Functions
std::pair< GlobalPoint, double >	nearestPoint (const GlobalPoint &vertex, reco::Track iclus)

TransientVertex	weightedMeanOutlierRejection (const std::vector< std::pair< GlobalPoint, GlobalPoint >> &points, std::vector< reco::TransientTrack > iclus)

TransientVertex	weightedMeanOutlierRejectionBeamSpot (const std::vector< std::pair< GlobalPoint, GlobalPoint >> &points, std::vector< reco::TransientTrack > iclus, const reco::BeamSpot &beamSpot)

TransientVertex	weightedMeanOutlierRejectionVarianceAsError (const std::vector< std::pair< GlobalPoint, GlobalPoint >> &points, std::vector< std::vector< reco::TransientTrack >>::const_iterator iclus)

Variables
constexpr float	corr_x = 1.2

constexpr float	corr_x_bs = 1.0

constexpr float	corr_z = 1.4

constexpr int	maxIterations = 50

constexpr float	muSquare = 9.

constexpr float	precision = 1e-24

constexpr float	startError = 20.0

Function Documentation

◆ nearestPoint()

std::pair<GlobalPoint, double> WeightedMeanFitter::nearestPoint	(	const GlobalPoint &	vertex,
		reco::Track	iclus
	)

inline

Definition at line 21 of file WeightedMeanFitter.h.

References AlCaHLTBitMon_ParallelJobs::p, funct::pow(), reco::TrackBase::px(), reco::TrackBase::py(), reco::TrackBase::pz(), mathSSE::sqrt(), submitPVValidationJobs::t, bphysicsOniaDQM_cfi::vertex, btvMC_cff::vx, reco::TrackBase::vx(), btvMC_cff::vy, reco::TrackBase::vy(), btvMC_cff::vz, and reco::TrackBase::vz().

Referenced by weightedMeanOutlierRejection(), and weightedMeanOutlierRejectionBeamSpot().

                                                                                                {
     double ox = iclus.vx();
     double oy = iclus.vy();
     double oz = iclus.vz();
 
     double vx = iclus.px();
     double vy = iclus.py();
     double vz = iclus.pz();
 
     double opx = vertex.x() - ox;
     double opy = vertex.y() - oy;
     double opz = vertex.z() - oz;
 
     double vnorm2 = (vx * vx + vy * vy + vz * vz);
     double t = (vx * opx + vy * opy + vz * opz) / (vnorm2);
 
     GlobalPoint p(ox + t * vx, oy + t * vy, oz + t * vz);
     return std::pair<GlobalPoint, double>(
         p,
         std::sqrt(std::pow(p.x() - vertex.x(), 2) + std::pow(p.y() - vertex.y(), 2) + std::pow(p.z() - vertex.z(), 2)));
   }

◆ weightedMeanOutlierRejection()

TransientVertex WeightedMeanFitter::weightedMeanOutlierRejection	(	const std::vector< std::pair< GlobalPoint, GlobalPoint >> &	points,
		std::vector< reco::TransientTrack >	iclus
	)

inline

Definition at line 43 of file WeightedMeanFitter.h.

References funct::abs(), nano_mu_local_reco_cff::chi2, corr_x, corr_z, submitPVResolutionJobs::err, err_x, err_z, mps_fire::i, createfilelist::int, muSquare, ndof, nearestPoint(), AlCaHLTBitMon_ParallelJobs::p, hiPixelPairStep_cff::points, funct::pow(), precision, edm::second(), startError, HLT_2024v12_cff::track, findQualityFiles::v, and x.

Referenced by WeightedMeanPrimaryVertexEstimator::fit().

