FindSXPeaksHelper.cpp

Go to the documentation of this file.

// Mantid Repository : https://github.com/mantidproject/mantid
//
// Copyright © 2018 ISIS Rutherford Appleton Laboratory UKRI,
//   NScD Oak Ridge National Laboratory, European Spallation Source,
//   Institut Laue - Langevin & CSNS, Institute of High Energy Physics, CAS
// SPDX - License - Identifier: GPL - 3.0 +
#include "MantidCrystal/FindSXPeaksHelper.h"
#include "MantidAPI/Progress.h"
#include "MantidGeometry/IDetector.h"
#include "MantidKernel/ConfigService.h"
#include "MantidKernel/FloatingPointComparison.h"
#include "MantidKernel/Logger.h"
#include "MantidKernel/PhysicalConstants.h"
#include "MantidKernel/Unit.h"
#include "MantidKernel/UnitFactory.h"
 
#include "MantidTypes/SpectrumDefinition.h"
 
#define BOOST_ALLOW_DEPRECATED_HEADERS
#include <boost/graph/adjacency_list.hpp>
#undef BOOST_ALLOW_DEPRECATED_HEADERS
 
#include <boost/graph/connected_components.hpp>
#include <cmath>
 
namespace {
 
const double TWO_PI = 2 * M_PI;
 
bool isDifferenceLargerThanTolerance(const double angle1, const double angle2, const double tolerance) {
  auto difference = std::abs(angle1 - angle2);
 
  // If we have more than 360 degree angle difference then we need to wrap it
  // back to 360
  if (difference > TWO_PI) {
    difference = std::fmod(difference, TWO_PI);
  }
 
  // If we have more than 180 degrees then we must take the smaller angle
  if (difference > M_PI) {
    difference = TWO_PI - difference;
  }
 
  return difference > tolerance;
}
} // namespace
 
using namespace boost;
using Mantid::Kernel::UnitFactory;
 
namespace Mantid::Crystal::FindSXPeaksHelper {
 
Mantid::Kernel::Logger g_log("FindSXPeaksHelper");
 
/* ------------------------------------------------------------------------------------------
 * Single Crystal peak representation
 * ------------------------------------------------------------------------------------------
 */
 
SXPeak::SXPeak(double t, double phi, double intensity, const std::vector<int> &spectral, const size_t wsIndex,
               const API::SpectrumInfo &spectrumInfo)
    : m_tof(t), m_phi(phi), m_intensity(intensity), m_spectra(spectral), m_wsIndex(wsIndex) {
  // Sanity checks
  if (intensity < 0) {
    throw std::invalid_argument("SXPeak: Cannot have an intensity < 0");
  }
  if (spectral.empty()) {
    throw std::invalid_argument("SXPeak: Cannot have zero sized spectral list");
  }
  if (!spectrumInfo.hasDetectors(m_wsIndex)) {
    throw std::invalid_argument("SXPeak: Spectrum at ws index " + std::to_string(wsIndex) + " doesn't have detectors");
  }
 
  const auto l1 = spectrumInfo.l1();
  const auto l2 = spectrumInfo.l2(m_wsIndex);
 
  m_twoTheta = spectrumInfo.twoTheta(m_wsIndex);
  m_LTotal = l1 + l2;
  if (m_LTotal < 0) {
    throw std::invalid_argument("SXPeak: Cannot have detector distance < 0");
  }
  m_detId = spectrumInfo.detector(m_wsIndex).getID();
  m_nPixels = 1;
 
  Mantid::Kernel::Units::TOF tof;
  const auto unit = Mantid::Kernel::UnitFactory::Instance().create("dSpacing");
  Kernel::UnitParametersMap pmap{};
  spectrumInfo.getDetectorValues(tof, *unit, Kernel::DeltaEMode::Elastic, false, m_wsIndex, pmap);
  unit->initialize(l1, 0, pmap);
  try {
    m_dSpacing = unit->singleFromTOF(m_tof);
  } catch (std::exception &) {
    m_dSpacing = 0;
  }
 
