LinearLeastSquares.h

Go to the documentation of this file.

#ifndef GRT_LINEAR_LEAST_SQUARES_HEADER
#define GRT_LINEAR_LEAST_SQUARES_HEADER

#include "../../CoreModules/MLBase.h"

GRT_BEGIN_NAMESPACE

class LinearLeastSquares : public MLBase{
public:
    LinearLeastSquares() : MLBase("LinearLeastSquares") {        
        m = 0;
        b = 0;
        r = 0;
    }
    
    virtual ~LinearLeastSquares(){
        
    }

    virtual bool clear() override {
        m = 0;
        b = 0;
        r = 0;
        return MLBase::clear();
    }
    
    bool solve( const VectorFloat &x, const VectorFloat &y ){

        //Reset any previous results
        clear();
        
        if( x.getSize() == 0 && y.getSize() == 0 ){
            warningLog << __GRT_LOG__ << " Failed to compute solution, input vectors are empty!" << std::endl;
            return false;
        }
        
        if( x.getSize() != y.getSize() ){
            warningLog << __GRT_LOG__ << " Failed to compute solution, input vectors do not have the same size!" << std::endl;
            return false;
        }
        
        const unsigned int N = x.getSize();
        Float sumx = 0.0;    //Stores the sum of x
        Float sumx2 = 0.0;   //Stores the sum of x^2
        Float sumxy = 0.0;   //Stores the sum of x * y
        Float sumy = 0.0;    //Stores the sum of y
        Float sumy2 = 0.0;   //Stores the sum of y**2
        Float denom = 0;
        
        m = 0;
        b = 0;
        r = 0;
        
        for(unsigned int i=0; i<N; i++){
            sumx  += x[i];
            sumx2 += SQR(x[i]);
            sumxy += x[i] * y[i];
            sumy  += y[i];
            sumy2 += SQR(y[i]);
        }
        
        denom = (N * sumx2 - SQR(sumx));
        if (denom == 0) {
            // singular matrix. can't solve the problem.
            warningLog << __GRT_LOG__ << " Failed to compute solution, singular matrix detected!" << std::endl;
            return false;
        }
        
        m = (N*sumxy  -  sumx*sumy) / denom;
        b = (sumy*sumx2  -  sumx*sumxy) / denom;
        
        //compute correlation coeff
        Float norm = 1.0/N;
        sumx *= norm;
        sumy *= norm;
        sumxy *= norm;
        sumx2 *= norm;
        sumy2 *= norm;
        r = (sumxy - sumx*sumy) / sqrt( (sumx2 - SQR(sumx)) * (sumy2 - SQR(sumy)) );

        trained = true;
        converged = true;
        
        return true;
    }

    Float getM() const { return m; }

    Float getB() const { return b; }

    Float getR() const { return r; }
    
protected:

    Float m;
    Float b;
    Float r;
    
};

GRT_END_NAMESPACE

#endif //GRT_MEAN_SHIFT_HEADER