIsClustererSet() ){
 return clusterer->getPredictedClusterLabel();
 }
 return 0;
}
 
UINT GestureRecognitionPipeline::getNumTrainingSamples() const{
 return numTrainingSamples;
}

UINT GestureRecognitionPipeline::getNumTestSamples() const{
 return numTestSamples;
}

Float GestureRecognitionPipeline::getMaximumLikelihood() const{
 if( getIsClassifierSet() ){
 return classifier->getMaximumLikelihood();
 }
 if( getIsClustererSet() ){
 return clusterer->getMaximumLikelihood();
 }
 return 0;
}
 
Float GestureRecognitionPipeline::getPhase() const{
 if( getIsClassifierSet() ){
 return classifier->getPhase();
 }
 return 0;
}
 
Float GestureRecognitionPipeline::getCrossValidationAccuracy() const{ 
 return (getIsClassifierSet()||getIsRegressifierSet() ? testAccuracy : 0);
}
 
Float GestureRecognitionPipeline::getTestAccuracy() const{ 
 return testAccuracy; 
}
 
Float GestureRecognitionPipeline::getTestRMSError() const{ 
 return testRMSError; 
}

Float GestureRecognitionPipeline::getTestSSError() const{
 return testSquaredError;
}
 
Float GestureRecognitionPipeline::getTestFMeasure(const UINT classLabel) const{
 
 if( !getIsClassifierSet() ) return -1;
 if( getClassLabels().size() != testFMeasure.size() ) return -1;
 
 for(UINT i=0; i<testFMeasure.size(); i++){
 if( getClassLabels()[i] == classLabel ){
 return testFMeasure[i];
 }
 }
 return -1;
}

Float GestureRecognitionPipeline::getTestPrecision(const UINT classLabel) const{
 
 if( !getIsClassifierSet() ) return -1;
 if( getClassLabels().size() != testPrecision.size() ) return -1;
 
 for(UINT i=0; i<testPrecision.size(); i++){
 if( getClassLabels()[i] == classLabel ){
 return testPrecision[i];
 }
 }
 return -1;
}

Float GestureRecognitionPipeline::getTestRecall(const UINT classLabel) const{
 
 if( !getIsClassifierSet() ) return -1;
 if( getClassLabels().getSize() != testRecall.getSize() ) return -1;
 
 for(UINT i=0; i<testRecall.getSize(); i++){
 if( getClassLabels()[i] == classLabel ){
 return testRecall[i];
 }
 }
 return -1;
}

Float GestureRecognitionPipeline::getTestRejectionPrecision() const{ 
 return testRejectionPrecision; 
}
 
Float GestureRecognitionPipeline::getTestRejectionRecall() const{ 
 return testRejectionRecall; 
}
 
Float GestureRecognitionPipeline::getTestTime() const{
 return testTime;
}

Float GestureRecognitionPipeline::getTrainingTime() const{
 return trainingTime;
}

Float GestureRecognitionPipeline::getTrainingRMSError() const{
 return getIsRegressifierSet() ? regressifier->getRootMeanSquaredTrainingError() : 0;
}

Float GestureRecognitionPipeline::getTrainingSSError() const{
 return getIsRegressifierSet() ? regressifier->getTotalSquaredTrainingError() : 0;
}
 
MatrixFloat GestureRecognitionPipeline::getTestConfusionMatrix() const{ 
 return testConfusionMatrix; 
}
 
TestResult GestureRecognitionPipeline::getTestResults() const {
 TestResult testResults;
 testResults.numTrainingSamples = numTrainingSamples;
 testResults.numTestSamples = numTestSamples;
 testResults.accuracy = testAccuracy;
 testResults.rmsError = testRMSError;
 testResults.totalSquaredError = testSquaredError;
 testResults.trainingTime = trainingTime;
 testResults.testTime = testTime;
 testResults.rejectionPrecision = testRejectionPrecision;
 testResults.rejectionRecall = testRejectionRecall;
 testResults.precision = testPrecision;
 testResults.recall = testRecall;
 testResults.fMeasure = testFMeasure;
 testResults.confusionMatrix = testConfusionMatrix;
 return testResults;
}
 
VectorFloat GestureRecognitionPipeline::getTestPrecision() const{
 return testPrecision;
}

VectorFloat GestureRecognitionPipeline::getTestRecall() const{
 return testRecall;
}