  const auto samplePos = spectrumInfo.samplePosition();
  const auto sourcePos = spectrumInfo.sourcePosition();
  const auto detPos = spectrumInfo.position(m_wsIndex);
  // Normalized beam direction
  const auto beamDir = normalize(samplePos - sourcePos);
  // Normalized detector direction
  const auto detDir = normalize(detPos - samplePos);
  m_unitWaveVector = beamDir - detDir;
  m_qConvention = Kernel::ConfigService::Instance().getString("Q.convention");
}
 
bool SXPeak::compare(const SXPeak &rhs, double tolerance) const {
  if (!Kernel::withinRelativeDifference(m_tof / m_nPixels, rhs.m_tof / rhs.m_nPixels, tolerance))
    return false;
  if (!Kernel::withinRelativeDifference(m_phi / m_nPixels, rhs.m_phi / rhs.m_nPixels, tolerance))
    return false;
  if (!Kernel::withinRelativeDifference(m_twoTheta / m_nPixels, rhs.m_twoTheta / rhs.m_nPixels, tolerance))
    return false;
  return true;
}
 
bool SXPeak::compare(const SXPeak &rhs, const double xTolerance, const double phiTolerance, const double thetaTolerance,
                     const XAxisUnit units) const {
 
  const auto x_1 = (units == XAxisUnit::TOF) ? m_tof : m_dSpacing;
  const auto x_2 = (units == XAxisUnit::TOF) ? rhs.m_tof : rhs.m_dSpacing;
 
  if (std::abs(x_1 - x_2) > xTolerance) {
    return false;
  }
 
  if (isDifferenceLargerThanTolerance(m_phi, rhs.m_phi, phiTolerance)) {
    return false;
  }
  if (isDifferenceLargerThanTolerance(m_twoTheta, rhs.m_twoTheta, thetaTolerance)) {
    return false;
  }
  return true;
}
 
Mantid::Kernel::V3D SXPeak::getQ() const {
  double qSign = 1.0;
  if (m_qConvention == "Crystallography") {
    qSign = -1.0;
  }
  double vi = m_LTotal / (m_tof * 1e-6);
  // wavenumber = h_bar / mv
  double wi = PhysicalConstants::h_bar / (PhysicalConstants::NeutronMass * vi);
  // in angstroms
  wi *= 1e10;
  // wavevector=1/wavenumber = 2pi/wavelength
  double wvi = 1.0 / wi;
  // Now calculate the wavevector of the scattered neutron
  return m_unitWaveVector * wvi * qSign;
}
 
SXPeak &SXPeak::operator+=(const SXPeak &rhs) {
  m_tof += rhs.m_tof;
  m_phi += rhs.m_phi;
  m_twoTheta += rhs.m_twoTheta;
  m_intensity += rhs.m_intensity;
  m_LTotal += rhs.m_LTotal;
  m_nPixels += 1;
  m_spectra.insert(m_spectra.end(), rhs.m_spectra.cbegin(), rhs.m_spectra.cend());
  return *this;
}
 
void SXPeak::reduce() {
  m_tof /= m_nPixels;
  m_phi /= m_nPixels;
  m_twoTheta /= m_nPixels;
  m_intensity /= m_nPixels;
  m_LTotal /= m_nPixels;
  m_nPixels = 1;
}
 
const double &SXPeak::getIntensity() const { return m_intensity; }
 
detid_t SXPeak::getDetectorId() const { return m_detId; }
 
const std::vector<int> &SXPeak::getPeakSpectras() const { return m_spectra; }
 
PeakContainer::PeakContainer(const HistogramData::HistogramY &y)
    : m_y(y), m_startIndex(0), m_stopIndex(m_y.size() - 1), m_maxIndex(0) {}
 
void PeakContainer::startRecord(yIt item) {
  m_startIndex = std::distance(m_y.begin(), item);
  m_maxIndex = m_startIndex;
  m_maxSignal = *item;
}
 
void PeakContainer::record(yIt item) {
  if (*item > m_maxSignal) {
    m_maxIndex = std::distance(m_y.begin(), item);
    m_maxSignal = *item;
  }
}
 
void PeakContainer::stopRecord(yIt item) {
  if (*item > m_maxSignal) {
    m_maxIndex = std::distance(m_y.begin(), item);
    m_maxSignal = *item;
  }
  m_stopIndex = std::distance(m_y.begin(), item);
 
