/*=========================================================================
 *
 *  Copyright NumFOCUS
 *
 *  Licensed under the Apache License, Version 2.0 (the "License");
 *  you may not use this file except in compliance with the License.
 *  You may obtain a copy of the License at
 *
 *         https://www.apache.org/licenses/LICENSE-2.0.txt
 *
 *  Unless required by applicable law or agreed to in writing, software
 *  distributed under the License is distributed on an "AS IS" BASIS,
 *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 *  See the License for the specific language governing permissions and
 *  limitations under the License.
 *
 *=========================================================================*/
#ifndef itkLBFGSBOptimizer_h
#define itkLBFGSBOptimizer_h

#include "itkIntTypes.h"
#include "itkSingleValuedNonLinearVnlOptimizer.h"
#include "ITKOptimizersExport.h"
#include <memory> // For unique_ptr.

namespace itk
{
/* Necessary forward declaration see below for definition */
/** \class LBFGSBOptimizerHelper
 * \brief Wrapper helper around vnl_lbfgsb.
 *
 * This class is used to translate iteration events, etc, from
 * vnl_lbfgsb into iteration events in ITK.
 * \ingroup ITKOptimizers
 */
class ITK_FORWARD_EXPORT LBFGSBOptimizerHelper;

/** \class LBFGSBOptimizer
 * \brief Limited memory Broyden Fletcher Goldfarb Shannon minimization with simple bounds.
 *
 * This class is a wrapper for converted Fortran code for performing limited
 * memory Broyden Fletcher Goldfarb Shannon minimization with simple bounds.
 * The algorithm minimizes a nonlinear function f(x) of n variables subject to
 * simple bound constraints of l <= x <= u.
 *
 * See also the documentation in Numerics/lbfgsb.c
 *
 * References:
 *
 * [1] R. H. Byrd, P. Lu and J. Nocedal.
 * A Limited Memory Algorithm for Bound Constrained Optimization, (1995),
 * SIAM Journal on Scientific and Statistical Computing ,
 * 16, 5, pp. 1190-1208.
 *
 * [2] C. Zhu, R. H. Byrd and J. Nocedal.
 * L-BFGS-B: Algorithm 778: L-BFGS-B, FORTRAN routines for large scale
 * bound constrained optimization (1997),
 * ACM Transactions on Mathematical Software,
 * Vol 23, Num. 4, pp. 550 - 560.
 *
 * \ingroup Numerics Optimizers
 * \ingroup ITKOptimizers
 */
class ITKOptimizers_EXPORT LBFGSBOptimizer : public SingleValuedNonLinearVnlOptimizer
{
public:
  ITK_DISALLOW_COPY_AND_MOVE(LBFGSBOptimizer);

  /** Standard "Self" type alias. */
  using Self = LBFGSBOptimizer;
  using Superclass = SingleValuedNonLinearVnlOptimizer;
  using Pointer = SmartPointer<Self>;
  using ConstPointer = SmartPointer<const Self>;

  /** Method for creation through the object factory. */
  itkNewMacro(Self);

  /** \see LightObject::GetNameOfClass() */
  itkOverrideGetNameOfClassMacro(LBFGSBOptimizer);

  /**  BoundValue type.
   *  Use for defining the lower and upper bounds on the variables.
   */
  using BoundValueType = Array<double>;

  /** BoundSelection type
   * Use for defining the boundary condition for each variables.
   */
  using BoundSelectionType = Array<long>;

  /** Internal boundary value storage type */
  using InternalBoundValueType = vnl_vector<double>;

  /** Internal boundary selection storage type */
  using InternalBoundSelectionType = vnl_vector<long>;

  /** The vnl optimizer */
  using InternalOptimizerType = LBFGSBOptimizerHelper;

  /** Start optimization with an initial value. */
  void
  StartOptimization() override;

  /** Plug in a Cost Function into the optimizer  */
  void
  SetCostFunction(SingleValuedCostFunction * costFunction) override;

  /** Set/Get the optimizer trace flag. If set to true, the optimizer
   * prints out information every iteration.
   */
  virtual void
  SetTrace(bool flag);

  itkGetMacro(Trace, bool);
  itkBooleanMacro(Trace);