VectorFloat GestureRecognitionPipeline::getTestFMeasure() const{
 return testFMeasure;
}

VectorFloat GestureRecognitionPipeline::getClassLikelihoods() const{ 
 if( getIsClassifierSet() ){ return classifier->getClassLikelihoods(); }
 if( getIsClustererSet() ){ return clusterer->getClusterLikelihoods(); }
 else{ return VectorFloat(); } 
}

VectorFloat GestureRecognitionPipeline::getClassDistances() const{ 
 if( getIsClassifierSet() ){ return classifier->getClassDistances(); }
 if( getIsClustererSet() ){ return clusterer->getClusterDistances(); }
 else{ return VectorFloat(); } 
}

VectorFloat GestureRecognitionPipeline::getNullRejectionThresholds() const{
 if( getIsClassifierSet() ){ return classifier->getNullRejectionThresholds(); }
 else{ return VectorFloat(); } 
}

VectorFloat GestureRecognitionPipeline::getRegressionData() const{ 
 return regressionData;
}

VectorFloat GestureRecognitionPipeline::getUnProcessedRegressionData() const{ 
 if( getIsRegressifierSet() ) {
 return regressifier->getRegressionData();
 }
 return VectorFloat();
}
 
VectorFloat GestureRecognitionPipeline::getPreProcessedData() const{
 if( getIsPreProcessingSet() ){ 
 return preProcessingModules[ preProcessingModules.size()-1 ]->getProcessedData(); 
 }
 return VectorFloat();
}

VectorFloat GestureRecognitionPipeline::getPreProcessedData(const UINT moduleIndex) const{
 if( getIsPreProcessingSet() ){ 
 if( moduleIndex < preProcessingModules.size() ){
 return preProcessingModules[ moduleIndex ]->getProcessedData(); 
 }
 }
 return VectorFloat();
}

VectorFloat GestureRecognitionPipeline::getFeatureExtractionData() const{
 if( getIsFeatureExtractionSet() ){ 
 return featureExtractionModules[ featureExtractionModules.size()-1 ]->getFeatureVector(); 
 }
 return VectorFloat();
}
 
VectorFloat GestureRecognitionPipeline::getFeatureExtractionData(const UINT moduleIndex) const{
 if( getIsFeatureExtractionSet() ){ 
 if( moduleIndex < featureExtractionModules.size() ){
 return featureExtractionModules[ moduleIndex ]->getFeatureVector(); 
 }
 }
 warningLog << "getFeatureExtractionData(UINT moduleIndex) - Failed to get class labels!" << std::endl;
 return VectorFloat();
}
 
Vector< UINT > GestureRecognitionPipeline::getClassLabels() const{ 
 if( getIsClassifierSet() ){
 return classifier->getClassLabels();
 }
 if( getIsClustererSet() ){
 return clusterer->getClusterLabels();
 }
 warningLog << "getClassLabels() - Failed to get class labels!" << std::endl;
 return Vector< UINT>(); 
}
 
Vector< TestInstanceResult > GestureRecognitionPipeline::getTestInstanceResults() const{
 return testResults;
}

Vector< TestResult > GestureRecognitionPipeline::getCrossValidationResults() const{
 return crossValidationResults;
}
 
PreProcessing* GestureRecognitionPipeline::getPreProcessingModule(const UINT moduleIndex) const{
 if( moduleIndex < preProcessingModules.size() ){
 return preProcessingModules[ moduleIndex ];
 }
 warningLog << "getPreProcessingModule(const UINT moduleIndex) - Failed to get pre processing module!" << std::endl;
 return NULL;
}
 
FeatureExtraction* GestureRecognitionPipeline::getFeatureExtractionModule(const UINT moduleIndex) const{
 if( moduleIndex < featureExtractionModules.size() ){
 return featureExtractionModules[ moduleIndex ];
 }
 warningLog << "getFeatureExtractionModule(const UINT moduleIndex) - Failed to get feature extraction module!" << std::endl;
 return NULL;
}
 
Classifier* GestureRecognitionPipeline::getClassifier() const{
 return classifier; 
}
 
Regressifier* GestureRecognitionPipeline::getRegressifier() const{
 return regressifier; 
}
 
Clusterer* GestureRecognitionPipeline::getClusterer() const{
 return clusterer;
}
 