                                                                                            {
     float x = 0., y = 0., z = 0.;
     float s_wx = 0., s_wz = 0.;
     float s2_wx = 0., s2_wz = 0.;
     float wx = 0., wz = 0., chi2 = 0.;
     float ndof_x = 0.;
 
     AlgebraicSymMatrix33 err;
     err(0, 0) = startError / 10 * startError / 10;
     err(1, 1) = startError / 10 * startError / 10;
     err(2, 2) = startError * startError;  // error is 20 cm, so cov -> is 20 ^ 2
     for (const auto& p : points) {
       wx = p.second.x() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(p.second.x(), 2);
 
       wz = p.second.z() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(p.second.z(), 2);
 
       x += p.first.x() * wx;
       y += p.first.y() * wx;
       z += p.first.z() * wz;
 
       s_wx += wx;
       s_wz += wz;
     }
 
     if (s_wx == 0. || s_wz == 0.) {
       edm::LogWarning("WeightedMeanFitter") << "Vertex fitting failed at beginning\n";
       return TransientVertex(GlobalPoint(0, 0, 0), err, iclus, 0, 0);
     }
 
     x /= s_wx;
     y /= s_wx;
     z /= s_wz;
 
     float old_x, old_y, old_z;
 
     float xpull;
     int niter = 0;
 
     float err_x, err_z;
 
     err_x = 1. / s_wx;
     err_z = 1. / s_wz;
 
     while ((niter++) < 2) {
       old_x = x;
       old_y = y;
       old_z = z;
       s_wx = 0;
       s_wz = 0;
       s2_wx = 0;
       s2_wz = 0;
       x = 0;
       y = 0;
       z = 0;
       ndof_x = 0;
 
       for (unsigned int i = 0; i < (unsigned int)points.size(); i++) {
         std::pair<GlobalPoint, double> p = nearestPoint(GlobalPoint(old_x, old_y, old_z), (iclus)[i].track());
 
         wx = points[i].second.x() <= precision ? std::pow(precision, 2) : std::pow(points[i].second.x(), 2);
         wz = points[i].second.z() <= precision ? std::pow(precision, 2) : std::pow(points[i].second.z(), 2);
 
         xpull = 0.;
 
         if (std::pow(p.first.x() - old_x, 2) / (wx + err_x) < muSquare &&
             std::pow(p.first.y() - old_y, 2) / (wx + err_x) < muSquare &&
             std::pow(p.first.z() - old_z, 2) / (wz + err_z) < muSquare)
           xpull = 1.;
 
         ndof_x += xpull;
 
         wx = xpull / wx;
         wz = xpull / wz;
 
         x += wx * p.first.x();
         y += wx * p.first.y();
         z += wz * p.first.z();
 
         s_wx += wx;
         s_wz += wz;
 
         s2_wx += wx * xpull;
         s2_wz += wz * xpull;
       }
 
       if (s_wx == 0. || s_wz == 0.) {
         edm::LogWarning("WeightedMeanFitter")
             << "Vertex fitting failed, either all tracks are outliers or they have a very large error\n";
         return TransientVertex(GlobalPoint(0, 0, 0), err, iclus, 0, 0);
       }
       x /= s_wx;
       y /= s_wx;
       z /= s_wz;
 
       err_x = (s2_wx / std::pow(s_wx, 2));
       err_z = (s2_wz / std::pow(s_wz, 2));
 
       if (std::abs(x - old_x) < (precision / 1.) && std::abs(y - old_y) < (precision / 1.) &&
           std::abs(z - old_z) < (precision / 1.)) {
         break;
       }
     }
 
     err(0, 0) = err_x * corr_x * corr_x;
     err(1, 1) = err_x * corr_x * corr_x;
     err(2, 2) = err_z * corr_z * corr_z;
 
     float dist = 0;
     for (const auto& p : points) {
       wx = p.second.x();
       wx = wx <= precision ? precision : wx;
 
       wz = p.second.z();
       wz = wz <= precision ? precision : wz;
 
       dist = std::pow(p.first.x() - x, 2) / (std::pow(wx, 2) + err(0, 0));
       dist += std::pow(p.first.y() - y, 2) / (std::pow(wx, 2) + err(1, 1));
       dist += std::pow(p.first.z() - z, 2) / (std::pow(wz, 2) + err(2, 2));
       chi2 += dist;
     }
     float ndof =
         ndof_x > 1 ? (2 * ndof_x - 3) : 0.00001;  // ndof_x is actually the number of tracks with non-zero weight
     TransientVertex v(GlobalPoint(x, y, z), err, iclus, chi2, ndof);
     return v;
   }

◆ weightedMeanOutlierRejectionBeamSpot()

TransientVertex WeightedMeanFitter::weightedMeanOutlierRejectionBeamSpot	(	const std::vector< std::pair< GlobalPoint, GlobalPoint >> &	points,
		std::vector< reco::TransientTrack >	iclus,
		const reco::BeamSpot &	beamSpot
	)

inline

Definition at line 170 of file WeightedMeanFitter.h.

