#include <L1TrackQuality.h>

Public Types
enum	QualityAlgorithm { QualityAlgorithm::Cut, QualityAlgorithm::GBDT, QualityAlgorithm::NN, QualityAlgorithm::None }

Public Member Functions
std::vector< float >	featureTransform (TTTrack< Ref_Phase2TrackerDigi_ > &aTrack, std::vector< std::string > const &featureNames)

	L1TrackQuality ()

	L1TrackQuality (const edm::ParameterSet &qualityParams)

void	setCutParameters (std::string const &AlgorithmString, float maxZ0, float maxEta, float chi2dofMax, float bendchi2Max, float minPt, int nStubmin)

void	setL1TrackQuality (TTTrack< Ref_Phase2TrackerDigi_ > &aTrack)

void	setONNXModel (std::string const &AlgorithmString, edm::FileInPath const &ONNXmodel, std::string const &ONNXInputName, std::vector< std::string > const &featureNames)

	~L1TrackQuality ()=default

Private Attributes
float	bendchi2Max_

float	chi2dofMax_

std::vector< std::string >	featureNames_

float	maxEta_

float	maxZ0_

float	minPt_

int	nStubsmin_

std::string	ONNXInputName_

float	ONNXInvRScaling_

edm::FileInPath	ONNXmodel_

QualityAlgorithm	qualityAlgorithm_ = QualityAlgorithm::None

Detailed Description

Definition at line 27 of file L1TrackQuality.h.

Member Enumeration Documentation

◆ QualityAlgorithm

enum L1TrackQuality::QualityAlgorithm

strong

Enumerator
Cut
GBDT
NN
None

Definition at line 30 of file L1TrackQuality.h.

30 { Cut, GBDT, NN, None };

mkfit::NN

constexpr Matriplex::idx_t NN

Definition: Matrix.h:43

None

Definition: APVGainStruct.h:56

Constructor & Destructor Documentation

◆ L1TrackQuality() [1/2]

L1TrackQuality::L1TrackQuality ( )

Definition at line 12 of file L1TrackQuality.cc.

12 {}

◆ L1TrackQuality() [2/2]

L1TrackQuality::L1TrackQuality ( const edm::ParameterSet & qualityParams )

Definition at line 14 of file L1TrackQuality.cc.

References edm::ParameterSet::getParameter(), ONNXInvRScaling_, setCutParameters(), setONNXModel(), and AlCaHLTBitMon_QueryRunRegistry::string.

                                                                    {
   std::string AlgorithmString = qualityParams.getParameter<std::string>("qualityAlgorithm");
   // Unpacks EDM parameter set itself to save unecessary processing within TrackProducers
   if (AlgorithmString == "Cut") {
     setCutParameters(AlgorithmString,
                      (float)qualityParams.getParameter<double>("maxZ0"),
                      (float)qualityParams.getParameter<double>("maxEta"),
                      (float)qualityParams.getParameter<double>("chi2dofMax"),
                      (float)qualityParams.getParameter<double>("bendchi2Max"),
                      (float)qualityParams.getParameter<double>("minPt"),
                      qualityParams.getParameter<int>("nStubsmin"));
   }
 
   else {
     setONNXModel(AlgorithmString,
                  qualityParams.getParameter<edm::FileInPath>("ONNXmodel"),
                  qualityParams.getParameter<std::string>("ONNXInputName"),
                  qualityParams.getParameter<std::vector<std::string>>("featureNames"));
     ONNXInvRScaling_ = qualityParams.getParameter<double>("ONNXInvRScale");
   }
 }

◆ ~L1TrackQuality()

L1TrackQuality::~L1TrackQuality ( )

default

Member Function Documentation

◆ featureTransform()

std::vector< float > L1TrackQuality::featureTransform	(	TTTrack< Ref_Phase2TrackerDigi_ > &	aTrack,
		std::vector< std::string > const &	featureNames
	)

Definition at line 36 of file L1TrackQuality.cc.