PostProcessing* GestureRecognitionPipeline::getPostProcessingModule(UINT moduleIndex) const{
 if( moduleIndex < postProcessingModules.size() ){
 return postProcessingModules[ moduleIndex ];
 }
 warningLog << "getPostProcessingModule(UINT moduleIndex) - Failed to get post processing module!" << std::endl;
 return NULL;
}
 
Context* GestureRecognitionPipeline::getContextModule(UINT contextLevel,UINT moduleIndex) const{
 if( contextLevel < contextModules.size() ){
 if( moduleIndex < contextModules[ contextLevel ].size() ){
 return contextModules[ contextLevel ][ moduleIndex ];
 }
 }
 warningLog << "getContextModule(UINT contextLevel,UINT moduleIndex) - Failed to get context module!" << std::endl;
 return NULL;
}
 
 
bool GestureRecognitionPipeline::addPreProcessingModule(const PreProcessing &preProcessingModule,UINT insertIndex){
 
 //Validate the insertIndex is valid
 if( insertIndex != INSERT_AT_END_INDEX && insertIndex >= preProcessingModules.size() ){
 errorLog << "addPreProcessingModule(const PreProcessing &preProcessingModule) - Invalid insertIndex value!" << std::endl;
 return false;
 }
 
 //Create a new instance of the preProcessing and then clone the values across from the reference preProcessing
 PreProcessing *newInstance = preProcessingModule.createNewInstance();
 
 //Verify that the clone was successful
 if( !newInstance->deepCopyFrom( &preProcessingModule ) ){
 delete newInstance;
 newInstance = NULL;
 errorLog << "addPreProcessingModule(const PreProcessing &preProcessingModule) - PreProcessing Module Not Set!" << std::endl;
 return false;
 }
 
 //Add the new instance to the preProcessingModules
 Vector< PreProcessing* >::iterator iter = preProcessingModules.begin();
 
 if( insertIndex == INSERT_AT_END_INDEX ) iter = preProcessingModules.end(); 
 else iter = preProcessingModules.begin() + insertIndex;
 
 preProcessingModules.insert(iter, newInstance);

 //The pipeline has been changed, so flag that the pipeline is no longer trained
 trained = false;
 
 return true;
}

bool GestureRecognitionPipeline::setPreProcessingModule(const PreProcessing &preProcessingModule){
 removeAllPreProcessingModules();
 return addPreProcessingModule( preProcessingModule );
}

bool GestureRecognitionPipeline::addFeatureExtractionModule(const FeatureExtraction &featureExtractionModule,UINT insertIndex){
 
 //Validate the insertIndex is valid
 if( insertIndex != INSERT_AT_END_INDEX && insertIndex >= featureExtractionModules.size() ){
 errorLog << "addFeatureExtractionModule(const FeatureExtraction &featureExtractionModule) - Invalid insertIndex value!" << std::endl;
 return false;
 }
 
 //Create a new instance of the preProcessing and then clone the values across from the reference preProcessing
 FeatureExtraction *newInstance = featureExtractionModule.createNewInstance();
 
 //Verify that the clone was successful
 if( !newInstance->deepCopyFrom( &featureExtractionModule ) ){
 delete newInstance;
 newInstance = NULL;
 errorLog << "addFeatureExtractionModule(const FeatureExtraction &featureExtractionModule - FeatureExtraction Module Not Set!" << std::endl;
 return false;
 }
 
 //Add the new instance to the preProcessingModules
 Vector< FeatureExtraction* >::iterator iter = featureExtractionModules.begin();
 
 if( insertIndex == INSERT_AT_END_INDEX ) iter = featureExtractionModules.end(); 
 else iter = featureExtractionModules.begin() + insertIndex;
 
 featureExtractionModules.insert(iter, newInstance);

 //The pipeline has been changed, so flag that the pipeline is no longer trained
 trained = false;
 
 return true;
}

bool GestureRecognitionPipeline::setFeatureExtractionModule(const FeatureExtraction &featureExtractionModule){
 removeAllFeatureExtractionModules();
 return addFeatureExtractionModule( featureExtractionModule );
}

bool GestureRecognitionPipeline::setClassifier(const Classifier &classifier){
 
 //Delete any previous classifier, regressifier or clusterer
 deleteClassifier();
 deleteRegressifier();
 deleteClusterer();

