/*=========================================================================
 *
 *  Copyright UMC Utrecht and contributors
 *
 *  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
 *
 *        http://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 elxAdaGrad_hxx
#define elxAdaGrad_hxx

#include "elxAdaGrad.h"
#include <itkDeref.h>

#include <cmath> // For abs.
#include <iomanip>
#include <string>
#include <vector>
#include <sstream>
#include <algorithm>
#include <utility>
#include "itkAdvancedImageToImageMetric.h"
#include "itkTimeProbe.h"

namespace elastix
{

/**
 * ********************** Constructor ***********************
 */

template <class TElastix>
AdaGrad<TElastix>::AdaGrad()
{
  this->m_MaximumNumberOfSamplingAttempts = 0;
  this->m_CurrentNumberOfSamplingAttempts = 0;
  this->m_PreviousErrorAtIteration = 0;
  this->m_AutomaticParameterEstimationDone = false;

  this->m_AutomaticParameterEstimation = false;
  this->m_MaximumStepLength = 1.0;
  this->m_MaximumStepLengthRatio = 1.0;
  this->m_RegularizationKappa = 0.8;
  this->m_ConditionNumber = 2.0;
  this->m_NoiseFactor = 1.0;

  this->m_NumberOfGradientMeasurements = 0;
  this->m_NumberOfJacobianMeasurements = 0;
  this->m_NumberOfSamplesForPrecondition = 0;
  this->m_NumberOfSamplesForNoiseCompensationFactor = 0;
  this->m_NumberOfSpatialSamples = 5000;
  this->m_SigmoidScaleFactor = 0.1;
  this->m_GlobalStepSize = 0;

  this->m_RandomGenerator = RandomGeneratorType::GetInstance();
  this->m_AdvancedTransform = nullptr;

  this->m_UseNoiseCompensation = true;

} // Constructor


/**
 * ***************** BeforeRegistration ***********************
 */

template <class TElastix>
void
AdaGrad<TElastix>::BeforeRegistration()
{
  /** Add the target cell "stepsize" to IterationInfo. */
  this->AddTargetCellToIterationInfo("2:Metric");
  this->AddTargetCellToIterationInfo("3a:Time");
  this->AddTargetCellToIterationInfo("3b:StepSize");
  this->AddTargetCellToIterationInfo("4a:||Gradient||");
  this->AddTargetCellToIterationInfo("4b:||SearchDirection||");

  /** Format the metric and stepsize as floats. */
  this->GetIterationInfoAt("2:Metric") << std::showpoint << std::fixed;
  this->GetIterationInfoAt("3a:Time") << std::showpoint << std::fixed;
  this->GetIterationInfoAt("3b:StepSize") << std::showpoint << std::fixed;
  this->GetIterationInfoAt("4:||Gradient||") << std::showpoint << std::fixed;
  this->GetIterationInfoAt("4b:||SearchDirection||") << std::showpoint << std::fixed;

  this->m_SettingsVector.clear();

} // end BeforeRegistration()


/**
 * ***************** BeforeEachResolution ***********************
 */

template <class TElastix>
void
AdaGrad<TElastix>::BeforeEachResolution()
{
  /** Get the current resolution level. */
  unsigned int level = static_cast<unsigned int>(this->m_Registration->GetAsITKBaseType()->GetCurrentLevel());

  const unsigned int P = this->GetElastix()->GetElxTransformBase()->GetAsITKBaseType()->GetNumberOfParameters();

  const Configuration & configuration = itk::Deref(Superclass2::GetConfiguration());

  /** Set the maximumNumberOfIterations. */
  SizeValueType maximumNumberOfIterations = 500;
  configuration.ReadParameter(
    maximumNumberOfIterations, "MaximumNumberOfIterations", this->GetComponentLabel(), level, 0);
  this->SetNumberOfIterations(maximumNumberOfIterations);

  /** Set the gain parameter A. */
  double A = 20.0;
  configuration.ReadParameter(A, "SP_A", this->GetComponentLabel(), level, 0);
  this->SetParam_A(A);

  /** Set the MaximumNumberOfSamplingAttempts. check if needed? */
  SizeValueType maximumNumberOfSamplingAttempts = 0;
  configuration.ReadParameter(
    maximumNumberOfSamplingAttempts, "MaximumNumberOfSamplingAttempts", this->GetComponentLabel(), level, 0);
  this->SetMaximumNumberOfSamplingAttempts(maximumNumberOfSamplingAttempts);
  if (maximumNumberOfSamplingAttempts > 5)
  {
    log::warn(
      std::ostringstream{} << "\nWARNING: You have set MaximumNumberOfSamplingAttempts to "
                           << maximumNumberOfSamplingAttempts << ".\n"
                           << "  This functionality is known to cause problems (stack overflow) for large values.\n"
                           << "  If elastix stops or segfaults for no obvious reason, reduce this value.\n"
                           << "  You may select the RandomSparseMask image sampler to fix mask-related problems.\n");
  }