References funct::abs(), pwdgSkimBPark_cfi::beamSpot, nano_mu_local_reco_cff::chi2, corr_x_bs, corr_z, submitPVResolutionJobs::err, err_x, err_y, err_z, mps_fire::i, createfilelist::int, muSquare, nearestPoint(), AlCaHLTBitMon_ParallelJobs::p, hiPixelPairStep_cff::points, funct::pow(), precision, edm::second(), startError, HLT_2024v12_cff::track, findQualityFiles::v, and x.

Referenced by WeightedMeanPrimaryVertexEstimator::fit().

                                     {
     float x = 0., y = 0., z = 0.;
     float s_wx = 0., s_wz = 0.;
     float s2_wx = 0., s2_wz = 0.;
     float wx = 0., wz = 0., chi2 = 0.;
     float wy = 0., s_wy = 0., s2_wy = 0.;
     float ndof_x = 0.;
 
     AlgebraicSymMatrix33 err;
     err(0, 0) = startError / 10 * startError / 10;
     err(1, 1) = startError / 10 * startError / 10;
     err(2, 2) = startError * startError;  // error is 20 cm, so cov -> is 20 ^ 2
 
     GlobalError bse(beamSpot.rotatedCovariance3D());
     GlobalPoint bsp(Basic3DVector<float>(beamSpot.position()));
 
     for (const auto& p : points) {
       wx = p.second.x() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(p.second.x(), 2);
       wy = p.second.y() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(p.second.y(), 2);
 
       wz = p.second.z() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(p.second.z(), 2);
 
       x += p.first.x() * wx;
       y += p.first.y() * wy;
       z += p.first.z() * wz;
 
       s_wx += wx;
       s_wy += wy;
       s_wz += wz;
     }
 
     if (s_wx == 0. || s_wy == 0. || s_wz == 0.) {
       edm::LogWarning("WeightedMeanFitter") << "Vertex fitting failed at beginning\n";
       return TransientVertex(GlobalPoint(0, 0, 0), err, iclus, 0, 0);
     }
     // use the square of covariance element to increase it's weight: it will be the most important
     wx = bse.cxx() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(bse.cxx(), 2);
     wy = bse.cyy() <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(bse.cyy(), 2);
 
     x += bsp.x() * wx;
     y += bsp.y() * wy;
 
     x /= (s_wx + wx);
     y /= (s_wy + wy);
     z /= s_wz;
 
     float old_x, old_y, old_z;
 
     float xpull;
     int niter = 0;
 
     float err_x, err_y, err_z;
 
     err_x = 1. / s_wx;
     err_y = 1. / s_wy;
     err_z = 1. / s_wz;
 
     while ((niter++) < 2) {
       old_x = x;
       old_y = y;
       old_z = z;
       s_wx = 0;
       s_wz = 0;
       s2_wx = 0;
       s2_wz = 0;
 
       s_wy = 0;
       s2_wy = 0;
 
       x = 0;
       y = 0;
       z = 0;
       ndof_x = 0;
 
       for (unsigned int i = 0; i < (unsigned int)points.size(); i++) {
         std::pair<GlobalPoint, double> p = nearestPoint(GlobalPoint(old_x, old_y, old_z), (iclus)[i].track());
 
         wx = points[i].second.x() <= precision ? std::pow(precision, 2) : std::pow(points[i].second.x(), 2);
         wy = points[i].second.y() <= precision ? std::pow(precision, 2) : std::pow(points[i].second.y(), 2);
 
         wz = points[i].second.z() <= precision ? std::pow(precision, 2) : std::pow(points[i].second.z(), 2);
 