 //Create a new instance of the classifier and then clone the values across from the reference classifier
 this->classifier = classifier.createNewInstance();
 
 if( this->classifier == NULL ){
 errorLog << "setClassifier(const Classifier classifier) - Classifier Module Not Set!" << std::endl;
 return false;
 }
 
 //Deep copy the data from the rhs classifier into this classifier
 if( !this->classifier->deepCopyFrom( &classifier ) ){
 deleteClassifier();
 pipelineMode = PIPELINE_MODE_NOT_SET;
 errorLog << "setClassifier(const Classifier classifier) - Classifier Module Not Set!" << std::endl;
 return false;
 }
 
 //Set the mode of the pipeline to classification mode
 pipelineMode = CLASSIFICATION_MODE;
 
 //Flag that the key part of the pipeline has now been initialized
 initialized = true;
 
 //If there is no preprocessing / feature extraction and the classifier is trained, then flag the pipeline is trained
 //Otherwise the user needs to train the pipeline
 if( !getIsPreProcessingSet() && !getIsFeatureExtractionSet() && classifier.getTrained() ){
 inputVectorDimensions = classifier.getNumInputDimensions();
 trained = true;
 }else trained = false;

 return true;
}

bool GestureRecognitionPipeline::setRegressifier(const Regressifier &regressifier){
 
 //Delete any previous classifier, regressifier or clusterer
 deleteClassifier();
 deleteRegressifier();
 deleteClusterer();
 
 //Set the mode of the pipeline to regression mode
 pipelineMode = REGRESSION_MODE;
 
 //Create a new instance of the regressifier and then clone the values across from the reference regressifier
 this->regressifier = regressifier.createNewInstance();
 
 //Validate that the regressifier was cloned correctly
 if( !this->regressifier->deepCopyFrom( &regressifier ) ){
 deleteRegressifier();
 pipelineMode = PIPELINE_MODE_NOT_SET;
 errorLog << "setRegressifier(const Regressifier &regressifier) - Regressifier Module Not Set!" << std::endl;
 return false;
 }
 
 //Flag that the key part of the pipeline has now been initialized
 initialized = true;

 //If there is no preprocessing / feature extraction and the regressifier is trained, then flag the pipeline is trained
 //Otherwise the user needs to train the pipeline
 if( !getIsPreProcessingSet() && !getIsFeatureExtractionSet() ){
 trained = regressifier.getTrained();
 }else trained = false;
 
 return true;
}
 
bool GestureRecognitionPipeline::setClusterer(const Clusterer &clusterer){
 
 //Delete any previous classifier, regressifier or clusterer
 deleteClassifier();
 deleteRegressifier();
 deleteClusterer();
 
 //Set the mode of the pipeline to cluster mode
 pipelineMode = CLUSTER_MODE;
 
 //Create a new instance of the clusterer and then clone the values across from the reference clusterer
 this->clusterer = clusterer.createNewInstance();
 
 //Validate that the regressifier was cloned correctly
 if( !this->clusterer->deepCopyFrom( &clusterer ) ){
 deleteClusterer();
 pipelineMode = PIPELINE_MODE_NOT_SET;
 errorLog << "setClusterer(const Clusterer &clusterer) - Clusterer Module Not Set!" << std::endl;
 return false;
 }
 
 //Flag that the key part of the pipeline has now been initialized
 initialized = true;
 
 //If there is no preprocessing / feature extraction and the regressifier is trained, then flag the pipeline is trained
 //Otherwise the user needs to train the pipeline
 if( !getIsPreProcessingSet() && !getIsFeatureExtractionSet() ){
 trained = clusterer.getTrained();
 }else trained = false;
 
 return true;
}

bool GestureRecognitionPipeline::addPostProcessingModule(const PostProcessing &postProcessingModule,UINT insertIndex){
 
 //Validate the insertIndex is valid
 if( insertIndex != INSERT_AT_END_INDEX && insertIndex >= postProcessingModules.size() ){
 errorLog << "addPostProcessingModule((const PostProcessing &postProcessingModule) - Invalid insertIndex value!" << std::endl;
 return false;
 }
 
 //Create a new instance of the preProcessing and then clone the values across from the reference preProcessing
 PostProcessing *newInstance = postProcessingModule.createNewInstance();
 