  // Peak end is one back though
  --m_stopIndex;
}
 
size_t PeakContainer::getNumberOfPointsInPeak() const {
  // If we didn't record anything then the start iterator is at the end
  if (m_startIndex >= m_y.size()) {
    return 0;
  }
  if (m_stopIndex >= m_startIndex) {
    return m_stopIndex - m_startIndex + 1;
  }
  return 0;
}
 
yIt PeakContainer::getMaxIterator() const { return m_y.begin() + m_maxIndex; }
 
double PeakContainer::getStartingSignal() const { return m_y[m_startIndex]; }
 
/* ------------------------------------------------------------------------------------------
 * Background
 * ------------------------------------------------------------------------------------------
 */
AbsoluteBackgroundStrategy::AbsoluteBackgroundStrategy(const double background) : m_background(background) {}
 
bool AbsoluteBackgroundStrategy::isBelowBackground(const double intensity,
                                                   const HistogramData::HistogramY & /*y*/) const {
  return intensity < m_background;
}
 
PerSpectrumBackgroundStrategy::PerSpectrumBackgroundStrategy(const double backgroundMultiplier)
    : m_backgroundMultiplier(backgroundMultiplier) {}
 
bool PerSpectrumBackgroundStrategy::isBelowBackground(const double intensity,
                                                      const HistogramData::HistogramY &y) const {
  auto background = 0.5 * (1.0 + y.front() + y.back());
  background *= m_backgroundMultiplier;
  return intensity < background;
}
 
/* ------------------------------------------------------------------------------------------
 * Peak Finding Strategy
 * ------------------------------------------------------------------------------------------
 */
 
PeakFindingStrategy::PeakFindingStrategy(const BackgroundStrategy *backgroundStrategy,
                                         const API::SpectrumInfo &spectrumInfo, const double minValue,
                                         const double maxValue, const XAxisUnit units)
    : m_backgroundStrategy(backgroundStrategy), m_minValue(minValue), m_maxValue(maxValue),
      m_spectrumInfo(spectrumInfo), m_units(units) {}
 
PeakList PeakFindingStrategy::findSXPeaks(const HistogramData::HistogramX &x, const HistogramData::HistogramY &y,
                                          const HistogramData::HistogramE &e, const int workspaceIndex) const {
  // ---------------------------------------
  // Get the lower and upper bound iterators
  // ---------------------------------------
  auto boundsIterator = getBounds(x);
  auto lowit = boundsIterator.first;
  auto highit = boundsIterator.second;
 
  // If range specified doesn't overlap with this spectrum then bail out
  if (lowit == x.end() || highit == x.begin()) {
    return PeakList();
  }
 
  // Upper limit is the bin before, i.e. the last value smaller than MaxRange
  --highit;
 
  // ---------------------------------------
  // Perform the search of the peaks
  // ---------------------------------------
  return dofindSXPeaks(x, y, e, lowit, highit, workspaceIndex);
}
 
void PeakFindingStrategy::setMinNBinsPerPeak(int minNBinsPerPeak) { m_minNBinsPerPeak = minNBinsPerPeak; }
 
void PeakFindingStrategy::filterPeaksForMinBins(std::vector<std::unique_ptr<PeakContainer>> &inputPeakList) const {
  if (m_minNBinsPerPeak == EMPTY_INT() || inputPeakList.empty()) {
    return;
  }
 
  for (auto inputIt = inputPeakList.begin(); inputIt != inputPeakList.end();) {
    if (static_cast<int>((*inputIt)->getNumberOfPointsInPeak()) < m_minNBinsPerPeak) {
      inputIt = inputPeakList.erase(inputIt);
    } else {
      ++inputIt;
    }
  }
}
 
BoundsIterator PeakFindingStrategy::getBounds(const HistogramData::HistogramX &x) const {
  // Find the range [min,max]
  auto lowit = (m_minValue == EMPTY_DBL()) ? x.begin() : std::lower_bound(x.begin(), x.end(), m_minValue);
  using std::placeholders::_1;
  auto highit = (m_maxValue == EMPTY_DBL())
                    ? x.end()
                    : std::find_if(lowit, x.end(), std::bind(std::greater<double>(), _1, m_maxValue));
 
  return std::make_pair(lowit, highit);
}
 
double PeakFindingStrategy::calculatePhi(size_t workspaceIndex) const {
  double phi = m_spectrumInfo.phi(workspaceIndex);
 