  /** Set/Get the initial time. Default: 0.0. Should be >= 0. */
  double initialTime = 0.0;
  configuration.ReadParameter(initialTime, "SigmoidInitialTime", this->GetComponentLabel(), level, 0);
  this->SetInitialTime(initialTime);

  /** Set/Get whether the adaptive step size mechanism is desired. Default: true
   * NB: the setting is turned off in case of UseRandomSampleRegion == true.
   */
  /** Set whether automatic gain estimation is required; default: true. */
  this->m_AutomaticParameterEstimation = true;
  configuration.ReadParameter(
    this->m_AutomaticParameterEstimation, "AutomaticParameterEstimation", this->GetComponentLabel(), level, 0);

  std::string stepSizeStrategy = "Adaptive";
  configuration.ReadParameter(stepSizeStrategy, "StepSizeStrategy", this->GetComponentLabel(), 0, 0);
  this->m_StepSizeStrategy = stepSizeStrategy;

  if (this->m_AutomaticParameterEstimation)
  {
    /** Read user setting. */
    configuration.ReadParameter(
      this->m_MaximumStepLengthRatio, "MaximumStepLengthRatio", this->GetComponentLabel(), level, 0);

    /** Set the maximum step length: the maximum displacement of a voxel in mm.
     * Compute default value: mean in-plane spacing of fixed and moving image.
     */
    const unsigned int fixdim = std::min((unsigned int)this->GetElastix()->FixedDimension, (unsigned int)2);
    const unsigned int movdim = std::min((unsigned int)this->GetElastix()->MovingDimension, (unsigned int)2);
    double             sum = 0.0;
    for (unsigned int d = 0; d < fixdim; ++d)
    {
      sum += this->GetElastix()->GetFixedImage()->GetSpacing()[d];
    }
    for (unsigned int d = 0; d < movdim; ++d)
    {
      sum += this->GetElastix()->GetMovingImage()->GetSpacing()[d];
    }
    this->m_MaximumStepLength = this->m_MaximumStepLengthRatio * sum / static_cast<double>(fixdim + movdim);

    /** Read user setting. */
    configuration.ReadParameter(this->m_MaximumStepLength, "MaximumStepLength", this->GetComponentLabel(), level, 0);

    /** Number of gradients N to estimate the average magnitudes
     * of the exact preconditioned gradient and the approximation error.
     */
    this->m_NumberOfGradientMeasurements = 0;
    configuration.ReadParameter(
      this->m_NumberOfGradientMeasurements, "NumberOfGradientMeasurements", this->GetComponentLabel(), level, 0);
    this->m_NumberOfGradientMeasurements =
      std::max(static_cast<SizeValueType>(2), this->m_NumberOfGradientMeasurements);

    /** Set the number of Jacobian measurements M.
     * By default, if nothing specified by the user, M is determined as:
     * M = max( 1000, nrofparams );
     * This is a rather crude rule of thumb, which seems to work in practice.
     */
    this->m_NumberOfJacobianMeasurements = std::max(static_cast<unsigned int>(5000), static_cast<unsigned int>(2 * P));
    configuration.ReadParameter(
      this->m_NumberOfJacobianMeasurements, "NumberOfJacobianMeasurements", this->GetComponentLabel(), level, 0);

    /** Set the NumberOfSpatialSamples. */
    unsigned long numberOfSpatialSamples = 5000;
    configuration.ReadParameter(numberOfSpatialSamples, "NumberOfSpatialSamples", this->GetComponentLabel(), level, 0);
    this->m_NumberOfSpatialSamples = numberOfSpatialSamples;

    /** Set the number of samples for precondition matrix computation.
     * By default, if nothing specified by the user, M is determined as:
     * P = max( 1000, nrofparams );
     * This is a rather crude rule of thumb, which seems to work in practice.
     */
    this->m_NumberOfSamplesForPrecondition = std::max(static_cast<unsigned int>(1000), static_cast<unsigned int>(P));
    configuration.ReadParameter(
      this->m_NumberOfSamplesForPrecondition, "NumberOfSamplesForPrecondition", this->GetComponentLabel(), level, 0);

    /** Set the number of image samples used to compute the 'exact' gradient.
     * By default, if nothing supplied by the user, 100000. This works in general.
     * If the image is smaller, the number of samples is automatically reduced later.
     */
    this->m_NumberOfSamplesForNoiseCompensationFactor = 100000;
    configuration.ReadParameter(this->m_NumberOfSamplesForNoiseCompensationFactor,
                                "NumberOfSamplesForNoiseCompensationFactor",
                                this->GetComponentLabel(),
                                level,
                                0);