 //Verify that the clone was successful
 if( !newInstance->deepCopyFrom( &postProcessingModule ) ){
 delete newInstance;
 newInstance = NULL;
 errorLog << "addPostProcessingModule(const PostProcessing &postProcessingModule) - PostProcessing Module Not Set!" << std::endl;
 return false;
 }
 
 //Add the new instance to the postProcessingModules
 Vector< PostProcessing* >::iterator iter = postProcessingModules.begin();
 
 if( insertIndex == INSERT_AT_END_INDEX ) iter = postProcessingModules.end(); 
 else iter = postProcessingModules.begin() + insertIndex;
 
 postProcessingModules.insert(iter, newInstance);

 //Note, we don't change the trained state of the pipeline for post processing modules, as they are added after the core ML module

 return true;
}

bool GestureRecognitionPipeline::setPostProcessingModule(const PostProcessing &postProcessingModule){
 removeAllPostProcessingModules();
 return addPostProcessingModule( postProcessingModule );
}

bool GestureRecognitionPipeline::addContextModule(const Context &contextModule,UINT contextLevel,UINT insertIndex){
 
 //Validate the contextLevel is valid
 if( contextLevel >= contextModules.size() ){
 errorLog << "addContextModule(...) - Invalid contextLevel value!" << std::endl;
 return false;
 }

 //Validate the insertIndex is valid
 if( insertIndex != INSERT_AT_END_INDEX && insertIndex >= contextModules[contextLevel].size() ){
 errorLog << "addContextModule(...) - Invalid insertIndex value!" << std::endl;
 return false;
 }
 
 //Create a new instance of the preProcessing and then clone the values across from the reference preProcessing
 Context *newInstance = contextModule.createNewInstance();
 
 //Verify that the clone was successful
 if( !newInstance->deepCopyFrom( &contextModule ) ){
 delete newInstance;
 newInstance = NULL;
 errorLog << "addContextModule(...) - Context Module Not Set!" << std::endl;
 return false;
 }
 
 //Add the new instance to the contextModules
 Vector< Context* >::iterator iter = contextModules[ contextLevel ].begin();
 
 if( insertIndex == INSERT_AT_END_INDEX ) iter = contextModules[ contextLevel ].end(); 
 else iter = contextModules[ contextLevel ].begin() + insertIndex;
 
 contextModules[ contextLevel ].insert(iter, newInstance);
 
 return true;
}

bool GestureRecognitionPipeline::updateContextModule(bool value,UINT contextLevel,UINT moduleIndex){
 
 //Validate the contextLevel is valid
 if( contextLevel >= contextModules.size() ){
 errorLog << "updateContextModule(...) - Context Level is out of bounds!" << std::endl;
 return false;
 }
 
 //Validate the moduleIndex is valid
 if( moduleIndex >= contextModules[contextLevel].size() ){
 errorLog << "updateContextModule(...) - Invalid contextLevel value!" << std::endl;
 return false;
 }
 
 return contextModules[contextLevel][moduleIndex]->updateContext( value );
}
 
bool GestureRecognitionPipeline::removeAllPreProcessingModules(){
 deleteAllPreProcessingModules();
 return true;
}
 
bool GestureRecognitionPipeline::removePreProcessingModule(UINT moduleIndex){
 if( moduleIndex >= preProcessingModules.size() ){
 errorLog << "removePreProcessingModule(UINT moduleIndex) - Invalid moduleIndex " << moduleIndex << ". The size of the preProcessingModules Vector is " << int(preProcessingModules.size()) << std::endl;
 return false;
 }
 
 //Delete the module
 delete preProcessingModules[ moduleIndex ];
 preProcessingModules[ moduleIndex ] = NULL;
 preProcessingModules.erase( preProcessingModules.begin() + moduleIndex );

 //The pipeline has been changed, so flag that the pipeline is no longer trained
 trained = false;

 return true;
}
 
bool GestureRecognitionPipeline::removeAllFeatureExtractionModules(){
 deleteAllFeatureExtractionModules();
 return true;
}
 
bool GestureRecognitionPipeline::removeFeatureExtractionModule(UINT moduleIndex){
 if( moduleIndex >= featureExtractionModules.size() ){
 errorLog << "removeFeatureExtractionModule(UINT moduleIndex) - Invalid moduleIndex " << moduleIndex << ". The size of the featureExtractionModules Vector is " << int(featureExtractionModules.size()) << std::endl;
 return false;
 }
 