References funct::abs(), TTTrack< T >::chi2(), TTTrack< T >::chi2XY(), TTTrack< T >::chi2Z(), PVValHelper::eta, TTTrack< T >::eta(), TrackQualityParams_cfi::featureNames, dqmMemoryStats::float, TTTrack< T >::getStubRefs(), TTTrack< T >::hitPattern(), mps_fire::i, dqmiolumiharvest::j, dqmdumpme::k, dqm-mbProfile::log, TTTrack< T >::momentum(), ONNXInvRScaling_, PV3DBase< T, PVType, FrameType >::perp(), TTTrack< T >::phi(), TTTrack< T >::rInv(), cmsswSequenceInfo::seq, AlCaHLTBitMon_QueryRunRegistry::string, TTTrack< T >::stubPtConsistency(), TTTrack< T >::tanL(), and TTTrack< T >::z0().

Referenced by setL1TrackQuality().

                                                                                               {
   // List input features for MVA in proper order below, the features options are
   // {"log_chi2","log_chi2rphi","log_chi2rz","log_bendchi2","nstubs","lay1_hits","lay2_hits",
   // "lay3_hits","lay4_hits","lay5_hits","lay6_hits","disk1_hits","disk2_hits","disk3_hits",
   // "disk4_hits","disk5_hits","rinv","tanl","z0","dtot","ltot","chi2","chi2rz","chi2rphi",
   // "bendchi2","pt","eta","nlaymiss_interior","phi","bendchi2_bin","chi2rz_bin","chi2rphi_bin"}
 
   std::vector<float> transformedFeatures;
 
   // Define feature map, filled as features are generated
   std::map<std::string, float> feature_map;
 
   // The following converts the 7 bit hitmask in the TTTrackword to an expected
   // 11 bit hitmask based on the eta of the track
   std::vector<int> hitpattern_binary = {0, 0, 0, 0, 0, 0, 0};
   std::vector<int> hitpattern_expanded_binary = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
   std::vector<float> eta_bins = {0.0, 0.2, 0.41, 0.62, 0.9, 1.26, 1.68, 2.08, 2.4};
 
   // Expected hitmap table, each row corresponds to an eta bin, each value corresponds to
   // the expected layer in the expanded hit pattern. The expanded hit pattern should be
   // 11 bits but contains a 12th element so this hitmap table is symmetric
   int hitmap[8][7] = {{0, 1, 2, 3, 4, 5, 11},
                       {0, 1, 2, 3, 4, 5, 11},
                       {0, 1, 2, 3, 4, 5, 11},
                       {0, 1, 2, 3, 4, 5, 11},
                       {0, 1, 2, 3, 4, 5, 11},
                       {0, 1, 2, 6, 7, 8, 9},
                       {0, 1, 7, 8, 9, 10, 11},
                       {0, 6, 7, 8, 9, 10, 11}};
 
   // iterate through bits of the hitpattern and compare to 1 filling the hitpattern binary vector
   int tmp_trk_hitpattern = aTrack.hitPattern();
   for (int i = 6; i >= 0; i--) {
     int k = tmp_trk_hitpattern >> i;
     if (k & 1)
       hitpattern_binary[i] = 1;
   }
 
   // calculate number of missed interior layers from hitpattern
   int nbits = floor(log2(tmp_trk_hitpattern)) + 1;
   int lay_i = 0;
   int tmp_trk_nlaymiss_interior = 0;
   bool seq = false;
   for (int i = 0; i < nbits; i++) {
     lay_i = ((1 << i) & tmp_trk_hitpattern) >> i;  //0 or 1 in ith bit (right to left)
 
     if (lay_i && !seq)
       seq = true;  //sequence starts when first 1 found
     if (!lay_i && seq)
       tmp_trk_nlaymiss_interior++;
   }
 
   float eta = abs(aTrack.eta());
   int eta_size = static_cast<int>(eta_bins.size());
   // First iterate through eta bins
 
   for (int j = 1; j < eta_size; j++) {
     if (eta < eta_bins[j] && eta >= eta_bins[j - 1])  // if track in eta bin
     {
       // Iterate through hitpattern binary
       for (int k = 0; k <= 6; k++)
         // Fill expanded binary entries using the expected hitmap table positions
         hitpattern_expanded_binary[hitmap[j - 1][k]] = hitpattern_binary[k];
       break;
     }
   }
 
   int tmp_trk_ltot = 0;
   //calculate number of layer hits
   for (int i = 0; i < 6; ++i) {
     tmp_trk_ltot += hitpattern_expanded_binary[i];
   }
 
   int tmp_trk_dtot = 0;
   //calculate number of disk hits
   for (int i = 6; i < 11; ++i) {
     tmp_trk_dtot += hitpattern_expanded_binary[i];
   }
 