    /** Set/Get the scaling factor zeta of the sigmoid width. Large values
     * cause a more wide sigmoid. Default: 0.1. Should be > 0.
     */
    double sigmoidScaleFactor = 0.1;
    configuration.ReadParameter(sigmoidScaleFactor, "SigmoidScaleFactor", this->GetComponentLabel(), level, 0);
    this->m_SigmoidScaleFactor = sigmoidScaleFactor;

    /** Set the regularization factor kappa. */
    this->m_RegularizationKappa = 0.8;
    configuration.ReadParameter(
      this->m_RegularizationKappa, "RegularizationKappa", this->GetComponentLabel(), level, 0);

    /** Set the regularization factor kappa. */
    this->m_ConditionNumber = 2.0;
    configuration.ReadParameter(this->m_ConditionNumber, "ConditionNumber", this->GetComponentLabel(), level, 0);

  } // end if automatic parameter estimation
  else
  {
    /** If no automatic parameter estimation is used, a and alpha also need
     * to be specified.
     */
    double a = 400.0; // arbitrary guess
    double alpha = 0.602;
    configuration.ReadParameter(a, "SP_a", this->GetComponentLabel(), level, 0);
    configuration.ReadParameter(alpha, "SP_alpha", this->GetComponentLabel(), level, 0);
    this->SetParam_a(a);
    this->SetParam_alpha(alpha);

    /** Set/Get the maximum of the sigmoid. Should be > 0. Default: 1.0. */
    double sigmoidMax = 1.0;
    configuration.ReadParameter(sigmoidMax, "SigmoidMax", this->GetComponentLabel(), level, 0);
    this->SetSigmoidMax(sigmoidMax);

    /** Set/Get the minimum of the sigmoid. Should be < 0. Default: -0.8. */
    double sigmoidMin = -0.8;
    configuration.ReadParameter(sigmoidMin, "SigmoidMin", this->GetComponentLabel(), level, 0);
    this->SetSigmoidMin(sigmoidMin);

    /** Set/Get the scaling of the sigmoid width. Large values
     * cause a more wide sigmoid. Default: 1e-8. Should be >0.
     */
    double sigmoidScale = 1e-8;
    configuration.ReadParameter(sigmoidScale, "SigmoidScale", this->GetComponentLabel(), level, 0);
    this->SetSigmoidScale(sigmoidScale);

  } // end else: no automatic parameter estimation

} // end BeforeEachResolution()


/**
 * ***************** AfterEachIteration *************************
 */

template <class TElastix>
void
AdaGrad<TElastix>::AfterEachIteration()
{
  /** Print some information. */
  this->GetIterationInfoAt("2:Metric") << this->GetValue();
  this->GetIterationInfoAt("3a:Time") << this->GetCurrentTime();
  this->GetIterationInfoAt("3b:StepSize") << this->GetLearningRate() * this->m_NoiseFactor;

  bool asFastAsPossible = false;
  if (asFastAsPossible)
  {
    this->GetIterationInfoAt("4a:||Gradient||") << "---";
    this->GetIterationInfoAt("4b:||SearchDirection||") << "---";
  }
  else
  {
    this->GetIterationInfoAt("4a:||Gradient||") << this->GetGradient().magnitude();
    this->GetIterationInfoAt("4b:||SearchDirection||") << this->GetSearchDirection().magnitude();
  }

  /** Select new spatial samples for the computation of the metric. */
  if (this->GetNewSamplesEveryIteration())
  {
    this->SelectNewSamples();
  }

} // end AfterEachIteration()


/**
 * ***************** AfterEachResolution *************************
 */

template <class TElastix>
void
AdaGrad<TElastix>::AfterEachResolution()
{
  /** Get the current resolution level. */
  unsigned int level = static_cast<unsigned int>(this->m_Registration->GetAsITKBaseType()->GetCurrentLevel());

  /**
   * enum StopConditionType {
   *   MaximumNumberOfIterations,
   *   MetricError,
   *   MinimumStepSize } ;
   */
  std::string stopcondition;

  switch (this->GetStopCondition())
  {
    case MaximumNumberOfIterations:
      stopcondition = "Maximum number of iterations has been reached";
      break;

    case MetricError:
      stopcondition = "Error in metric";
      break;

    case MinimumStepSize:
      stopcondition = "The minimum step length has been reached";
      break;

    default:
      stopcondition = "Unknown";
      break;
  }

  /** Print the stopping condition. */
  log::info(std::ostringstream{} << "Stopping condition: " << stopcondition << ".");