  if (phi < 0) {
    phi += 2.0 * M_PI;
  }
  return phi;
}
 
double PeakFindingStrategy::getXValue(const HistogramData::HistogramX &x, const size_t peakLocation) const {
  auto leftBinPosition = x.begin() + peakLocation;
  const double leftBinEdge = *leftBinPosition;
  const double rightBinEdge = *std::next(leftBinPosition);
  return 0.5 * (leftBinEdge + rightBinEdge);
}
 
double PeakFindingStrategy::convertToTOF(const double xValue, const size_t workspaceIndex) const {
  if (m_units == XAxisUnit::TOF) {
    // we're already using TOF units
    return xValue;
  } else {
    const auto unit = UnitFactory::Instance().create("dSpacing");
    Mantid::Kernel::Units::TOF tof;
    Kernel::UnitParametersMap pmap{};
    m_spectrumInfo.getDetectorValues(*unit, tof, Kernel::DeltaEMode::Elastic, false, workspaceIndex, pmap);
    // we're using d-spacing, convert the point to TOF
    unit->initialize(m_spectrumInfo.l1(), 0, pmap);
    return unit->singleToTOF(xValue);
  }
}
 
PeakList PeakFindingStrategy::convertToSXPeaks(const HistogramData::HistogramX &x, const HistogramData::HistogramY &y,
                                               const std::vector<std::unique_ptr<PeakContainer>> &foundPeaks,
                                               const int workspaceIndex) const {
  PeakList peaks;
 
  if (foundPeaks.empty()) {
    return peaks;
  }
 
  // Add a vector to the std::optional
  peaks = std::vector<FindSXPeaksHelper::SXPeak>();
 
  for (const auto &peak : foundPeaks) {
    // Get the index of the bin
    auto maxY = peak->getMaxIterator();
    const auto distance = std::distance(y.begin(), maxY);
    const auto xValue = getXValue(x, distance);
    const auto tof = convertToTOF(xValue, workspaceIndex);
    const double phi = calculatePhi(workspaceIndex);
 
    std::vector<int> specs(1, workspaceIndex);
    (*peaks).emplace_back(tof, phi, *maxY, specs, workspaceIndex, m_spectrumInfo);
  }
 
  return peaks;
}
 
StrongestPeaksStrategy::StrongestPeaksStrategy(const BackgroundStrategy *backgroundStrategy,
                                               const API::SpectrumInfo &spectrumInfo, const double minValue,
                                               const double maxValue, const XAxisUnit units)
    : PeakFindingStrategy(backgroundStrategy, spectrumInfo, minValue, maxValue, units) {}
 
PeakList StrongestPeaksStrategy::dofindSXPeaks(const HistogramData::HistogramX &x, const HistogramData::HistogramY &y,
                                               [[maybe_unused]] const HistogramData::HistogramE &e, Bound low,
                                               Bound high, const int workspaceIndex) const {
  auto distmin = std::distance(x.begin(), low);
  auto distmax = std::distance(x.begin(), high);
 
  // Find the max element
  auto maxY = (y.size() > 1) ? std::max_element(y.begin() + distmin, y.begin() + distmax) : y.begin();
 
  // Perform a check against the background
  double intensity = (*maxY);
  if (m_backgroundStrategy->isBelowBackground(intensity, y)) {
    return PeakList();
  }
 
  // Create the SXPeak information
  const auto distance = std::distance(y.begin(), maxY);
  const auto xValue = getXValue(x, distance);
  const auto tof = convertToTOF(xValue, workspaceIndex);
  const double phi = calculatePhi(workspaceIndex);
 
  std::vector<int> specs(1, workspaceIndex);
  std::vector<SXPeak> peaks;
  peaks.emplace_back(tof, phi, *maxY, specs, workspaceIndex, m_spectrumInfo);
  return peaks;
}
 
AllPeaksStrategy::AllPeaksStrategy(const BackgroundStrategy *backgroundStrategy, const API::SpectrumInfo &spectrumInfo,
                                   const double minValue, const double maxValue, const XAxisUnit units)
    : PeakFindingStrategy(backgroundStrategy, spectrumInfo, minValue, maxValue, units) {
  // We only allow the AbsoluteBackgroundStrategy for now
  if (!dynamic_cast<const AbsoluteBackgroundStrategy *>(m_backgroundStrategy)) {
    throw std::invalid_argument("The AllPeaksStrategy has to be initialized "
                                "with the AbsoluteBackgroundStrategy.");
  }
}
 