   // bin bendchi2 variable (bins from https://twiki.cern.ch/twiki/bin/viewauth/CMS/HybridDataFormat#Fitted_Tracks_written_by_KalmanF)
   float tmp_trk_bendchi2 = aTrack.stubPtConsistency();
   std::array<float, 8> bendchi2_bins{{0, 0.5, 1.25, 2, 3, 5, 10, 50}};
   int n_bendchi2 = static_cast<int>(bendchi2_bins.size());
   float tmp_trk_bendchi2_bin = -1;
   for (int i = 0; i < (n_bendchi2 - 1); i++) {
     if (tmp_trk_bendchi2 >= bendchi2_bins[i] && tmp_trk_bendchi2 < bendchi2_bins[i + 1]) {
       tmp_trk_bendchi2_bin = i;
       break;
     }
   }
   if (tmp_trk_bendchi2_bin < 0)
     tmp_trk_bendchi2_bin = n_bendchi2;
 
   // bin chi2rphi variable (bins from https://twiki.cern.ch/twiki/bin/viewauth/CMS/HybridDataFormat#Fitted_Tracks_written_by_KalmanF)
   float tmp_trk_chi2rphi = aTrack.chi2XY();
   std::array<float, 16> chi2rphi_bins{{0, 0.25, 0.5, 1, 2, 3, 5, 7, 10, 20, 40, 100, 200, 500, 1000, 3000}};
   int n_chi2rphi = static_cast<int>(chi2rphi_bins.size());
   float tmp_trk_chi2rphi_bin = -1;
   for (int i = 0; i < (n_chi2rphi - 1); i++) {
     if (tmp_trk_chi2rphi >= chi2rphi_bins[i] && tmp_trk_chi2rphi < chi2rphi_bins[i + 1]) {
       tmp_trk_chi2rphi_bin = i;
       break;
     }
   }
   if (tmp_trk_chi2rphi_bin < 0)
     tmp_trk_chi2rphi_bin = n_chi2rphi;
 
   // bin chi2rz variable (bins from https://twiki.cern.ch/twiki/bin/viewauth/CMS/HybridDataFormat#Fitted_Tracks_written_by_KalmanF)
   float tmp_trk_chi2rz = aTrack.chi2Z();
   std::array<float, 16> chi2rz_bins{{0, 0.25, 0.5, 1, 2, 3, 5, 7, 10, 20, 40, 100, 200, 500, 1000, 3000}};
   int n_chi2rz = static_cast<int>(chi2rz_bins.size());
   float tmp_trk_chi2rz_bin = -1;
   for (int i = 0; i < (n_chi2rz - 1); i++) {
     if (tmp_trk_chi2rz >= chi2rz_bins[i] && tmp_trk_chi2rz < chi2rz_bins[i + 1]) {
       tmp_trk_chi2rz_bin = i;
       break;
     }
   }
   if (tmp_trk_chi2rz_bin < 0)
     tmp_trk_chi2rz_bin = n_chi2rz;
 
   // get the nstub
   std::vector<edm::Ref<edmNew::DetSetVector<TTStub<Ref_Phase2TrackerDigi_>>, TTStub<Ref_Phase2TrackerDigi_>>> stubRefs =
       aTrack.getStubRefs();
 