  /** Store the used parameters, for later printing to screen. */
  SettingsType settings;
  settings.a = this->GetParam_a();
  settings.A = this->GetParam_A();
  settings.alpha = this->GetParam_alpha();
  settings.fmax = this->GetSigmoidMax();
  settings.fmin = this->GetSigmoidMin();
  settings.omega = this->GetSigmoidScale();
  this->m_SettingsVector.push_back(settings);

  /** Print settings that were used in this resolution. */
  SettingsVectorType tempSettingsVector;
  tempSettingsVector.push_back(settings);
  log::info(std::ostringstream{} << "Settings of " << this->elxGetClassName() << " in resolution " << level << ":");
  Superclass2::PrintSettingsVector(tempSettingsVector);

} // end AfterEachResolution()


/**
 * ******************* AfterRegistration ************************
 */

template <class TElastix>
void
AdaGrad<TElastix>::AfterRegistration()
{
  /** Print the best metric value. */
  double bestValue = this->GetValue();
  log::info(std::ostringstream{} << '\n'
                                 << "Final metric value  = " << bestValue << '\n'

                                 << "Settings of " << this->elxGetClassName() << " for all resolutions:");
  Superclass2::PrintSettingsVector(this->m_SettingsVector);

} // end AfterRegistration()


/**
 * ****************** StartOptimization *************************
 */

template <class TElastix>
void
AdaGrad<TElastix>::StartOptimization()
{
  /** As this optimizer estimates the scales itself, no other scales are used. */
  this->SetUseScales(false);

  this->m_AutomaticParameterEstimationDone = false;
  this->Superclass1::StartOptimization();

} // end StartOptimization()


/**
 * ********************** AdvanceOneStep **********************
 */

template <class TElastix>
void
AdaGrad<TElastix>::AdvanceOneStep()
{
  /** Get space dimension. */
  const unsigned int spaceDimension = this->GetScaledCostFunction()->GetNumberOfParameters();

  /** Compute and set the learning rate. */
  double lamda = this->GetParam_a() / (1.0 + this->Superclass1::GetCurrentTime() / this->GetParam_A());
  this->SetLearningRate(lamda);

  DerivativeType & searchDirection = this->m_SearchDirection;

  /** Get a reference to the previously allocated newPosition. */
  ParametersType & newPosition = this->m_ScaledCurrentPosition;

  /** Get a reference to the current position. */
  const ParametersType & currentPosition = this->GetScaledCurrentPosition();

  /** Update the new position. */
  const double eta = 1e-14;
  const double lamda2 = lamda * this->m_NoiseFactor;
  //  const double lamda2 = 0.01;
  for (unsigned int j = 0; j < spaceDimension; ++j)
  {
    this->m_PreconditionVector[j] += this->m_Gradient[j] * this->m_Gradient[j];
    searchDirection[j] = this->m_Gradient[j] / (std::sqrt(this->m_PreconditionVector[j] + eta));
    newPosition[j] = currentPosition[j] - lamda2 * searchDirection[j];
  }

  this->Superclass1::UpdateCurrentTime();
  this->InvokeEvent(itk::IterationEvent());

} // end AdvanceOneStep()


/**
 * ********************** ResumeOptimization **********************
 */

template <class TElastix>
void
AdaGrad<TElastix>::ResumeOptimization()
{
  /** The following code relies on the fact that all components have been set up and
   * that the initial position has been set, so must be called in this function.
   */
  if (this->GetAutomaticParameterEstimation() && !this->m_AutomaticParameterEstimationDone)
  {
    this->AutomaticPreconditionerEstimation();
    this->m_AutomaticParameterEstimationDone = true; // hack
  }

  this->Superclass1::ResumeOptimization();

} // end ResumeOptimization()


/**
 * ****************** MetricErrorResponse *************************
 */

template <class TElastix>
void
AdaGrad<TElastix>::MetricErrorResponse(itk::ExceptionObject & err)
{
  if (this->GetCurrentIteration() != this->m_PreviousErrorAtIteration)
  {
    this->m_PreviousErrorAtIteration = this->GetCurrentIteration();
    this->m_CurrentNumberOfSamplingAttempts = 1;
  }
  else
  {
    this->m_CurrentNumberOfSamplingAttempts++;
  }

  if (this->m_CurrentNumberOfSamplingAttempts <= this->m_MaximumNumberOfSamplingAttempts)
  {
    this->SelectNewSamples();
    this->ResumeOptimization();
  }
  else
  {
    /** Stop optimization and pass on exception. */
    this->Superclass1::MetricErrorResponse(err);
  }

} // end MetricErrorResponse()


