misfit.3.man

来自「COOOL：CWP面向对象最优化库（CWP Object Oriented Op」· MAN 代码 · 共 95 行

MISFITFCN(derived)  OBJECTIVE FUNCTIONS   MISFITFCN(derived)

                                                     Jun  1 15:19
NAME
    MisFitFcn

SYNOPSIS
    #include <MisFit.hh>

    class MisFitFcn: public ObjectiveFunction

        \fIPublic members\fP
            MisFitFcn();
            MisFitFcn(int, Forward*, Vector<double>*, int);
            ~MisFitFcn();
            int	 getOrder();
            int	 experiments();
            const char* 	className() const;
            const char* 	forwardName() const;
            Vector<double>*	getGradient(Model<double>& m);
            Vector<double>*	getGradient(Model<long>& m);
            Vector<double>	updateError(Model<double>& m);
            Vector<double>	updateError(Model<long>& m);
            Vector<double>	observedData();
            Vector<double>	currentError();
            Vector<double>	operateOn(Vector<double>&);
            Vector<double>	adjointOperation(Vector<double> &);

        \fIProtected members\fP
            double 		realPerformance(Model<long>&);
            double 		realPerformance(Model<double>&);

        \fIPrivate members\fP
            Forward*		forOp;
            Vector<double>*	refData;
            Vector<double>*	errData;
            int			lp;
            int 		ndata;
            int			isUpdated;
            Vector<double> 	modeling(Model<double>&);
            Vector<double> 	modeling(Model<long>&);

DESCRIPTION
    MisFitFcn
     This is a derived class of objective function which measures
     the least-square error between the observed and calculated 
     data set. The calculated data set are passed by the Forward
     operator which needs to be defined when constructing the
     MisFitFcn class.
     See also Forward, LinForward, ObjFcn.
    

DESCRIPTION
    Constructor MisFitFcn(n, *p, *d, lp):
       int n:  number of dimensions of the Model space;
       *p: pointer to a Forward operator class;
       *d: pointer to the data Vector;
       int lp: lp norm. lp=2, least-square criterian;
    Fetching information of the MisFitFcn:
       int getOrder(): returns lp;
       int experiments(): returns number of iterations;
       char* className(): returns name of the current class;
       char* forwardName(): returns name of the Forward class;
       observedData(): returns the observed data Vector;
       currentError(): returns the error Vector between observed
          and calculated;
    
    Other operations:
      getGradient(m): evaluate gradient Vector for Model m;
      updateError(m): evaluate lp-norm error for Model m;
      operateOn(v): Forward modeling on Vector v;
      adjointOperation(v): adjoint of the Forward operator 
         operates on Vector v; mostly used when the Forward
         is a LinForward.
    

CAVEATS
     MisFitFcn currently only takes Models either double or 
     long integer. It is mostly because I tried to avoid using
     templates, due to the guliness of template features in
     G++. Hopefully, this could be changed soon.

DEFINED MACROS
    MISFIT_FCN_HH

INCLUDED FILES
    <Vector.hh>
    "Forward.hh"
    "ObjFcn.hh"

SOURCE FILES
    MisFit.cc