   // fill the feature map
   feature_map["nstub"] = stubRefs.size();
   feature_map["rinv"] = ONNXInvRScaling_ * abs(aTrack.rInv());
   feature_map["tanl"] = abs(aTrack.tanL());
   feature_map["z0"] = aTrack.z0();
   feature_map["phi"] = aTrack.phi();
   feature_map["pt"] = aTrack.momentum().perp();
   feature_map["eta"] = aTrack.eta();
 
   float tmp_trk_chi2 = aTrack.chi2();
   feature_map["chi2"] = tmp_trk_chi2;
   feature_map["log_chi2"] = log(tmp_trk_chi2);
 
   feature_map["chi2rphi"] = tmp_trk_chi2rphi;
   feature_map["log_chi2rphi"] = log(tmp_trk_chi2rphi);
 
   feature_map["chi2rz"] = tmp_trk_chi2rz;
   feature_map["log_chi2rz"] = log(tmp_trk_chi2rz);
 
   feature_map["chi2rz"] = tmp_trk_chi2rz;
   feature_map["log_chi2rz"] = log(tmp_trk_chi2rz);
 
   feature_map["bendchi2"] = tmp_trk_bendchi2;
   feature_map["log_bendchi2"] = log(tmp_trk_bendchi2);
 
   feature_map["lay1_hits"] = float(hitpattern_expanded_binary[0]);
   feature_map["lay2_hits"] = float(hitpattern_expanded_binary[1]);
   feature_map["lay3_hits"] = float(hitpattern_expanded_binary[2]);
   feature_map["lay4_hits"] = float(hitpattern_expanded_binary[3]);
   feature_map["lay5_hits"] = float(hitpattern_expanded_binary[4]);
   feature_map["lay6_hits"] = float(hitpattern_expanded_binary[5]);
   feature_map["disk1_hits"] = float(hitpattern_expanded_binary[6]);
   feature_map["disk2_hits"] = float(hitpattern_expanded_binary[7]);
   feature_map["disk3_hits"] = float(hitpattern_expanded_binary[8]);
   feature_map["disk4_hits"] = float(hitpattern_expanded_binary[9]);
   feature_map["disk5_hits"] = float(hitpattern_expanded_binary[10]);
 
   feature_map["dtot"] = float(tmp_trk_dtot);
   feature_map["ltot"] = float(tmp_trk_ltot);
 
   feature_map["nlaymiss_interior"] = float(tmp_trk_nlaymiss_interior);
   feature_map["bendchi2_bin"] = tmp_trk_bendchi2_bin;
   feature_map["chi2rphi_bin"] = tmp_trk_chi2rphi_bin;
   feature_map["chi2rz_bin"] = tmp_trk_chi2rz_bin;
 
   // fill tensor with track params
   transformedFeatures.reserve(featureNames.size());
   for (const std::string& feature : featureNames)
     transformedFeatures.push_back(feature_map[feature]);
 
   return transformedFeatures;
 }

◆ setCutParameters()

void L1TrackQuality::setCutParameters	(	std::string const &	AlgorithmString,
		float	maxZ0,
		float	maxEta,
		float	chi2dofMax,
		float	bendchi2Max,
		float	minPt,
		int	nStubmin
	)

Definition at line 275 of file L1TrackQuality.cc.

References L1TPhase2Offline_cfi::bendchi2Max, bendchi2Max_, L1TPhase2Offline_cfi::chi2dofMax, chi2dofMax_, Cut, maxEta, maxEta_, L1TPhase2Offline_cfi::maxZ0, maxZ0_, beam_dqm_sourceclient-live_cfg::minPt, minPt_, nStubsmin_, and qualityAlgorithm_.

Referenced by L1TrackQuality().

                                                     {
   qualityAlgorithm_ = QualityAlgorithm::Cut;
   maxZ0_ = maxZ0;
   maxEta_ = maxEta;
   chi2dofMax_ = chi2dofMax;
   bendchi2Max_ = bendchi2Max;
   minPt_ = minPt;
   nStubsmin_ = nStubmin;
 }

◆ setL1TrackQuality()

void L1TrackQuality::setL1TrackQuality ( TTTrack< Ref_Phase2TrackerDigi_ > & aTrack )

Definition at line 215 of file L1TrackQuality.cc.

References funct::abs(), bendchi2Max_, TTTrack< T >::chi2(), chi2dofMax_, Cut, PV3DBase< T, PVType, FrameType >::eta(), featureNames_, featureTransform(), edm::FileInPath::fullPath(), GBDT, cms::Ort::ONNXRuntime::getOutputNames(), TTTrack< T >::getStubRefs(), maxEta_, maxZ0_, minPt_, TTTrack< T >::momentum(), NN, nStubsmin_, ONNXInputName_, ONNXmodel_, PV3DBase< T, PVType, FrameType >::perp(), qualityAlgorithm_, TTTrack< T >::settrkMVA1(), TTTrack< T >::stubPtConsistency(), and TTTrack< T >::z0().