/**
 * ******************* AutomaticPreconditionerEstimation **********************
 */

template <class TElastix>
void
AdaGrad<TElastix>::AutomaticPreconditionerEstimation()
{
  /** Total time. */
  itk::TimeProbe timer, timer4;
  timer.Start();
  log::info(std::ostringstream{} << "Starting preconditioner estimation for " << this->elxGetClassName() << " ...");

  /** Get current position to start the parameter estimation. */
  this->GetRegistration()->GetAsITKBaseType()->GetModifiableTransform()->SetParameters(this->GetCurrentPosition());

  /** Get the number of parameters. */
  unsigned int P =
    static_cast<unsigned int>(this->GetRegistration()->GetAsITKBaseType()->GetTransform()->GetNumberOfParameters());

  this->m_SearchDirection = ParametersType(P);
  this->m_SearchDirection.Fill(0.0); // if the print out is not needed, this could be removed. YQ

  /** Cast to advanced metric type. */
  using MetricType = typename ElastixType::MetricBaseType::AdvancedMetricType;
  MetricType * testPtr = dynamic_cast<MetricType *>(this->GetElastix()->GetElxMetricBase()->GetAsITKBaseType());
  if (!testPtr)
  {
    itkExceptionMacro("ERROR: VoxelWiseASGD expects the metric to be of type AdvancedImageToImageMetric!");
  }

  /** Getting pointers to the samplers. */
  const unsigned int                   M = this->GetElastix()->GetNumberOfMetrics();
  std::vector<ImageSamplerBasePointer> originalSampler(M);
  for (unsigned int m = 0; m < M; ++m)
  {
    ImageSamplerBasePointer sampler = this->GetElastix()->GetElxMetricBase(m)->GetAdvancedMetricImageSampler();
    originalSampler[m] = sampler.GetPointer();
  }

#if 0
  /** Get the current resolution level. */
  unsigned int level = static_cast<unsigned int>(this->m_Registration->GetAsITKBaseType()->GetCurrentLevel());

  /** Create some samplers that can be used for the pre-conditioner computation. */
  //std::vector< ImageRandomCoordinateSamplerPointer > preconditionSamplers( M, 0 ); // very slow, leave this for reminder. YQ
  std::vector< ImageRandomSamplerPointer > preconditionSamplers( M );
  for( unsigned int m = 0; m < M; ++m )
  {
    ImageSamplerBasePointer sampler =
      this->GetElastix()->GetElxMetricBase( m )->GetAdvancedMetricImageSampler();
    //preconditionSamplers[ m ] = ImageRandomCoordinateSamplerType::New();
    preconditionSamplers[ m ] = ImageRandomSamplerType::New();
    preconditionSamplers[ m ]->SetInput( sampler->GetInput() );
    preconditionSamplers[ m ]->SetInputImageRegion( sampler->GetInputImageRegion() );
    preconditionSamplers[ m ]->SetMask( sampler->GetMask() );
    preconditionSamplers[ m ]->SetNumberOfSamples( this->m_NumberOfSamplesForPrecondition );
    preconditionSamplers[ m ]->Update();
    this->GetElastix()->GetElxMetricBase( m )
      ->SetAdvancedMetricImageSampler( preconditionSamplers[ m ] );
  }

  /** Construct preconditionerEstimator to initialize the preconditioner estimation. */
  PreconditionerEstimationPointer preconditionerEstimator = PreconditionerEstimationType::New();
  preconditionerEstimator->SetFixedImage( testPtr->GetFixedImage() );
  preconditionerEstimator->SetFixedImageRegion( testPtr->GetFixedImageRegion() );
  preconditionerEstimator->SetFixedImageMask( testPtr->GetFixedImageMask() );
  preconditionerEstimator->SetTransform(
    this->GetRegistration()->GetAsITKBaseType()->GetTransform() );
  preconditionerEstimator->SetCostFunction( this->m_CostFunction );
  preconditionerEstimator->SetNumberOfJacobianMeasurements(
    this->m_NumberOfJacobianMeasurements );
  preconditionerEstimator->SetRegularizationKappa( this->m_RegularizationKappa );
  preconditionerEstimator->SetMaximumStepLength( this->m_MaximumStepLength );
  preconditionerEstimator->SetConditionNumber( this->m_ConditionNumber );
  preconditionerEstimator->SetUseScales( false ); // Make sure scales are not used
#endif
  const Configuration & configuration = itk::Deref(Superclass2::GetConfiguration());

