gpNonStatReg.hpp
Go to the documentation of this file.
00001 #ifndef _GP_HET_REGRESSION_HPP_
00002 #define _GP_HET_REGRESSION_HPP_
00003 
00004 #include "gpReg/types.hpp"
00005 #include "gpReg/global.h"
00006 #include "gpRegression.hpp"
00007 
00008 #include <assert.h>
00009 #include "covarianceFunction.hpp"
00010 #ifdef DO_PROFILING
00011   #include "profiler.hpp"
00012 #endif
00013 
00014 
00015 
00016 // -----------------------------------------------------------------------
00017 class CovNonStatSE : public CovFunc<TDoubleVector>
00018 {
00019   public:
00020 
00021     CovNonStatSE( GPReg<TDoubleVector> &ellGP, double ellMean, double sf2 ) : 
00022         m_ellGP(ellGP), m_ellMean(ellMean), m_sf2(sf2)
00023     {
00024       m_ell = 0.13;
00025     }
00026     //Cov1dSEPolar( ConfigFile &cfg, const char *cfgSection );
00027 
00028     virtual double getCov( const TDoubleVector &x, const TDoubleVector &y )
00029     {
00030       double dist = fabs(x[0]-y[0]);
00032       double covBlock = 0.0;
00033       double a, b;
00034       if (x[0]<y[0]) {
00035         a = x[0];
00036         b = y[0];
00037       } else {
00038         a = y[0];
00039         b = x[0];
00040       }
00041       double temp;
00042       TDoubleVector z(1);
00043       for (z[0]=a; z[0]<=b; z[0]+=0.01) {
00044         m_ellGP.evalGP( z, temp );
00045         temp += m_ellMean;
00046         covBlock += 1.0 / (1.0+exp(-temp));
00047       }
00048       double f = (1.0-covBlock/0.01);
00049       if (f<0) f=0.0;
00050       //*/
00051       
00052       /*
00053       double covBlock = 1.0;
00054       double f = 1.0;
00055       double blockPoint = 0.5;
00056       if ( (x[0]<blockPoint && y[0]>blockPoint) || (x[0]>blockPoint && y[0]<blockPoint) ) {
00057         f=0.0;
00058       }
00059       */
00060 
00061       double cov = (m_sf2*m_sf2) * f * exp( -0.5 * pow(dist,2.0)/m_ell ); 
00062 
00063       //fprintf( stderr, "[%f %f] (%f %f) %f\n", x[0], y[0], ell_x, ell_y, cov );
00064       FILE *covFile = fopen( "cov.dat", "a" );
00065       fprintf( covFile, "%f %f %f %f %f\n", x[0], y[0], cov, covBlock, f );
00066       fclose( covFile );
00067       return cov; 
00068       //return (m_sf2*m_sf2) * f1 * f2 * f3 * exp( -0.5 * dist*dist/(ell_x-ell_y) ); 
00069     }
00070 
00071     // paciorek 1d
00072     /*
00073     virtual double getCov( const TDoubleVector &x, const TDoubleVector &y )
00074     {
00075       double ell_x, ell_y;
00076       m_ellGP.evalGP( x, ell_x );
00077       m_ellGP.evalGP( y, ell_y );
00078       double dist = x[0]-y[0];
00079       double f1 = pow( ell_x, 0.25 );
00080       double f2 = pow( ell_y, 0.25 );
00081       double f3 = pow( (ell_x+ell_y)/2.0, -0.5 );
00082 
00083       double cov = (m_sf2*m_sf2) * f1 * f2 * f3 * exp( -0.5 * pow(dist,2.0)/((ell_x+ell_y)/2.0) ); 
00084       //double cov = (m_sf2*m_sf2) * exp( -0.5 * pow(dist,2.0)/((ell_x+ell_y)/2.0) ); 
00085 
00086       //fprintf( stderr, "[%f %f] (%f %f) %f\n", x[0], y[0], ell_x, ell_y, cov );
00087       FILE *covFile = fopen( "cov.dat", "a" );
00088       fprintf( covFile, "%f %f %f %f %f %f\n", x[0], y[0], ell_x, ell_y, f1*f2*f3, cov );
00089       fclose( covFile );
00090       return cov;
00091       //return (m_sf2*m_sf2) * f1 * f2 * f3 * exp( -0.5 * dist*dist/(ell_x-ell_y) ); 
00092     }
00093     */
00094 
00095 
00096 
00097   private:
00098     ConfigFile m_cfg;
00099   
00100   public:
00101     GPReg<TDoubleVector> &m_ellGP;
00102     double m_ellMean;
00103     double m_ell;
00104     double m_sf2;
00105 };
00106 
00107 
00108 
00109 // -----------------------------------------------------------------------
00110 template <class TInput>
00111 class GPNonStatReg
00112 {
00113   
00114   public:
00115   
00116 
00117  
00118     // -----------------------------------------------------------------------
00119     GPNonStatReg(
00120         double ell, double ellMean, double sf2, double noise ) :
00121       m_numDataPoints(0), m_dataPoints(0), m_targets(0),
00122       m_ellPriorCovFunc(ell,0.0,sf2), m_ellPriorGP(m_ellPriorCovFunc,noise),
00123       m_covFunc(m_ellPriorGP,ellMean,sf2), m_dataGP(m_covFunc,noise)
00124     {}
00125 
00126 
00127 
00128     // -----------------------------------------------------------------------
00129     ~GPNonStatReg()
00130     {}
00131 
00132 
00133 
00134     // -----------------------------------------------------------------------
00135     void setDataPoints( TVector<TInput> &dataPoints, TVector<double> &targets, TVector<double> &ellTargets )
00136     {
00137       m_dataPoints = dataPoints;
00138       m_numDataPoints = dataPoints.size();
00139       m_targets = targets;
00140       m_ellTargets = ellTargets;
00141     }      
00142 
00143 
00144 
00145     // -----------------------------------------------------------------------
00146     void buildGP()
00147     {
00148       // dummy ellPrior (all targets = ell)
00149       m_ellPriorGP.setDataPoints( m_dataPoints, m_ellTargets );
00150       m_ellPriorGP.buildGP();
00151       
00152       m_dataGP.setDataPoints( m_dataPoints, m_targets );
00153       m_dataGP.buildGP();
00154     }
00155 
00156 
00157 
00158     // -----------------------------------------------------------------------
00159     void evalGP( const TInput &x, double &mean, double &var )
00160     {
00161       m_dataGP.evalGP( x, mean, var );
00162     }
00163 
00164 
00165 
00166     // -----------------------------------------------------------------------
00167     void evalGP( const TInput &x, double &mean )
00168     {
00169       m_dataGP.evalGP( x, mean );
00170     }
00171 
00172 
00173 
00174   // -----------------------------------------------------------------------
00175   public:
00176     int m_numDataPoints;
00177     TVector<TInput> m_dataPoints;
00178     TVector<double> m_targets;
00179 
00180     Cov1dSE         m_ellPriorCovFunc;
00181     GPReg<TInput>   m_ellPriorGP;
00182     CovNonStatSE    m_covFunc;
00183     GPReg<TInput>   m_dataGP;
00184 
00185     TVector<double> m_ellTargets;
00186     double          m_noise;
00187 
00188 };
00189 
00190 
00191 
00192 #endif //_GP_HET_REGRESSION_HPP_


gaussian_process
Author(s): Maintained by Juergen Sturm
autogenerated on Mon Oct 6 2014 00:09:34