                                                                               {
   if (this->qualityAlgorithm_ == QualityAlgorithm::Cut) {
     // Get Track parameters
     float trk_pt = aTrack.momentum().perp();
     float trk_bend_chi2 = aTrack.stubPtConsistency();
     float trk_z0 = aTrack.z0();
     float trk_eta = aTrack.momentum().eta();
     float trk_chi2 = aTrack.chi2();
     const auto& stubRefs = aTrack.getStubRefs();
     int nStubs = stubRefs.size();
 
     float classification = 0.0;  // Default classification is 0
 
     if (trk_pt >= this->minPt_ && abs(trk_z0) < this->maxZ0_ && abs(trk_eta) < this->maxEta_ &&
         trk_chi2 < this->chi2dofMax_ && trk_bend_chi2 < this->bendchi2Max_ && nStubs >= this->nStubsmin_)
       classification = 1.0;
     // Classification updated to 1 if conditions are met
 
     aTrack.settrkMVA1(classification);
   }
 
   if ((this->qualityAlgorithm_ == QualityAlgorithm::NN) || (this->qualityAlgorithm_ == QualityAlgorithm::GBDT)) {
     // Setup ONNX input and output names and arrays
     std::vector<std::string> ortinput_names;
     std::vector<std::string> ortoutput_names;
 
     cms::Ort::FloatArrays ortinput;
     cms::Ort::FloatArrays ortoutputs;
 
     std::vector<float> Transformed_features = featureTransform(aTrack, this->featureNames_);
     cms::Ort::ONNXRuntime Runtime(this->ONNXmodel_.fullPath());  //Setup ONNX runtime
 
     ortinput_names.push_back(this->ONNXInputName_);
     ortoutput_names = Runtime.getOutputNames();
 
     //ONNX runtime recieves a vector of vectors of floats so push back the input
     // vector of float to create a 1,1,21 ortinput
     ortinput.push_back(Transformed_features);
 
     // batch_size 1 as only one set of transformed features is being processed
     int batch_size = 1;
     // Run classification
     ortoutputs = Runtime.run(ortinput_names, ortinput, {}, ortoutput_names, batch_size);
 
     if (this->qualityAlgorithm_ == QualityAlgorithm::NN) {
       aTrack.settrkMVA1(ortoutputs[0][0]);
     }
 
     else if (this->qualityAlgorithm_ == QualityAlgorithm::GBDT) {
       aTrack.settrkMVA1(ortoutputs[1][1]);
     }
     // Slight differences in the ONNX models of the GBDTs and NNs mean different
     // indices of the ortoutput need to be accessed
   }
 
   else {
     aTrack.settrkMVA1(-999);
   }
 }

◆ setONNXModel()

void L1TrackQuality::setONNXModel	(	std::string const &	AlgorithmString,
		edm::FileInPath const &	ONNXmodel,
		std::string const &	ONNXInputName,
		std::vector< std::string > const &	featureNames
	)

Definition at line 291 of file L1TrackQuality.cc.

References TrackQualityParams_cfi::featureNames, featureNames_, GBDT, NN, None, TrackQualityParams_cfi::ONNXInputName, ONNXInputName_, TrackQualityParams_cfi::ONNXmodel, ONNXmodel_, and qualityAlgorithm_.

Referenced by L1TrackQuality().

                                                                             {
   //Convert algorithm string to Enum class for track by track comparison
   if (AlgorithmString == "NN") {
     qualityAlgorithm_ = QualityAlgorithm::NN;
   } else if (AlgorithmString == "GBDT") {
     qualityAlgorithm_ = QualityAlgorithm::GBDT;
   } else {
     qualityAlgorithm_ = QualityAlgorithm::None;
   }
   ONNXmodel_ = ONNXmodel;
   ONNXInputName_ = ONNXInputName;
   featureNames_ = featureNames;
 }