  /** Construct the preconditioner and initialize. */
  this->m_PreconditionVector = ParametersType(P);
  this->m_PreconditionVector.Fill(0.0);
#if 0
  /** Compute the preconditioner. */
  itk::TimeProbe timer_P; timer_P.Start();
  log::info(std::ostringstream{}  << "  Computing preconditioner ...");
  double maxJJ = 0; // needed for the noise compensation term

  bool JacobiType = false;
  configuration.ReadParameter( JacobiType,
    "JacobiTypePreconditioner", this->GetComponentLabel(), level, 0 );

  if( JacobiType )
  {
    preconditionerEstimator->ComputeJacobiTypePreconditioner(
      maxJJ, this->m_PreconditionVector );
  }
  else
  {
    preconditionerEstimator->Compute( this->GetScaledCurrentPosition(),
      maxJJ, this->m_PreconditionVector );
  }

  timer_P.Stop();
  elxout << "  Computing the preconditioner took " <<  Conversion::SecondsToDHMS( timer_P.GetMean(), 6 ) <<  std::endl;


  elxout << std::scientific << "The preconditioner: [ ";
  for( unsigned int i = 0; i < P; ++i ) elxout << m_PreconditionVector[ i ] << " ";
  elxout << "]" <<  std::endl << std::fixed;


  /** Set the sampler back to the original. */
  for( unsigned int m = 0; m < M; ++m )
  {
    this->GetElastix()->GetElxMetricBase( m )->SetAdvancedMetricImageSampler( originalSampler[ m ] );
  }
#endif
  /** Construct computeJacobianTerms to initialize the parameter estimation. */
  double                                                jacg = 0.0;
  double                                                maxJJ = 0.0;
  typename ComputeDisplacementDistributionType::Pointer computeDisplacementDistribution =
    ComputeDisplacementDistributionType::New();
  computeDisplacementDistribution->SetFixedImage(testPtr->GetFixedImage());
  computeDisplacementDistribution->SetFixedImageRegion(testPtr->GetFixedImageRegion());
  computeDisplacementDistribution->SetFixedImageMask(testPtr->GetFixedImageMask());
  computeDisplacementDistribution->SetTransform(this->GetRegistration()->GetAsITKBaseType()->GetModifiableTransform());
  computeDisplacementDistribution->SetCostFunction(this->m_CostFunction);
  computeDisplacementDistribution->SetNumberOfJacobianMeasurements(this->m_NumberOfJacobianMeasurements);


  std::string maximumDisplacementEstimationMethod = "2sigma";
  configuration.ReadParameter(
    maximumDisplacementEstimationMethod, "MaximumDisplacementEstimationMethod", this->GetComponentLabel(), 0, 0);

  /** Compute the Jacobian terms. */
  log::info("  Computing displacement distribution ...");
  timer4.Start();
  computeDisplacementDistribution->Compute(
    this->GetScaledCurrentPosition(), jacg, maxJJ, maximumDisplacementEstimationMethod);
  timer4.Stop();
  log::info(std::ostringstream{} << "  Computing the displacement distribution took "
                                 << Conversion::SecondsToDHMS(timer4.GetMean(), 6));

  /** Sample the fixed image to estimate the noise factor. */
  itk::TimeProbe timer_noise;
  timer_noise.Start();
  double sigma4factor = 1.0;
  double sigma4 = 0.0;
  log::info(std::ostringstream{} << "  The estimated MaxJJ is: " << maxJJ);
  if (maxJJ > 1e-14)
  {
    sigma4 = sigma4factor * this->m_MaximumStepLength / std::sqrt(maxJJ);
  }
  double gg = 0.0;
  double ee = 0.0;
  this->SampleGradients(this->GetScaledCurrentPosition(), sigma4, gg, ee);
  this->m_NoiseFactor = gg / (gg + ee);
  timer_noise.Stop();
  log::info(std::ostringstream{} << "  The MaxJJ used for noisefactor is: " << maxJJ << '\n'
                                 << "  The NoiseFactor is: " << m_NoiseFactor << '\n'
                                 << "  Compute the noise compensation took "
                                 << Conversion::SecondsToDHMS(timer_noise.GetMean(), 6));

  // MS: the following can probably be removed or moved.
  // YQ: these variables are used to update the time for adaptive step size.
  // See in itkPreconditionedASGDOptimizer.cxx
  /** Initial of the variables. */
  const double alpha = 1.0;
  const double fmax = 1.0;
  const double fmin = -0.8;
  /** Set parameters in superclass. */
  this->SetParam_alpha(alpha);
  this->SetSigmoidMax(fmax);
  this->SetSigmoidMin(fmin);
  /** Initial of the variables. */
  double       a = 1.0;
  const double A = this->GetParam_A();
  const double delta = this->GetMaximumStepLength();

  a = delta * std::pow(A + 1.0, alpha) / (jacg + 1e-14);

  this->SetParam_a(a);
  /** Print the elapsed time. */
  timer.Stop();
  log::info(std::ostringstream{} << "Automatic preconditioner estimation took "
                                 << Conversion::SecondsToDHMS(timer.GetMean(), 2));