 //Delete the module
 delete featureExtractionModules[ moduleIndex ];
 featureExtractionModules[ moduleIndex ] = NULL;
 featureExtractionModules.erase( featureExtractionModules.begin() + moduleIndex );

 //The pipeline has been changed, so flag that the pipeline is no longer trained
 trained = false;
 
 return true;
}

bool GestureRecognitionPipeline::removeAllPostProcessingModules(){
 deleteAllPostProcessingModules();
 return true;
}

bool GestureRecognitionPipeline::removePostProcessingModule(UINT moduleIndex){
 if( moduleIndex >= postProcessingModules.size() ){
 errorLog << "removePostProcessingModule(UINT moduleIndex) - Invalid moduleIndex " << moduleIndex << ". The size of the postProcessingModules Vector is " << int(postProcessingModules.size()) << std::endl;
 return false;
 }
 
 //Delete the module
 delete postProcessingModules[ moduleIndex ];
 postProcessingModules[ moduleIndex ] = NULL;
 postProcessingModules.erase( postProcessingModules.begin() + moduleIndex );

 //The pipeline has been changed, so flag that the pipeline is no longer trained
 trained = false;
 
 return true;
}
 
bool GestureRecognitionPipeline::removeContextModule(UINT contextLevel,UINT moduleIndex){
 if( contextLevel >= NUM_CONTEXT_LEVELS ){
 errorLog << "removeContextModule(UINT contextLevel,UINT moduleIndex) - Invalid moduleIndex " << moduleIndex << " is out of bounds!" << std::endl;
 return false;
 }
 
 if( moduleIndex >= contextModules[contextLevel].size() ){
 errorLog << "removePostProcessingModule(UINT moduleIndex) - Invalid moduleIndex " << moduleIndex << ". The size of the contextModules Vector at context level " << " is " << int(contextModules[contextLevel].size()) << std::endl;
 return false;
 }
 
 //Delete the module
 delete contextModules[contextLevel][moduleIndex];
 contextModules[contextLevel][moduleIndex] = NULL;
 contextModules[contextLevel].erase( contextModules[contextLevel].begin() + moduleIndex );
 return true;
}
 
bool GestureRecognitionPipeline::removeAllContextModules(){
 deleteAllContextModules();
 return true;
}
 
bool GestureRecognitionPipeline::clearTestResults(){
 
 numTestSamples = 0;
 testAccuracy = 0;
 testRMSError = 0;
 testSquaredError = 0;
 testTime = 0;
 testFMeasure.clear();
 testPrecision.clear();
 testRecall.clear();
 testRejectionPrecision = 0;
 testRejectionRecall = 0;
 testConfusionMatrix.clear();
 testResults.clear();
 crossValidationResults.clear();
 
 return true;
}
 
bool GestureRecognitionPipeline::setInfo(const std::string &info){
 this->info = info;
 return true;
}

 
void GestureRecognitionPipeline::deleteAllPreProcessingModules(){
 if( preProcessingModules.size() != 0 ){
 for(UINT i=0; i<preProcessingModules.size(); i++){
 delete preProcessingModules[i];
 preProcessingModules[i] = NULL;
 }
 preProcessingModules.clear();
 trained = false;
 }
}
 
void GestureRecognitionPipeline::deleteAllFeatureExtractionModules(){
 if( featureExtractionModules.size() != 0 ){
 for(UINT i=0; i<featureExtractionModules.size(); i++){
 delete featureExtractionModules[i];
 featureExtractionModules[i] = NULL;
 }
 featureExtractionModules.clear();
 trained = false;
 }
}
 
void GestureRecognitionPipeline::deleteClassifier(){
 if( classifier != NULL ){
 delete classifier;
 classifier = NULL;
 }
 trained = false;
 initialized = false;
}
 
void GestureRecognitionPipeline::deleteRegressifier(){
 if( regressifier != NULL ){
 delete regressifier;
 regressifier = NULL;
 }
 trained = false;
 initialized = false;
}
 
void GestureRecognitionPipeline::deleteClusterer(){
 if( clusterer != NULL ){
 delete clusterer;
 clusterer = NULL;
 }
 trained = false;
 initialized = false;
}
 
void GestureRecognitionPipeline::deleteAllPostProcessingModules(){
 if( postProcessingModules.size() != 0 ){
 for(UINT i=0; i<postProcessingModules.size(); i++){
 delete postProcessingModules[i];
 postProcessingModules[i] = NULL;
 }
 postProcessingModules.clear();
 trained = false;
 }
}
 