PeakList AllPeaksStrategy::dofindSXPeaks(const HistogramData::HistogramX &x, const HistogramData::HistogramY &y,
                                         [[maybe_unused]] const HistogramData::HistogramE &e, Bound low, Bound high,
                                         const int workspaceIndex) const {
  // Get all peaks from the container
  auto foundPeaks = getAllPeaks(x, y, low, high, m_backgroundStrategy);
 
  // Filter the found peaks having the mininum number of bins
  filterPeaksForMinBins(foundPeaks);
 
  // Convert the found peaks to SXPeaks
  auto peaks = convertToSXPeaks(x, y, foundPeaks, workspaceIndex);
 
  return peaks;
}
 
std::vector<std::unique_ptr<PeakContainer>>
AllPeaksStrategy::getAllPeaks(const HistogramData::HistogramX &x, const HistogramData::HistogramY &y, Bound low,
                              Bound high,
                              const Mantid::Crystal::FindSXPeaksHelper::BackgroundStrategy *backgroundStrategy) const {
  // We iterate over the data and only consider data which is above the
  // threshold.
  // Once data starts to be above the threshold we start to record it and add it
  // to a peak. Once it falls below, it concludes recording of that particular
  // peak
  bool isRecording = false;
 
  std::unique_ptr<PeakContainer> currentPeak = nullptr;
  std::vector<std::unique_ptr<PeakContainer>> peaks;
 
  // We want to the upper boundary to be inclusive hence we need to increment it
  // by one
  if (high != x.end()) {
    ++high;
  }
  auto distanceMin = std::distance(x.begin(), low);
  auto distanceMax = std::distance(x.begin(), high);
 
  const auto lowY = y.begin() + distanceMin;
  auto highY = distanceMax < static_cast<int>(y.size()) ? y.begin() + distanceMax : y.end();
 
  for (auto it = lowY; it != highY; ++it) {
    const auto signal = *it;
    const auto isAboveThreshold = !backgroundStrategy->isBelowBackground(signal, y);
 
    // There are four scenarios:
    // 1. Not recording + below threshold => continue
    // 2. Not recording + above treshold => start recording
    // 3. Recording + below threshold => stop recording
    // 4. Recording + above threshold => continue recording
    if (!isRecording && (!std::isfinite(signal) || !isAboveThreshold)) {
      continue;
    } else if (!isRecording && isAboveThreshold && std::isfinite(signal)) {
      // only start recording if is finite as NaN values will be found to be above threshold
      currentPeak = std::make_unique<PeakContainer>(y);
      currentPeak->startRecord(it);
      isRecording = true;
    } else if (isRecording && !isAboveThreshold) {
      currentPeak->stopRecord(it);
      peaks.emplace_back(std::move(currentPeak));
      currentPeak = nullptr;
      isRecording = false;
    } else {
      // this will continue to record NaN if previous point was above the background
      currentPeak->record(it);
    }
  }
 
  // Handle a peak on the edge if it exists
  if (isRecording) {
    if (highY == y.end()) {
      --highY;
    }
    currentPeak->stopRecord(highY);
    peaks.emplace_back(std::move(currentPeak));
    currentPeak = nullptr;
  }
 
  return peaks;
}
 
NSigmaPeaksStrategy::NSigmaPeaksStrategy(const API::SpectrumInfo &spectrumInfo, const double nsigma,
                                         const double minValue, const double maxValue, const XAxisUnit units)
    : PeakFindingStrategy(nullptr, spectrumInfo, minValue, maxValue, units), m_nsigma(nsigma) {}
 
PeakList NSigmaPeaksStrategy::dofindSXPeaks(const HistogramData::HistogramX &x, const HistogramData::HistogramY &y,
                                            const HistogramData::HistogramE &e, Bound low, Bound high,
                                            const int workspaceIndex) const {
  auto nsigmaPeaks = getAllNSigmaPeaks(x, y, e, low, high);
 