} // end AutomaticPreconditionerEstimation()


/**
 * ******************** SampleGradients **********************
 */

template <class TElastix>
void
AdaGrad<TElastix>::SampleGradients(const ParametersType & mu0, double perturbationSigma, double & gg, double & ee)
{
  /** Some shortcuts. */
  const unsigned int M = this->GetElastix()->GetNumberOfMetrics();

  /** Variables for sampler support. Each metric may have a sampler. */
  std::vector<bool>                                useRandomSampleRegionVec(M, false);
  std::vector<ImageRandomSamplerBasePointer>       randomSamplerVec(M);
  std::vector<ImageRandomCoordinateSamplerPointer> randomCoordinateSamplerVec(M);
  std::vector<ImageGridSamplerPointer>             gridSamplerVec(M);

  /** If new samples every iteration, get each sampler, and check if it is
   * a kind of random sampler. If yes, prepare an additional grid sampler
   * for the exact gradients, and set the stochasticgradients flag to true.
   */
  bool stochasticgradients = false;
  if (this->GetNewSamplesEveryIteration())
  {
    for (unsigned int m = 0; m < M; ++m)
    {
      /** Get the sampler. */
      ImageSamplerBasePointer sampler = this->GetElastix()->GetElxMetricBase(m)->GetAdvancedMetricImageSampler();
      randomSamplerVec[m] = dynamic_cast<ImageRandomSamplerBaseType *>(sampler.GetPointer());
      randomCoordinateSamplerVec[m] = dynamic_cast<ImageRandomCoordinateSamplerType *>(sampler.GetPointer());

      if (randomSamplerVec[m].IsNotNull())
      {
        /** At least one of the metric has a random sampler. */
        stochasticgradients |= true;

        /** If the sampler is a randomCoordinateSampler set the UseRandomSampleRegion
         * property to false temporarily. It disturbs the parameter estimation.
         * At the end of this function the original setting is set back.
         * Also, the AdaptiveStepSize mechanism is turned off when any of the samplers
         * has UseRandomSampleRegion==true.
         * \todo Extend ASGD to really take into account random region sampling.
         * \todo This does not work for the MultiInputRandomCoordinateImageSampler,
         * because it does not inherit from the RandomCoordinateImageSampler
         */
        if (randomCoordinateSamplerVec[m].IsNotNull())
        {
          useRandomSampleRegionVec[m] = randomCoordinateSamplerVec[m]->GetUseRandomSampleRegion();
          if (useRandomSampleRegionVec[m])
          {
            if (this->m_StepSizeStrategy == "Adaptive")
            {
              log::warn(
                "WARNING: StepSizeStrategy is set to Constant, because UseRandomSampleRegion is set to \"true\".");
              this->m_StepSizeStrategy = "Constant";
            }
          }
          /** Do not turn it off yet, as it would go wrong if you multiple metrics are using
           * all the same sampler. */
          // randomCoordinateSamplerVec[ m ]->SetUseRandomSampleRegion( false );

        } // end if random coordinate sampler

        /** Set up the grid sampler for the "exact" gradients.
         * Copy settings from the random sampler and update.
         */
        gridSamplerVec[m] = ImageGridSamplerType::New();
        gridSamplerVec[m]->SetInput(randomSamplerVec[m]->GetInput());
        gridSamplerVec[m]->SetInputImageRegion(randomSamplerVec[m]->GetInputImageRegion());
        gridSamplerVec[m]->SetMask(randomSamplerVec[m]->GetMask());
        gridSamplerVec[m]->SetNumberOfSamples(this->m_NumberOfSamplesForNoiseCompensationFactor);
        gridSamplerVec[m]->Update();

      } // end if random sampler

    } // end for loop over metrics

    /** Start a second loop over all metrics to turn off the random region sampling. */
    for (unsigned int m = 0; m < M; ++m)
    {
      if (randomCoordinateSamplerVec[m].IsNotNull())
      {
        randomCoordinateSamplerVec[m]->SetUseRandomSampleRegion(false);
      }
    } // end loop over metrics

  } // end if NewSamplesEveryIteration.

  const Configuration & configuration = itk::Deref(Superclass2::GetConfiguration());

  /** Prepare for progress printing. */
  const bool showProgressPercentage = configuration.RetrieveParameterValue(false, "ShowProgressPercentage", 0, false);
  const auto progressObserver = showProgressPercentage
                                  ? ProgressCommand::CreateAndSetUpdateFrequency(this->m_NumberOfGradientMeasurements)
                                  : nullptr;
  log::info("  Sampling gradients ...");