void GestureRecognitionPipeline::deleteAllContextModules(){
 for(UINT i=0; i<contextModules.size(); i++){
 for(UINT j=0; j<contextModules[i].size(); j++){
 delete contextModules[i][j];
 contextModules[i][j] = NULL;
 }
 contextModules[i].clear();
 }
}

bool GestureRecognitionPipeline::init(){
 initialized = false;
 trained = false;
 info = "";
 pipelineMode = PIPELINE_MODE_NOT_SET;
 inputVectorDimensions = 0;
 outputVectorDimensions = 0;
 predictedClassLabel = 0;
 predictedClusterLabel = 0;
 predictionModuleIndex = 0;
 numTrainingSamples = 0;
 numTestSamples = 0;
 testAccuracy = 0;
 testRMSError = 0;
 testSquaredError = 0;
 testRejectionPrecision = 0;
 testRejectionRecall = 0;
 testTime = 0;
 trainingTime = 0;
 classifier = NULL;
 regressifier = NULL;
 clusterer = NULL;
 contextModules.resize( NUM_CONTEXT_LEVELS );
 return true;
}
 
bool GestureRecognitionPipeline::updateTestMetrics(const UINT classLabel,const UINT predictedClassLabel,VectorFloat &precisionCounter,VectorFloat &recallCounter,Float &rejectionPrecisionCounter,Float &rejectionRecallCounter,VectorFloat &confusionMatrixCounter){

 const bool nullRejectionEnabled = classifier->getNullRejectionEnabled();

 //Find the index of the classLabel
 UINT predictedClassLabelIndex =0;
 bool predictedClassLabelIndexFound = false;
 for(UINT k=0; k<getNumClassesInModel(); k++){
 if( predictedClassLabel == classifier->getClassLabels()[k] ){
 predictedClassLabelIndex = k;
 predictedClassLabelIndexFound = true;
 break;
 }
 }
 
 if( !predictedClassLabelIndexFound && (nullRejectionEnabled == false || predictedClassLabel != GRT_DEFAULT_NULL_CLASS_LABEL) ){
 errorLog << "Failed to find class label index for label: " << predictedClassLabel << std::endl;
 return false;
 }

 //Find the index of the class label
 UINT actualClassLabelIndex = 0;
 for(UINT k=0; k<getNumClassesInModel(); k++){
 if( classLabel == classifier->getClassLabels()[k] ){
 actualClassLabelIndex = k;
 break;
 }
 }

 //Update the classification accuracy
 if( classLabel == predictedClassLabel ){
 testAccuracy++;
 }

 if( nullRejectionEnabled == false ){

 //Update the precision
 if( classLabel == predictedClassLabel ){
 //Update the precision value
 testPrecision[ predictedClassLabelIndex ]++;
 }
 //Update the precision counter
 precisionCounter[ predictedClassLabelIndex ]++;

 //Update the recall
 if( classLabel == predictedClassLabel ){
 //Update the recall value
 testRecall[ predictedClassLabelIndex ]++;
 }
 //Update the recall counter
 recallCounter[ actualClassLabelIndex ]++;

 //Update the confusion matrix
 testConfusionMatrix[ actualClassLabelIndex ][ predictedClassLabelIndex ]++;
 confusionMatrixCounter[ actualClassLabelIndex ]++;

 }else{ //Null rejection is enabled
 //Update the precision
 if( predictedClassLabel != GRT_DEFAULT_NULL_CLASS_LABEL ){
 if( classLabel == predictedClassLabel ){
 //Update the precision value
 testPrecision[ predictedClassLabelIndex ]++;
 }
 //Update the precision counter
 precisionCounter[ predictedClassLabelIndex ]++;
 }

 //Update the recall
 if( classLabel != GRT_DEFAULT_NULL_CLASS_LABEL ){
 if( classLabel == predictedClassLabel ){
 //Update the recall value
 testRecall[ predictedClassLabelIndex ]++;
 }
 //Update the recall counter
 recallCounter[ actualClassLabelIndex ]++;
 }

 //Update the rejection precision
 if( predictedClassLabel == GRT_DEFAULT_NULL_CLASS_LABEL ){
 if( classLabel == GRT_DEFAULT_NULL_CLASS_LABEL ) testRejectionPrecision++;
 rejectionPrecisionCounter++;
 }