  // Filter the found peaks having the mininum number of bins
  filterPeaksForMinBins(nsigmaPeaks);
 
  auto sxPeaks = convertToSXPeaks(x, y, nsigmaPeaks, workspaceIndex);
  return sxPeaks;
}
 
std::vector<std::unique_ptr<PeakContainer>> NSigmaPeaksStrategy::getAllNSigmaPeaks(const HistogramData::HistogramX &x,
                                                                                   const HistogramData::HistogramY &y,
                                                                                   const HistogramData::HistogramE &e,
                                                                                   Bound low, Bound high) const {
  /*
  Credits to the author of SXD2001 for the idea of using NSigma as a threshold for peak finding:
    Gutmann, M. J. (2005). SXD2001. ISIS Facility, Rutherford Appleton Laboratory, Oxfordshire, England.
  */
 
  bool isRecording = false;
  std::unique_ptr<PeakContainer> currentPeak = nullptr;
  std::vector<std::unique_ptr<PeakContainer>> peaks;
 
  // We want to the upper boundary to be inclusive hence we need to increment by one
  if (high != x.end()) {
    ++high;
  }
  auto distanceMin = std::distance(x.begin(), low);
  auto distanceMax = std::distance(x.begin(), high);
 
  const auto lowY = y.begin() + distanceMin;
  auto highY = distanceMax < static_cast<int>(y.size()) ? y.begin() + distanceMax : y.end();
 
  const auto lowE = e.begin() + distanceMin;
  const auto highE = distanceMax < static_cast<int>(e.size()) ? e.begin() + distanceMax : e.begin();
 
  auto yIt = lowY + 1;
  auto eIt = lowE + 1;
 
  for (; yIt != highY && eIt != highE; ++yIt, ++eIt) {
    const auto signalDiff = *yIt - *(yIt - 1);
    const auto isStartPeak = signalDiff > (m_nsigma * (*eIt)) + NSIGMA_COMPARISON_THRESHOLD;
    const auto isSigDropSignificant = (signalDiff * (-1.)) > (m_nsigma * (*eIt)) + NSIGMA_COMPARISON_THRESHOLD;
    const auto isEndPeak = (currentPeak != nullptr ? (isSigDropSignificant || currentPeak->getStartingSignal() > *yIt)
                                                   : isSigDropSignificant);
 
    /*Possible scenarios
    1. isRecording is False and isStartPeak True(== isEndPeak is False) => start recording
    2. isRecording is True and isEndPeak False - continue recording
    3. isRecording is True and isEndpeak is True(== isStartpeak is False) - end peak
    4. else - continue
    */
    if (!isRecording && isStartPeak) {
      currentPeak = std::make_unique<PeakContainer>(y);
      currentPeak->startRecord(yIt);
      isRecording = true;
    } else if (isRecording && !isEndPeak) {
      currentPeak->record(yIt);
    } else if (isRecording && isEndPeak) {
      currentPeak->stopRecord(yIt);
      peaks.emplace_back(std::move(currentPeak));
      currentPeak = nullptr;
      isRecording = false;
    } else {
      continue;
    }
  }
 
  // Handle a peak on the edge if it exists
  if (isRecording) {
    if (highY == y.end()) {
      --highY;
    }
    currentPeak->stopRecord(highY);
    peaks.emplace_back(std::move(currentPeak));
    currentPeak = nullptr;
  }
 
  return peaks;
}
/* ------------------------------------------------------------------------------------------
 * PeakList Reduction Strategy
 * ------------------------------------------------------------------------------------------
 */
 
ReducePeakListStrategy::ReducePeakListStrategy(const CompareStrategy *compareStrategy)
    : m_compareStrategy(compareStrategy) {}
 
void ReducePeakListStrategy::setMinNSpectraPerPeak(int minNSpectraPerPeak) {
  m_minNSpectraPerPeak = minNSpectraPerPeak;
}
 
void ReducePeakListStrategy::setMaxNSpectraPerPeak(int maxSpectrasForPeak) {
  m_maxNSpectraPerPeak = maxSpectrasForPeak;
}
 