  /** Initialize some variables for storing gradients and their magnitudes. */
  const unsigned int P = this->GetElastix()->GetElxTransformBase()->GetAsITKBaseType()->GetNumberOfParameters();
  DerivativeType     approxgradient(P);
  DerivativeType     exactgradient(P);
  DerivativeType     searchDirection(P);
  DerivativeType     diffgradient;
  double             exactgg = 0.0;
  double             diffgg = 0.0;
  double             approxgg = 0.0;

  /** Compute gg for some random parameters. */
  for (unsigned int i = 0; i < this->m_NumberOfGradientMeasurements; ++i)
  {
    if (progressObserver != nullptr)
    {
      /** Show progress 0-100% */
      progressObserver->UpdateAndPrintProgress(i);
    }
    /** Generate a perturbation, according to:
     *    \mu_i ~ N( \mu_0, perturbationsigma^2 I ).
     */
    ParametersType perturbedMu0 = mu0;
    this->AddRandomPerturbation(perturbedMu0, perturbationSigma);

    /** Compute contribution to exactgg and diffgg. */
    if (stochasticgradients)
    {
      /** Set grid sampler(s) and get exact derivative. */
      for (unsigned int m = 0; m < M; ++m)
      {
        if (gridSamplerVec[m].IsNotNull())
        {
          this->GetElastix()->GetElxMetricBase(m)->SetAdvancedMetricImageSampler(gridSamplerVec[m]);
        }
      }
      this->GetScaledDerivativeWithExceptionHandling(perturbedMu0, exactgradient);

      exactgg += inner_product(exactgradient, exactgradient);

      /** Set random sampler(s), select new spatial samples and get approximate derivative. */
      for (unsigned int m = 0; m < M; ++m)
      {
        if (randomSamplerVec[m].IsNotNull())
        {
          this->GetElastix()->GetElxMetricBase(m)->SetAdvancedMetricImageSampler(randomSamplerVec[m]);
        }
      }
      this->SelectNewSamples();
      this->GetScaledDerivativeWithExceptionHandling(perturbedMu0, approxgradient);

      /** Compute error vector. */
      diffgradient = exactgradient - approxgradient;
      approxgg = inner_product(diffgradient, diffgradient);
      diffgg += approxgg;
    }
    else // no stochastic gradients
    {
      /** Get exact gradient. */
      this->GetScaledDerivativeWithExceptionHandling(perturbedMu0, exactgradient);

      /** Compute g^T g. NB: diffgg=0. */
      exactgg += exactgradient.squared_magnitude();
    } // end else: no stochastic gradients

  } // end for loop over gradient measurements

  if (progressObserver != nullptr)
  {
    progressObserver->PrintProgress(1.0);
  }

  /** Compute means. */
  exactgg /= this->m_NumberOfGradientMeasurements;
  diffgg /= this->m_NumberOfGradientMeasurements;

  /** For output: gg and ee.
   * gg and ee will be divided by Pd, but actually need to be divided by
   * the rank, in case of maximum likelihood. In case of no maximum likelihood,
   * the rank equals Pd.
   */
  gg = std::abs(exactgg);
  ee = std::abs(diffgg);

  /** Set back useRandomSampleRegion flag to what it was. */
  for (unsigned int m = 0; m < M; ++m)
  {
    if (randomCoordinateSamplerVec[m].IsNotNull())
    {
      randomCoordinateSamplerVec[m]->SetUseRandomSampleRegion(useRandomSampleRegionVec[m]);
    }
  }

} // end SampleGradients()


/**
 * *************** GetScaledDerivativeWithExceptionHandling ***************
 */

template <class TElastix>
void
AdaGrad<TElastix>::GetScaledDerivativeWithExceptionHandling(const ParametersType & parameters,
                                                            DerivativeType &       derivative)
{
  double dummyvalue = 0;
  try
  {
    this->GetScaledValueAndDerivative(parameters, dummyvalue, derivative);
  }
  catch (const itk::ExceptionObject &)
  {
    this->m_StopCondition = MetricError;
    this->StopOptimization();
    throw;
  }

} // end GetScaledDerivativeWithExceptionHandling()


/**
 * *************** AddRandomPerturbation ***************
 */

template <class TElastix>
void
AdaGrad<TElastix>::AddRandomPerturbation(ParametersType & parameters, double sigma)
{
  /** Add delta ~ sigma * N(0,I) to the input parameters. */
  for (unsigned int p = 0; p < parameters.GetSize(); ++p)
  {
    parameters[p] += sigma * this->m_RandomGenerator->GetNormalVariate(0.0, 1.0);
  }

} // end AddRandomPerturbation()


} // end namespace elastix

#endif // end #ifndef elxAdaGrad_hxx