PeakFitterMrqdt_tool.cc

Go to the documentation of this file.

#include "larcore/CoreUtils/ServiceUtil.h"
#include "larreco/HitFinder/HitFinderTools/IPeakFitter.h"
#include "larreco/RecoAlg/GausFitCache.h"      // hit::GausFitCache
#include "larvecutils/MarqFitAlg/MarqFitAlg.h" //marqfit functions

#include "art/Utilities/ToolMacros.h"
#include "cetlib_except/exception.h"
#include "larcore/Geometry/Geometry.h"
#include "messagefacility/MessageLogger/MessageLogger.h"

#include <cassert>
#include <cmath>
#include <fstream>

#include "larreco/HitFinder/HitFinderTools/ICandidateHitFinder.h"

namespace reco_tool {

  class PeakFitterMrqdt : IPeakFitter {
  public:
    explicit PeakFitterMrqdt(const fhicl::ParameterSet& pset);

    void findPeakParameters(const std::vector<float>&,
                            const ICandidateHitFinder::HitCandidateVec&,
                            PeakParamsVec&,
                            double&,
                            int&) const override;

  private:
    //Variables from the fhicl file
    const double fMinWidth;
    const double fMaxWidthMult;
    const double fPeakRange;
    const double fAmpRange;

    std::unique_ptr<gshf::MarqFitAlg> fMarqFitAlg;

    const geo::GeometryCore* fGeometry = lar::providerFrom<geo::Geometry>();
  };

  //--------------------------
  //constructor

  PeakFitterMrqdt::PeakFitterMrqdt(const fhicl::ParameterSet& pset)
    : fMinWidth(pset.get<double>("MinWidth", 0.5))
    , fMaxWidthMult(pset.get<double>("MaxWidthMult", 3.))
    , fPeakRange(pset.get<double>("PeakRangeFact", 2.))
    , fAmpRange(pset.get<double>("PeakAmpRange", 2.))
  {}

  //------------------------
  //output parameters should be replaced with a returned value
  void PeakFitterMrqdt::findPeakParameters(const std::vector<float>& signal,
                                           const ICandidateHitFinder::HitCandidateVec& fhc_vec,
                                           PeakParamsVec& mhpp_vec,
                                           double& chi2PerNDF,
                                           int& NDF) const
  {
    if (fhc_vec.empty()) return;

    const float chiCut = 1e-3;
    float lambda = 0.001; /* Marquardt damping parameter */
    float chiSqr = std::numeric_limits<float>::max(), dchiSqr = std::numeric_limits<float>::max();
    int nParams = 0;

    int startTime = fhc_vec.front().startTick;
    int endTime = fhc_vec.back().stopTick;

    int roiSize = endTime - startTime;

    // init to a large number in case the fit fails
    chi2PerNDF = chiSqr;

    std::vector<float> y(roiSize);
    std::vector<float> p(3 * fhc_vec.size());
    std::vector<float> plimmin(3 * fhc_vec.size());
    std::vector<float> plimmax(3 * fhc_vec.size());
    std::vector<float> perr(3 * fhc_vec.size());

    /* choose the fit function and set the parameters */
    nParams = 0;

    for (size_t ih = 0; ih < fhc_vec.size(); ih++) {
      float const peakMean =
        fhc_vec[ih].hitCenter - 0.5 - (float)startTime; //shift by 0.5 to account for bin width
      float const peakWidth = fhc_vec[ih].hitSigma;
      float const amplitude = fhc_vec[ih].hitHeight;
      float const meanLowLim = fmax(peakMean - fPeakRange * peakWidth, 0.);
      float const meanHiLim = fmin(peakMean + fPeakRange * peakWidth, (float)roiSize);
      p[0 + nParams] = amplitude;
      p[1 + nParams] = peakMean;
      p[2 + nParams] = peakWidth;

      plimmin[0 + nParams] = amplitude * 0.1;
      plimmin[1 + nParams] = meanLowLim;
      plimmin[2 + nParams] = std::max(fMinWidth, 0.1 * peakWidth);

      plimmax[0 + nParams] = amplitude * fAmpRange;
      plimmax[1 + nParams] = meanHiLim;
      plimmax[2 + nParams] = fMaxWidthMult * peakWidth;

      nParams += 3;
    }
    int fitResult = -1;

    for (size_t idx = 0; idx < size_t(roiSize); idx++) {
      y[idx] = signal[startTime + idx];
    }

    int trial = 0;
    lambda = -1.; /* initialize lambda on first call */
    do {
      fitResult = fMarqFitAlg->gshf::MarqFitAlg::mrqdtfit(
        lambda, &p[0], &plimmin[0], &plimmax[0], &y[0], nParams, roiSize, chiSqr, dchiSqr);
      trial++;
      if (fitResult || (trial > 100)) break;
    } while (fabs(dchiSqr) >= chiCut);

    if (!fitResult) {
      int fitResult2 =
        fMarqFitAlg->gshf::MarqFitAlg::cal_perr(&p[0], &y[0], nParams, roiSize, &perr[0]);
      if (!fitResult2) {
        NDF = roiSize - nParams;
        chi2PerNDF = chiSqr / NDF;
        int parIdx = 0;
        for (size_t i = 0; i < fhc_vec.size(); i++) {

          /* stand alone method
          mhpp_vec[i].peakAmplitude      = p[parIdx + 0];
          mhpp_vec[i].peakAmplitudeError = perr[parIdx + 0];
          mhpp_vec[i].peakCenter         = p[parIdx + 1] + 0.5 + float(startTime);
          mhpp_vec[i].peakCenterError    = perr[parIdx + 1];
          mhpp_vec[i].peakSigma          = p[parIdx + 2];
          mhpp_vec[i].peakSigmaError     = perr[parIdx + 2];
          */

          PeakFitParams_t mhpp;
          mhpp.peakAmplitude = p[parIdx + 0];
          mhpp.peakAmplitudeError = perr[parIdx + 0];
          mhpp.peakCenter =
            p[parIdx + 1] + 0.5 + float(startTime); //shift by 0.5 to account for bin width
          mhpp.peakCenterError = perr[parIdx + 1];
          mhpp.peakSigma = std::abs(p[parIdx + 2]);
          mhpp.peakSigmaError = perr[parIdx + 2];

          mhpp_vec.emplace_back(mhpp);

          parIdx += 3;
        }

        //      fitStat=0;
      }
    }
    return;
  }

  DEFINE_ART_CLASS_TOOL(PeakFitterMrqdt)
}