 //Update the rejection recall
 if( classLabel == GRT_DEFAULT_NULL_CLASS_LABEL ){
 if( predictedClassLabel == GRT_DEFAULT_NULL_CLASS_LABEL ) testRejectionRecall++;
 rejectionRecallCounter++;
 }

 //Update the confusion matrix
 if( classLabel == GRT_DEFAULT_NULL_CLASS_LABEL ) actualClassLabelIndex = 0;
 else actualClassLabelIndex++;
 if( predictedClassLabel == GRT_DEFAULT_NULL_CLASS_LABEL ) predictedClassLabelIndex = 0;
 else predictedClassLabelIndex++;
 testConfusionMatrix[ actualClassLabelIndex ][ predictedClassLabelIndex ]++;
 confusionMatrixCounter[ actualClassLabelIndex ]++;
 }
 
 return true;
}

bool GestureRecognitionPipeline::computeTestMetrics(VectorFloat &precisionCounter,VectorFloat &recallCounter,Float &rejectionPrecisionCounter,Float &rejectionRecallCounter,VectorFloat &confusionMatrixCounter,const UINT numTestSamples){
 
 //Compute the test metrics
 testAccuracy = testAccuracy/Float(numTestSamples) * 100.0;
 
 for(UINT k=0; k<getNumClassesInModel(); k++){
 if( precisionCounter[k] > 0 ) testPrecision[k] /= precisionCounter[k];
 else testPrecision[k] = 0;
 if( recallCounter[k] > 0 ) testRecall[k] /= recallCounter[k];
 else testRecall[k] = 0;
 
 if( precisionCounter[k] + recallCounter[k] > 0 )
 testFMeasure[k] = 2 * ((testPrecision[k]*testRecall[k])/(testPrecision[k]+testRecall[k]));
 else testFMeasure[k] = 0;
 }
 if( rejectionPrecisionCounter > 0 ) testRejectionPrecision /= rejectionPrecisionCounter;
 if( rejectionRecallCounter > 0 ) testRejectionRecall /= rejectionRecallCounter;
 
 
 for(UINT r=0; r<confusionMatrixCounter.size(); r++){
 if( confusionMatrixCounter[r] > 0 ){
 for(UINT c=0; c<testConfusionMatrix.getNumCols(); c++){
 testConfusionMatrix[r][c] /= confusionMatrixCounter[r];
 }
 }
 }
 
 return true;
}
 
std::string GestureRecognitionPipeline::getModelAsString() const{
 std::string model = "";
 
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 break;
 case CLASSIFICATION_MODE:
 if( getIsClassifierSet() ){
 model += "Classifier: " + classifier->getClassifierType() + "\n";
 model += classifier->getModelAsString();
 }
 break;
 case REGRESSION_MODE:
 if( getIsRegressifierSet() ){
 model += "Regressifier: " + regressifier->getRegressifierType() + "\n";
 model += regressifier->getModelAsString();
 }
 break;
 default:
 break;
 }
 
 return model;
}
 
std::string GestureRecognitionPipeline::getPipelineModeAsString() const{
 switch( pipelineMode ){
 case PIPELINE_MODE_NOT_SET:
 return "PIPELINE_MODE_NOT_SET";
 break;
 case CLASSIFICATION_MODE:
 return "CLASSIFICATION_MODE";
 break;
 case REGRESSION_MODE:
 return "REGRESSION_MODE";
 break;
 default:
 return "ERROR_UNKNWON_PIPELINE_MODE";
 break;
 }
 
 return "ERROR_UNKNWON_PIPELINE_MODE";
}
 
std::string GestureRecognitionPipeline::getInfo() const{
 return info;
}

UINT GestureRecognitionPipeline::getPipelineModeFromString(std::string pipelineModeAsString) const{
 if( pipelineModeAsString == "PIPELINE_MODE_NOT_SET" ){
 return PIPELINE_MODE_NOT_SET;
 }
 if( pipelineModeAsString == "CLASSIFICATION_MODE" ){
 return CLASSIFICATION_MODE;
 }
 if( pipelineModeAsString == "REGRESSION_MODE" ){
 return REGRESSION_MODE;
 }
 return PIPELINE_MODE_NOT_SET;
}

GRT_END_NAMESPACE