itkQuasiNewtonLBFGSOptimizer.h

Go to the documentation of this file.

/*======================================================================

  This file is part of the elastix software.

  Copyright (c) University Medical Center Utrecht. All rights reserved.
  See src/CopyrightElastix.txt or http://elastix.isi.uu.nl/legal.php for
  details.

     This software is distributed WITHOUT ANY WARRANTY; without even
     the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
     PURPOSE. See the above copyright notices for more information.

======================================================================*/

#ifndef __itkQuasiNewtonLBFGSOptimizer_h
#define __itkQuasiNewtonLBFGSOptimizer_h

#include "itkScaledSingleValuedNonLinearOptimizer.h"
#include "itkLineSearchOptimizer.h"
#include <vector>

namespace itk
{
  class QuasiNewtonLBFGSOptimizer : public ScaledSingleValuedNonLinearOptimizer
  {
  public:

    typedef QuasiNewtonLBFGSOptimizer             Self;
    typedef ScaledSingleValuedNonLinearOptimizer  Superclass;
    typedef SmartPointer<Self>                    Pointer;
    typedef SmartPointer<const Self>              ConstPointer;

    itkNewMacro(Self);
    itkTypeMacro(QuasiNewtonLBFGSOptimizer, ScaledSingleValuedNonLinearOptimizer);

    typedef Superclass::ParametersType            ParametersType;
    typedef Superclass::DerivativeType            DerivativeType;
    typedef Superclass::CostFunctionType          CostFunctionType;
    typedef Superclass::ScaledCostFunctionType    ScaledCostFunctionType;
    typedef Superclass::MeasureType               MeasureType;
    typedef Superclass::ScalesType                ScalesType;

    typedef Array<double>                         RhoType;
    typedef std::vector<ParametersType>           SType;
    typedef std::vector<DerivativeType>           YType;
    typedef Array<double>                         DiagonalMatrixType;
    typedef LineSearchOptimizer                   LineSearchOptimizerType;

    typedef LineSearchOptimizerType::Pointer      LineSearchOptimizerPointer;

    typedef enum {
      MetricError,
      LineSearchError,
      MaximumNumberOfIterations,
      InvalidDiagonalMatrix,
      GradientMagnitudeTolerance,
      ZeroStep,
      Unknown }                                   StopConditionType;

    virtual void StartOptimization(void);
    virtual void ResumeOptimization(void);
    virtual void StopOptimization(void);

    itkGetConstMacro(CurrentIteration, unsigned long);
    itkGetConstMacro(CurrentValue, MeasureType);
    itkGetConstReferenceMacro(CurrentGradient, DerivativeType);
    itkGetConstMacro(InLineSearch, bool);
    itkGetConstReferenceMacro(StopCondition, StopConditionType);
    itkGetConstMacro(CurrentStepLength, double);

    itkSetObjectMacro(LineSearchOptimizer, LineSearchOptimizerType);
    itkGetObjectMacro(LineSearchOptimizer, LineSearchOptimizerType);

    itkGetConstMacro(MaximumNumberOfIterations, unsigned long);
    itkSetClampMacro(MaximumNumberOfIterations, unsigned long,
      1, NumericTraits<unsigned long>::max());

    itkGetConstMacro(GradientMagnitudeTolerance, double);
    itkSetMacro(GradientMagnitudeTolerance, double);

    itkSetClampMacro(Memory,unsigned int,0,NumericTraits<unsigned int>::max());
    itkGetConstMacro(Memory,unsigned int);


  protected:
    QuasiNewtonLBFGSOptimizer();
    virtual ~QuasiNewtonLBFGSOptimizer(){};

    void PrintSelf(std::ostream& os, Indent indent) const {};

    DerivativeType                m_CurrentGradient;
    MeasureType                   m_CurrentValue;
    unsigned long                 m_CurrentIteration;
    StopConditionType             m_StopCondition;
    bool                          m_Stop;
    double                        m_CurrentStepLength;

    bool                          m_InLineSearch;

    RhoType                       m_Rho;
    SType                         m_S;
    YType                         m_Y;

    unsigned int                  m_Point;
    unsigned int                  m_PreviousPoint;
    unsigned int                  m_Bound;

    itkSetMacro(InLineSearch, bool);

    virtual void ComputeDiagonalMatrix(DiagonalMatrixType & diag_H0);

    virtual void ComputeSearchDirection(
      const DerivativeType & gradient,
      ParametersType & searchDir);

    virtual void LineSearch(
      const ParametersType searchDir,
      double & step,
      ParametersType & x,
      MeasureType & f,
      DerivativeType & g );

    virtual void StoreCurrentPoint(
      const ParametersType & step,
      const DerivativeType & grad_dif);

    virtual bool TestConvergence(bool firstLineSearchDone);

  private:
    QuasiNewtonLBFGSOptimizer(const Self&); //purposely not implemented
    void operator=(const Self&); //purposely not implemented

    unsigned long                 m_MaximumNumberOfIterations;
    double                        m_GradientMagnitudeTolerance;
    LineSearchOptimizerPointer    m_LineSearchOptimizer;
    unsigned int                  m_Memory;


