MixtureModel.h

Go to the documentation of this file.

#ifndef GRT_MIXTURE_MODEL_HEADER
#define GRT_MIXTURE_MODEL_HEADER

#include "../../CoreModules/Classifier.h"
#include "../../ClusteringModules/GaussianMixtureModels/GaussianMixtureModels.h"

GRT_BEGIN_NAMESPACE
 
class GuassModel{
public:
 GuassModel(){
 det = 0;
 }
 
 ~GuassModel(){
 
 }
 
 bool printModelValues() const{
 if( mu.size() == 0 ) return false;
 
 std::cout << "Determinate: " << det << std::endl;
 std::cout << "Mu: ";
 for(UINT i=0; i<mu.getSize(); i++)
 std::cout << mu[i] << "\t";
 std::cout << std::endl;
 
 std::cout << "Sigma: \n";
 for(UINT i=0; i<sigma.getNumRows(); i++){
 for(UINT j=0; j<sigma.getNumCols(); j++){
 std::cout << sigma[i][j] << "\t";
 }std::cout << std::endl;
 }std::cout << std::endl;
 
 std::cout << "InvSigma: \n";
 for(UINT i=0; i<invSigma.getNumRows(); i++){
 for(UINT j=0; j<invSigma.getNumCols(); j++){
 std::cout << invSigma[i][j] << "\t";
 }std::cout << std::endl;
 }std::cout << std::endl;
 
 return true;
 }
 
 Float det;
 VectorFloat mu;
 MatrixFloat sigma;
 MatrixFloat invSigma;
};
 
class MixtureModel{
public:
 MixtureModel(){
 classLabel = 0;
 K = 0;
 normFactor = 1;
 nullRejectionThreshold = 0;
 trainingMu = 0;
 trainingSigma = 0;
 gamma = 1;
 }
 ~MixtureModel(){
 gaussModels.clear();
 }
 
 inline GuassModel& operator[](const UINT i){
 return gaussModels[i];
 }
 
 inline const GuassModel& operator[](const UINT i) const{
 return gaussModels[i];
 }
 
 Float computeMixtureLikelihood( const VectorFloat &x ){
 Float sum = 0;
 for(UINT k=0; k<K; k++){
 sum += gauss(x,gaussModels[k].det,gaussModels[k].mu,gaussModels[k].invSigma);
 }
 //Normalize the mixture likelihood
 return sum/normFactor;
 }
 
 bool resize(UINT K){
 if( K > 0 ){
 this->K = K;
 gaussModels.clear();
 gaussModels.resize(K);
 return true;
 }
 return false;
 }
 
 bool recomputeNullRejectionThreshold(Float gamma){
 Float newRejectionThreshold = 0;
 //TODO - Need a way of improving the null rejection threshold for the GMMs!!!!
 newRejectionThreshold = trainingMu - (trainingSigma*gamma);
 newRejectionThreshold = 0.02;
 
 //Make sure that the new rejection threshold is greater than zero
 if( newRejectionThreshold > 0 ){
 this->gamma = gamma;
 this->nullRejectionThreshold = newRejectionThreshold;
 return true;
 }
 return false;
 }
 
 bool recomputeNormalizationFactor(){
 normFactor = 0;
 for(UINT k=0; k<K; k++){
 normFactor += gauss(gaussModels[k].mu,gaussModels[k].det,gaussModels[k].mu,gaussModels[k].invSigma);
 }
 return true;
 }
 
 bool printModelValues() const{
 if( gaussModels.getSize() > 0 ){
 for(UINT k=0; k<gaussModels.getSize(); k++){
 gaussModels[k].printModelValues();
 }
 }
 return false;
 }
 
 UINT getK() const { return K; }
 
 UINT getClassLabel() const { return classLabel; }
 
 Float getTrainingMu() const {
 return trainingMu;
 }
 
 Float getTrainingSigma() const {
 return trainingSigma;
 }
 
 Float getNullRejectionThreshold() const {
 return nullRejectionThreshold;
 }
 
 Float getNormalizationFactor() const {
 return normFactor;
 }
 
 bool setClassLabel(const UINT classLabel){
 this->classLabel = classLabel;
 return true;
 }
 
 bool setNormalizationFactor(const Float normFactor){
 this->normFactor = normFactor;
 return true;
 }
 
 bool setTrainingMuAndSigma(const Float trainingMu,const Float trainingSigma){
 this->trainingMu = trainingMu;
 this->trainingSigma = trainingSigma;
 return true;
 }
 
 bool setNullRejectionThreshold(const Float nullRejectionThreshold){
 this->nullRejectionThreshold = nullRejectionThreshold;
 return true;
 }
 
private: 
 Float gauss(const VectorFloat &x,Float det,const VectorFloat &mu,const MatrixFloat &invSigma){
 
 Float y = 0;
 Float sum = 0;
 const UINT N = x.getSize();
 VectorFloat temp(N,0);
 
 //Compute the first part of the equation
 y = (1.0/pow(TWO_PI,N/2.0)) * (1.0/pow(det,0.5));
 
 //Compute the later half 
 for(UINT i=0; i<N; i++){
 for(UINT j=0; j<N; j++){
 temp[i] += (x[j]-mu[j]) * invSigma[j][i];
 }
 sum += (x[i]-mu[i]) * temp[i];
 }
 
 return ( y*grt_exp( -0.5*sum ) );
 }
 
 UINT classLabel;
 UINT K;
 Float nullRejectionThreshold;
 Float gamma; //The number of standard deviations to use for the threshold
 Float trainingMu; //The average confidence value in the training data
 Float trainingSigma; //The simga confidence value in the training data
 Float normFactor;
 Vector< GuassModel > gaussModels;
 
};
 
GRT_END_NAMESPACE

#endif //GRT_MIXTURE_MODEL_HEADER