         xpull = 0.;
         if (std::pow(p.first.x() - old_x, 2) / (wx + err_x) < muSquare &&
             std::pow(p.first.y() - old_y, 2) / (wy + err_y) < muSquare &&
             std::pow(p.first.z() - old_z, 2) / (wz + err_z) < muSquare)
           xpull = 1.;
 
         ndof_x += xpull;
 
         wx = xpull / wx;
         wy = xpull / wy;
         wz = xpull / wz;
 
         x += wx * p.first.x();
         y += wy * p.first.y();
         z += wz * p.first.z();
 
         s_wx += wx;
         s_wy += wy;
         s_wz += wz;
 
         s2_wx += wx * xpull;
         s2_wy += wy * xpull;
         s2_wz += wz * xpull;
       }
 
       if (s_wx == 0. || s_wy == 0. || s_wz == 0.) {
         edm::LogWarning("WeightedMeanFitter")
             << "Vertex fitting failed, either all tracks are outliers or they have a very large error\n";
         return TransientVertex(GlobalPoint(0, 0, 0), err, iclus, 0, 0);
       }
       wx = bse.cxx() <= std::pow(precision, 2) ? 1. / std::pow(precision, 2) : 1. / bse.cxx();
       wy = bse.cyy() <= std::pow(precision, 2) ? 1. / std::pow(precision, 2) : 1. / bse.cyy();
 
       x += bsp.x() * wx;
       y += bsp.y() * wy;
       s_wx += wx;
       s2_wx += wx;
       s_wy += wy;
       s2_wy += wy;
 
       x /= s_wx;
       y /= s_wy;
       z /= s_wz;
 
       err_x = (s2_wx / std::pow(s_wx, 2));
       err_y = (s2_wy / std::pow(s_wy, 2));
       err_z = (s2_wz / std::pow(s_wz, 2));
 
       if (std::abs(x - old_x) < (precision) && std::abs(y - old_y) < (precision) && std::abs(z - old_z) < (precision)) {
         break;
       }
     }
     err(0, 0) = err_x * corr_x_bs * corr_x_bs;
     err(1, 1) = err_y * corr_x_bs * corr_x_bs;
     err(2, 2) = err_z * corr_z * corr_z;
 
     float dist = 0;
     for (const auto& p : points) {
       wx = p.second.x();
       wx = wx <= precision ? precision : wx;
 
       wz = p.second.z();
       wz = wz <= precision ? precision : wz;
 
       dist = std::pow(p.first.x() - x, 2) / (std::pow(wx, 2) + err(0, 0));
       dist += std::pow(p.first.y() - y, 2) / (std::pow(wx, 2) + err(1, 1));
       dist += std::pow(p.first.z() - z, 2) / (std::pow(wz, 2) + err(2, 2));
       chi2 += dist;
     }
     TransientVertex v(GlobalPoint(x, y, z), err, iclus, chi2, (int)ndof_x);
     return v;
   }

◆ weightedMeanOutlierRejectionVarianceAsError()

TransientVertex WeightedMeanFitter::weightedMeanOutlierRejectionVarianceAsError	(	const std::vector< std::pair< GlobalPoint, GlobalPoint >> &	points,
		std::vector< std::vector< reco::TransientTrack >>::const_iterator	iclus
	)

inline

Definition at line 328 of file WeightedMeanFitter.h.

References funct::abs(), nano_mu_local_reco_cff::chi2, submitPVResolutionJobs::err, err_x, err_y, err_z, JetChargeProducer_cfi::exp, maxIterations, muSquare, AlCaHLTBitMon_ParallelJobs::p, hiPixelPairStep_cff::points, funct::pow(), precision, startError, findQualityFiles::v, and x.