  }; // end class QuasiNewtonLBFGSOptimizer



} // end namespace itk

/*        LIMITED MEMORY BFGS METHOD FOR LARGE SCALE OPTIMIZATION */
/*                          JORGE NOCEDAL */
/*                        *** July 1990 *** */

/*     This subroutine solves the unconstrained minimization problem */

/*                      min F(x),    x= (x1,x2,...,xN), */

/*      using the limited memory BFGS method. The routine is especially */
/*      effective on problems involving a large number of variables. In */
/*      a typical iteration of this method an approximation Hk to the */
/*      inverse of the Hessian is obtained by applying M BFGS updates to */
/*      a diagonal matrix Hk0, using information from the previous M steps.  */
/*      The user specifies the number M, which determines the amount of */
/*      storage required by the routine. The user may also provide the */
/*      diagonal matrices Hk0 if not satisfied with the default choice. */
/*      The algorithm is described in "On the limited memory BFGS method */
/*      for large scale optimization", by D. Liu and J. Nocedal, */
/*      Mathematical Programming B 45 (1989) 503-528. */

/*      The user is required to calculate the function value F and its */
/*      gradient G. In order to allow the user complete control over */
/*      these computations, reverse  communication is used. The routine */
/*      must be called repeatedly under the control of the parameter */
/*      IFLAG. */

/*      The steplength is determined at each iteration by means of the */
/*      line search routine MCVSRCH, which is a slight modification of */
/*      the routine CSRCH written by More' and Thuente. */

/*      The calling statement is */

/*          CALL LBFGS(N,M,X,F,G,DIAGCO,DIAG,IPRINT,EPS,XTOL,W,IFLAG) */

/*      where */

/*     N       is an INTEGER variable that must be set by the user to the */
/*             number of variables. It is not altered by the routine. */
/*             Restriction: N>0. */

/*     M       is an INTEGER variable that must be set by the user to */
/*             the number of corrections used in the BFGS update. It */
/*             is not altered by the routine. Values of M less than 3 are */
/*             not recommended; large values of M will result in excessive */
/*             computing time. 3<= M <=7 is recommended. Restriction: M>0.  */

/*     X       is a DOUBLE PRECISION array of length N. On initial entry */
/*             it must be set by the user to the values of the initial */
/*             estimate of the solution vector. On exit with IFLAG=0, it */
/*             contains the values of the variables at the best point */
/*             found (usually a solution). */

/*     F       is a DOUBLE PRECISION variable. Before initial entry and on */
/*             a re-entry with IFLAG=1, it must be set by the user to */
/*             contain the value of the function F at the point X. */

/*     G       is a DOUBLE PRECISION array of length N. Before initial */
/*             entry and on a re-entry with IFLAG=1, it must be set by */
/*             the user to contain the components of the gradient G at */
/*             the point X. */

/*     DIAGCO  is a LOGICAL variable that must be set to .TRUE. if the */
/*             user  wishes to provide the diagonal matrix Hk0 at each */
/*             iteration. Otherwise it should be set to .FALSE., in which */
/*             case  LBFGS will use a default value described below. If */
/*             DIAGCO is set to .TRUE. the routine will return at each */
/*             iteration of the algorithm with IFLAG=2, and the diagonal */
/*              matrix Hk0  must be provided in the array DIAG. */

/*     DIAG    is a DOUBLE PRECISION array of length N. If DIAGCO=.TRUE., */
/*             then on initial entry or on re-entry with IFLAG=2, DIAG */
/*             it must be set by the user to contain the values of the */
/*             diagonal matrix Hk0.  Restriction: all elements of DIAG */
/*             must be positive. */

/*     IPRINT  is an INTEGER array of length two which must be set by the */
/*             user. */

/*             IPRINT(1) specifies the frequency of the output: */
/*                IPRINT(1) < 0 : no output is generated, */
/*                IPRINT(1) = 0 : output only at first and last iteration, */
/*                IPRINT(1) > 0 : output every IPRINT(1) iterations. */

/*             IPRINT(2) specifies the type of output generated: */
/*                IPRINT(2) = 0 : iteration count, number of function */
/*                                evaluations, function value, norm of the */
/*                                gradient, and steplength, */
/*                IPRINT(2) = 1 : same as IPRINT(2)=0, plus vector of */
/*                                variables and  gradient vector at the */
/*                                initial point, */
/*                IPRINT(2) = 2 : same as IPRINT(2)=1, plus vector of */
/*                                variables, */
/*                IPRINT(2) = 3 : same as IPRINT(2)=2, plus gradient vector.*/