  /** Set the lower bound value for each variable. */
  virtual void
  SetLowerBound(const BoundValueType & value);
  itkGetConstReferenceMacro(LowerBound, BoundValueType);

  /** Set the upper bound value for each variable. */
  virtual void
  SetUpperBound(const BoundValueType & value);
  itkGetConstReferenceMacro(UpperBound, BoundValueType);

  /** Set the boundary condition for each variable, where
   * select[i] = 0 if x[i] is unbounded,
   *           = 1 if x[i] has only a lower bound,
   *           = 2 if x[i] has both lower and upper bounds, and
   *           = 3 if x[1] has only an upper bound
   */
  virtual void
  SetBoundSelection(const BoundSelectionType & value);
  itkGetConstReferenceMacro(BoundSelection, BoundSelectionType);

  /** Set/Get the CostFunctionConvergenceFactor. Algorithm terminates
   * when the reduction in cost function is less than factor * epsmcj
   * where epsmch is the machine precision.
   * Typical values for factor: 1e+12 for low accuracy;
   * 1e+7 for moderate accuracy and 1e+1 for extremely high accuracy.
   */
  virtual void
  SetCostFunctionConvergenceFactor(double);

  itkGetMacro(CostFunctionConvergenceFactor, double);

  /** Set/Get the ProjectedGradientTolerance. Algorithm terminates
   * when the project gradient is below the tolerance. Default value
   * is 1e-5.
   */
  virtual void
  SetProjectedGradientTolerance(double);

  itkGetMacro(ProjectedGradientTolerance, double);

  /** Set/Get the MaximumNumberOfIterations. Default is 500 */
  virtual void
  SetMaximumNumberOfIterations(unsigned int);

  itkGetMacro(MaximumNumberOfIterations, unsigned int);

  /** Set/Get the MaximumNumberOfEvaluations. Default is 500 */
  virtual void
  SetMaximumNumberOfEvaluations(unsigned int);

  itkGetMacro(MaximumNumberOfEvaluations, unsigned int);

  /** Set/Get the MaximumNumberOfCorrections. Default is 5 */
  virtual void
  SetMaximumNumberOfCorrections(unsigned int);

  itkGetMacro(MaximumNumberOfCorrections, unsigned int);

  /** This optimizer does not support scaling of the derivatives. */
  void
  SetScales(const ScalesType &)
  {
    itkExceptionMacro("This optimizer does not support scales.");
  }

  /** Get the current iteration number. */
  itkGetConstReferenceMacro(CurrentIteration, unsigned int);

  /** Get the current cost function value. */
  MeasureType
  GetValue() const;

  /** Get the current infinity norm of the project gradient of the cost
   * function. */
  itkGetConstReferenceMacro(InfinityNormOfProjectedGradient, double);

  /** Get the reason for termination */
  const std::string
  GetStopConditionDescription() const override;

  /** Returns false unconditionally because LBFGSBOptimizer does not support using scales. */
  bool
  CanUseScales() const override
  {
    return false;
  }

protected:
  LBFGSBOptimizer();
  ~LBFGSBOptimizer() override;
  void
  PrintSelf(std::ostream & os, Indent indent) const override;

  using CostFunctionAdaptorType = Superclass::CostFunctionAdaptorType;

private:
  // give the helper access to member variables, to update iteration
  // counts, etc.
  friend class LBFGSBOptimizerHelper;

  bool         m_Trace{ false };
  bool         m_OptimizerInitialized{ false };
  double       m_CostFunctionConvergenceFactor{ 1e+7 };
  double       m_ProjectedGradientTolerance{ 1e-5 };
  unsigned int m_MaximumNumberOfIterations{ 500 };
  unsigned int m_MaximumNumberOfEvaluations{ 500 };
  unsigned int m_MaximumNumberOfCorrections{ 5 };
  unsigned int m_CurrentIteration{ 0 };
  double       m_InfinityNormOfProjectedGradient{ 0.0 };

  std::unique_ptr<InternalOptimizerType> m_VnlOptimizer;
  BoundValueType                         m_LowerBound{};
  BoundValueType                         m_UpperBound{};
  BoundSelectionType                     m_BoundSelection{};
};
} // end namespace itk

#endif