SimpleReduceStrategy::SimpleReduceStrategy(const CompareStrategy *compareStrategy)
    : ReducePeakListStrategy(compareStrategy) {}
 
std::vector<SXPeak> SimpleReduceStrategy::reduce(const std::vector<SXPeak> &peaks,
                                                 Mantid::Kernel::ProgressBase & /*progress*/) const {
  // If the peaks are empty then do nothing
  if (peaks.empty()) {
    return peaks;
  }
 
  std::vector<SXPeak> finalPeaks;
  for (const auto &currentPeak : peaks) {
    auto pos = std::find_if(finalPeaks.begin(), finalPeaks.end(), [&currentPeak, this](SXPeak &peak) {
      auto result = this->m_compareStrategy->compare(currentPeak, peak);
      // bool result = currentPeak.compare(peak,
      // resolution);
      if (result)
        peak += currentPeak;
      return result;
    });
    if (pos == finalPeaks.end()) {
      finalPeaks.emplace_back(currentPeak);
    }
  }
 
  reducePeaksFromNumberOfSpectras(finalPeaks);
 
  return finalPeaks;
}
 
void SimpleReduceStrategy::reducePeaksFromNumberOfSpectras(std::vector<SXPeak> &inputPeaks) const {
  if (m_minNSpectraPerPeak == EMPTY_INT() && m_maxNSpectraPerPeak == EMPTY_INT()) {
    return;
  }
 
  for (auto peakIt = inputPeaks.begin(); peakIt != inputPeaks.end();) {
    if (((m_minNSpectraPerPeak != EMPTY_INT()) &&
         (static_cast<int>((*peakIt).getPeakSpectras().size()) < m_minNSpectraPerPeak)) ||
        ((m_maxNSpectraPerPeak != EMPTY_INT()) &&
         (static_cast<int>((*peakIt).getPeakSpectras().size()) > m_maxNSpectraPerPeak))) {
      peakIt = inputPeaks.erase(peakIt);
    } else {
      ++peakIt;
    }
  }
}
 
FindMaxReduceStrategy::FindMaxReduceStrategy(const CompareStrategy *compareStrategy)
    : ReducePeakListStrategy(compareStrategy) {}
 
std::vector<SXPeak> FindMaxReduceStrategy::reduce(const std::vector<SXPeak> &peaks,
                                                  Mantid::Kernel::ProgressBase &progress) const {
  // If the peaks are empty then do nothing
  if (peaks.empty()) {
    return peaks;
  }
 
  // Groups the peaks into elements which are considered alike
  auto peakGroups = getPeakGroups(peaks, progress);
 
  // Now reduce the peaks groups
  return getFinalPeaks(peakGroups);
}
 
// Define some graph elements
using PeakGraph = adjacency_list<vecS, vecS, undirectedS, SXPeak *>;
using Vertex = boost::graph_traits<PeakGraph>::vertex_descriptor;
using Edge = boost::graph_traits<PeakGraph>::edge_descriptor;
 
std::vector<std::vector<SXPeak *>> FindMaxReduceStrategy::getPeakGroups(const std::vector<SXPeak> &peakList,
                                                                        Mantid::Kernel::ProgressBase &progress) const {
 
  // Create a vector of addresses. Note that the peaks live on the stack. This
  // here only works, because the peaks are always in a stack frame below.
  std::vector<SXPeak *> peaks;
  peaks.reserve(peakList.size());
  std::transform(peakList.cbegin(), peakList.cend(), std::back_inserter(peaks),
                 [](const auto &peak) { return &const_cast<SXPeak &>(peak); });
 
  // Add the peaks to a graph
  Edge edge;
  PeakGraph graph;
  PeakGraph::vertex_iterator vertexIt, vertexEnd;
 
  // Provide a warning if there are more than 500 peaks found.
  const size_t numberOfPeaksFound = peaks.size();
  if (numberOfPeaksFound > 500) {
    std::string warningMessage = std::string("There are ") + std::to_string(numberOfPeaksFound) +
                                 std::string(" peaks being processed. This might take a long time. "
                                             "Please check that the cutoff of the background that "
                                             "you have selected is high enough, else the algorithm will"
                                             " mistake background noise for peaks. The instrument view "
                                             "allows you to easily inspect the typical background level.");
    g_log.warning(warningMessage);
  }
 
  std::string message = std::string("There are ") + std::to_string(numberOfPeaksFound) +
                        std::string(" peaks. Investigating peak number ");
  int peakCounter = 0;
 
  for (auto peak : peaks) {
    ++peakCounter;
 