/*    EPS     is a positive DOUBLE PRECISION variable that must be set by */
/*            the user, and determines the accuracy with which the solution*/
/*            is to be found. The subroutine terminates when */

/*                         ||G|| < EPS max(1,||X||), */

/*            where ||.|| denotes the Euclidean norm. */

/*    XTOL    is a  positive DOUBLE PRECISION variable that must be set by */
/*            the user to an estimate of the machine precision (e.g. */
/*            10**(-16) on a SUN station 3/60). The line search routine will*/
/*            terminate if the relative width of the interval of uncertainty*/
/*            is less than XTOL. */

/*     W       is a DOUBLE PRECISION array of length N(2M+1)+2M used as */
/*             workspace for LBFGS. This array must not be altered by the */
/*             user. */

/*    IFLAG   is an INTEGER variable that must be set to 0 on initial entry*/
/*            to the subroutine. A return with IFLAG<0 indicates an error, */
/*            and IFLAG=0 indicates that the routine has terminated without*/
/*            detecting errors. On a return with IFLAG=1, the user must */
/*            evaluate the function F and gradient G. On a return with */
/*            IFLAG=2, the user must provide the diagonal matrix Hk0. */

/*            The following negative values of IFLAG, detecting an error, */
/*            are possible: */

/*              IFLAG=-1  The line search routine MCSRCH failed. The */
/*                        parameter INFO provides more detailed information */
/*                        (see also the documentation of MCSRCH): */

/*                       INFO = 0  IMPROPER INPUT PARAMETERS. */

/*                       INFO = 2  RELATIVE WIDTH OF THE INTERVAL OF */
/*                                 UNCERTAINTY IS AT MOST XTOL. */

/*                       INFO = 3  MORE THAN 20 FUNCTION EVALUATIONS WERE */
/*                                 REQUIRED AT THE PRESENT ITERATION. */

/*                       INFO = 4  THE STEP IS TOO SMALL. */

/*                       INFO = 5  THE STEP IS TOO LARGE. */

/*                       INFO = 6  ROUNDING ERRORS PREVENT FURTHER PROGRESS.*/
/*                                 THERE MAY NOT BE A STEP WHICH SATISFIES */
/*                                 THE SUFFICIENT DECREASE AND CURVATURE */
/*                                 CONDITIONS. TOLERANCES MAY BE TOO SMALL.  */

/*             IFLAG=-2  The i-th diagonal element of the diagonal inverse */
/*                       Hessian approximation, given in DIAG, is not */
/*                       positive. */

/*             IFLAG=-3  Improper input parameters for LBFGS (N or M are */
/*                       not positive). */


/*    ON THE DRIVER: */

/*    The program that calls LBFGS must contain the declaration: */

/*                       EXTERNAL LB2 */

/*    LB2 is a BLOCK DATA that defines the default values of several */
/*    parameters described in the COMMON section. */

/*    COMMON: */

/*     The subroutine contains one common area, which the user may wish to */
/*    reference: */

/* awf added stpawf */

/*    MP  is an INTEGER variable with default value 6. It is used as the */
/*        unit number for the printing of the monitoring information */
/*        controlled by IPRINT. */

/*    LP  is an INTEGER variable with default value 6. It is used as the */
/*        unit number for the printing of error messages. This printing */
/*        may be suppressed by setting LP to a non-positive value. */

/*    GTOL is a DOUBLE PRECISION variable with default value 0.9, which */
/*        controls the accuracy of the line search routine MCSRCH. If the */
/*        function and gradient evaluations are inexpensive with respect */
/*        to the cost of the iteration (which is sometimes the case when */
/*        solving very large problems) it may be advantageous to set GTOL */
/*        to a small value. A typical small value is 0.1.  Restriction: */
/*        GTOL should be greater than 1.D-04. */

/*    STPMIN and STPMAX are non-negative DOUBLE PRECISION variables which */
/*        specify lower and uper bounds for the step in the line search.  */
/*        Their default values are 1.D-20 and 1.D+20, respectively. These */
/*        values need not be modified unless the exponents are too large */
/*        for the machine being used, or unless the problem is extremely */
/*        badly scaled (in which case the exponents should be increased).  */


/*  MACHINE DEPENDENCIES */

/*        The only variables that are machine-dependent are XTOL, */
/*        STPMIN and STPMAX. */


/*  GENERAL INFORMATION */

/*    Other routines called directly:  DAXPY, DDOT, LB1, MCSRCH */

/*    Input/Output  :  No input; diagnostic messages on unit MP and */
/*                     error messages on unit LP. */


/*    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -*/


#endif //#ifndef __itkQuasiNewtonLBFGSOptimizer_h