recursive_least_squares.cpp

Go to the documentation of this file.

#include "recursive_least_squares.hpp"
#include <string>
 
namespace DDSPC
{
 
RecursiveLeastSquares::RecursiveLeastSquares(int num_predictors, int num_features, realT forgetting_factor, realT initial_covariance){
 
   _gamma = forgetting_factor;
   _inverse_gamma = 1 / _gamma;
 
   _initial_covariance = initial_covariance;
   _num_features = num_features;
   _num_predictors = num_predictors;
   
   // Set size of all the arrays
   prediction_matrix.resize(_num_predictors, _num_features);
   prediction_matrix.setZero();
    
    prediction_output.resize(_num_predictors, 1);
    prediction_output.setZero();
 
   inverse_covariance.resize(_num_features, _num_features);
   inverse_covariance.setZero();
   for(int i=0; i < _num_features; i++)
      inverse_covariance(i, i) = _initial_covariance;
 
   err.resize(_num_predictors, 1);
   err.setZero();
 
   K.resize(1, _num_features);
   K.setZero();
};
 
 
RecursiveLeastSquares::~RecursiveLeastSquares(){
 
}
 
void RecursiveLeastSquares::reset(){
   prediction_matrix.setZero();
   err.setZero();
   K.setZero();
 
   inverse_covariance.resize(_num_features, _num_features);
   inverse_covariance.setZero();
   for(int i=0; i < _num_features; i++)
      inverse_covariance(i, i) = _initial_covariance;
}
 
// I want to change this interface to make it easier to use.
void RecursiveLeastSquares::update(eigenImage<realT> *x, eigenImage<realT> *y){
   Matrix _x = (*x).matrix();
    err = (*y).matrix();
    err -= prediction_matrix * _x;
 
   xtP = (_inverse_gamma * _x).transpose() * inverse_covariance;
   realT cn = 1 + (xtP * _x)(0,0);
    K = xtP;
    K /= cn;
    prediction_matrix += err * K;
   
    inverse_covariance *= _inverse_gamma;
    inverse_covariance -= K.transpose() * xtP;
}
 
void RecursiveLeastSquares::update(Matrix *x, Matrix *y){
   Matrix _x = (*x);
    err = (*y);
    err -= prediction_matrix * _x;
 
   xtP = (_inverse_gamma * _x).transpose() * inverse_covariance;
   realT cn = 1 + (xtP * _x)(0,0);
    K = xtP;
    K /= cn;
    prediction_matrix += err * K;
   
    inverse_covariance *= _inverse_gamma;
    inverse_covariance -= K.transpose() * xtP;
}
 
 
// Matrix RecursiveLeastSquares::predict(eigenImage<realT> *x){
//    return prediction_matrix * (*x).matrix();
//}
 
void RecursiveLeastSquares::save_state(std::string filename){
 
}
 
}