    // 1. Add the vertex
    auto vertex = add_vertex(peak, graph);
 
    // 2. Iterate over all elements already in the graph and check if they need
    // to edstablish an edge between them.
    std::tie(vertexIt, vertexEnd) = vertices(graph);
 
    // Provide a progress report such that users can escape the graph generation
    if (peakCounter > 50) {
      progress.doReport(message + std::to_string(peakCounter));
    }
 
    for (; vertexIt != vertexEnd; ++vertexIt) {
      // 2.1 Check if we are looking at the new vertex itself. We don't want
      // self-loops
      if (vertex == *vertexIt) {
        continue;
      }
 
      // 2.2 Check if the edge exists already
      if (boost::edge(vertex, *vertexIt, graph).second) {
        continue;
      }
 
      // 2.3 Check if the two vertices should have an edge
      const auto toCheck = graph[*vertexIt];
      if (m_compareStrategy->compare(*peak, *toCheck)) {
        // We need to create an edge
        add_edge(vertex, *vertexIt, graph);
      }
    }
  }
 
  // Create disjoined graphs from graph above
  std::vector<int> components(boost::num_vertices(graph));
  const int numberOfPeaks = connected_components(graph, &components[0]);
 
  std::vector<std::vector<SXPeak *>> peakGroups(numberOfPeaks);
 
  for (auto i = 0u; i < components.size(); ++i) {
    auto index = components[i];
    peakGroups[index].emplace_back(graph[i]);
  }
 
  return peakGroups;
}
 
std::vector<SXPeak> FindMaxReduceStrategy::getFinalPeaks(const std::vector<std::vector<SXPeak *>> &peakGroups) const {
  std::vector<SXPeak> peaks;
  // For each peak groupf find one peak
  // Currently we select the peak with the largest signal (this strategy could
  // be changed to something like a weighted mean or similar)
  for (const auto &group : peakGroups) {
    // When MinNSpectraPerPeak or maxNSpectraPerPeak parameters are provided,
    // a group will be ignored if it does not satisfy the minimum or the maximum number of spectrums
    // required to identify as a peak.
    if ((m_minNSpectraPerPeak != EMPTY_INT() && static_cast<int>(group.size()) < m_minNSpectraPerPeak) ||
        (m_maxNSpectraPerPeak != EMPTY_INT() && static_cast<int>(group.size()) > m_maxNSpectraPerPeak)) {
      continue;
    }
 
    SXPeak *maxPeak = nullptr;
    double maxIntensity = std::numeric_limits<double>::min();
 
    for (auto *element : group) {
      if (element->getIntensity() > maxIntensity) {
        maxIntensity = element->getIntensity();
        maxPeak = element;
      }
    }
 
    // Add the max peak if valid
    if (maxPeak) {
      // check not null as is case when intensity is NaN (though this shouldn't occur now)
      peaks.emplace_back(*maxPeak);
    }
  }
 
  return peaks;
}
 
/* ------------------------------------------------------------------------------------------
 * Comparison Strategy
 * ------------------------------------------------------------------------------------------
 */
RelativeCompareStrategy::RelativeCompareStrategy(const double resolution) : m_resolution(resolution) {}
 
bool RelativeCompareStrategy::compare(const SXPeak &lhs, const SXPeak &rhs) const {
  return lhs.compare(rhs, m_resolution);
}
 
AbsoluteCompareStrategy::AbsoluteCompareStrategy(const double xUnitResolution, const double phiResolution,
                                                 const double twoThetaResolution, const XAxisUnit units)
    : m_xUnitResolution(xUnitResolution), m_phiResolution(phiResolution), m_twoThetaResolution(twoThetaResolution),
      m_units(units) {
  // Convert the input from degree to radians
  constexpr double rad2deg = M_PI / 180.;
  m_phiResolution *= rad2deg;
  m_twoThetaResolution *= rad2deg;
}
 
bool AbsoluteCompareStrategy::compare(const SXPeak &lhs, const SXPeak &rhs) const {
  return lhs.compare(rhs, m_xUnitResolution, m_phiResolution, m_twoThetaResolution, m_units);
}
 
} // namespace Mantid::Crystal::FindSXPeaksHelper