                                                                     {
     float x = 0, y = 0, z = 0, s_wx = 0, s_wy = 0, s_wz = 0, wx = 0, wy = 0, wz = 0, chi2 = 0;
     float ndof_x = 0;
     AlgebraicSymMatrix33 err;
     err(2, 2) = startError * startError;  // error is 20 cm, so cov -> is 20 ^ 2
     err(0, 0) = err(1, 1) = err(2, 2) / 100.;
 
     for (const auto& p : points) {
       wx = p.second.x();
       wx = wx <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(wx, 2);
 
       wz = p.second.z();
       wz = wz <= precision ? 1. / std::pow(precision, 2) : 1. / std::pow(wz, 2);
 
       x += p.first.x() * wx;
       y += p.first.y() * wx;
       z += p.first.z() * wz;
 
       s_wx += wx;
       s_wz += wz;
     }
 
     if (s_wx == 0. || s_wz == 0.) {
       edm::LogWarning("WeightedMeanFitter") << "Vertex fitting failed at beginning\n";
       return TransientVertex(GlobalPoint(0, 0, 0), err, *iclus, 0, 0);
     }
 
     x /= s_wx;
     y /= s_wx;
     z /= s_wz;
 
     float old_x, old_y, old_z;
     float xpull;
     int niter = 0;
     float err_x, err_y, err_z;
     err_x = 1. / s_wx;
     err_y = 1. / s_wx;
     err_z = 1. / s_wz;
     float s_err_x = 0., s_err_y = 0., s_err_z = 0.;
     while ((niter++) < maxIterations) {
       old_x = x;
       old_y = y;
       old_z = z;
       s_wx = 0;
       s_wy = 0;
       s_wz = 0;
       x = 0;
       y = 0;
       z = 0;
       s_err_x = 0.;
       s_err_y = 0.;
       s_err_z = 0.;
 
       for (const auto& p : points) {
         wx = p.second.x();
         wx = wx <= precision ? precision : wx;
 
         wy = wx * wx + err_y;
         wx = wx * wx + err_x;
 
         wz = p.second.z();
         wz = wz <= precision ? precision : wz;
         wz = wz * wz + err_z;
 
         xpull = std::pow((p.first.x() - old_x), 2) / wx;
         xpull += std::pow((p.first.y() - old_y), 2) / wy;
         xpull += std::pow((p.first.z() - old_z), 2) / wz;
         xpull = 1. / (1. + std::exp(-0.5 * ((muSquare)-xpull)));
         ndof_x += xpull;
 
         wx = 1. / wx;
         wy = 1. / wy;
         wz = 1. / wz;
 
         wx *= xpull;
         wy *= xpull;
         wz *= xpull;
 
         x += wx * p.first.x();
         y += wy * p.first.y();
         z += wz * p.first.z();
 
         s_wx += wx;
         s_wy += wy;
         s_wz += wz;
 
         s_err_x += wx * pow(p.first.x() - old_x, 2);
         s_err_y += wy * pow(p.first.y() - old_y, 2);
         s_err_z += wz * pow(p.first.z() - old_z, 2);
       }
       if (s_wx == 0. || s_wy == 0. || s_wz == 0.) {
         edm::LogWarning("WeightedMeanFitter")
             << "Vertex fitting failed, either all tracks are outliers or they have a very large error\n";
         return TransientVertex(GlobalPoint(0, 0, 0), err, *iclus, 0, 0);
       }
       x /= s_wx;
       y /= s_wy;
       z /= s_wz;
 
       err_x = s_err_x / s_wx;
       err_y = s_err_y / s_wy;
       err_z = s_err_z / s_wz;
 
       if (std::abs(x - old_x) < (precision / 1.) && std::abs(y - old_y) < (precision / 1.) &&
           std::abs(z - old_z) < (precision / 1.))
         break;
     }
 
     err(0, 0) = err_x;
     err(1, 1) = err_y;
     err(2, 2) = err_z;
 
     float dist = 0.f;
     for (const auto& p : points) {
       wx = p.second.x();
       wx = wx <= precision ? precision : wx;
 
       wz = p.second.z();
       wz = wz <= precision ? precision : wz;
 
       dist = std::pow(p.first.x() - x, 2) / (std::pow(wx, 2) + std::pow(err(0, 0), 2));
       dist += std::pow(p.first.y() - y, 2) / (std::pow(wx, 2) + std::pow(err(1, 1), 2));
       dist += std::pow(p.first.z() - z, 2) / (std::pow(wz, 2) + std::pow(err(2, 2), 2));
       chi2 += dist;
     }
     TransientVertex v(GlobalPoint(x, y, z), err, *iclus, chi2, (int)ndof_x);
     return v;
   }