#include <PatternRecognitionbyFastJet.h>

Inheritance diagram for ticl::PatternRecognitionbyFastJet< TILES >:

Public Member Functions
void	energyRegressionAndID (const std::vector< reco::CaloCluster > &layerClusters, const tensorflow::Session *, std::vector< Trackster > &result)

void	makeTracksters (const typename PatternRecognitionAlgoBaseT< TILES >::Inputs &input, std::vector< Trackster > &result, std::unordered_map< int, std::vector< int >> &seedToTracksterAssociation) override

	PatternRecognitionbyFastJet (const edm::ParameterSet &conf, edm::ConsumesCollector)

	~PatternRecognitionbyFastJet () override=default

Public Member Functions inherited from ticl::PatternRecognitionAlgoBaseT< TILES >
virtual void	makeTracksters (const Inputs &input, std::vector< Trackster > &result, std::unordered_map< int, std::vector< int >> &seedToTracksterAssociation)=0

	PatternRecognitionAlgoBaseT (const edm::ParameterSet &conf, edm::ConsumesCollector)

virtual	~PatternRecognitionAlgoBaseT ()

Static Public Member Functions
static void	fillPSetDescription (edm::ParameterSetDescription &iDesc)

Private Member Functions
void	buildJetAndTracksters (std::vector< fastjet::PseudoJet > &, std::vector< ticl::Trackster > &)

Private Attributes
const double	antikt_radius_

edm::ESGetToken< CaloGeometry, CaloGeometryRecord >	caloGeomToken_

const std::string	eidInputName_

const float	eidMinClusterEnergy_

const int	eidNClusters_

const int	eidNLayers_

const std::string	eidOutputNameEnergy_

const std::string	eidOutputNameId_

tensorflow::Session *	eidSession_

const int	minNumLayerCluster_

hgcal::RecHitTools	rhtools_

Static Private Attributes
static const int	eidNFeatures_ = 3

Additional Inherited Members
Public Types inherited from ticl::PatternRecognitionAlgoBaseT< TILES >
enum	VerbosityLevel { None = 0, Basic, Advanced, Expert, Guru }

Protected Attributes inherited from ticl::PatternRecognitionAlgoBaseT< TILES >
int	algo_verbosity_

Detailed Description

template<typename TILES>
class ticl::PatternRecognitionbyFastJet< TILES >

Definition at line 18 of file PatternRecognitionbyFastJet.h.

Constructor & Destructor Documentation

template<typename TILES >

PatternRecognitionbyFastJet::PatternRecognitionbyFastJet	(	const edm::ParameterSet &	conf,
		edm::ConsumesCollector	iC
	)

Definition at line 29 of file PatternRecognitionbyFastJet.cc.

     : PatternRecognitionAlgoBaseT<TILES>(conf, iC),
       caloGeomToken_(iC.esConsumes<CaloGeometry, CaloGeometryRecord>()),
       antikt_radius_(conf.getParameter<double>("antikt_radius")),
       minNumLayerCluster_(conf.getParameter<int>("minNumLayerCluster")),
       eidInputName_(conf.getParameter<std::string>("eid_input_name")),
       eidOutputNameEnergy_(conf.getParameter<std::string>("eid_output_name_energy")),
       eidOutputNameId_(conf.getParameter<std::string>("eid_output_name_id")),
       eidMinClusterEnergy_(conf.getParameter<double>("eid_min_cluster_energy")),
       eidNLayers_(conf.getParameter<int>("eid_n_layers")),
       eidNClusters_(conf.getParameter<int>("eid_n_clusters")){};

template<typename TILES >

ticl::PatternRecognitionbyFastJet< TILES >::~PatternRecognitionbyFastJet ( )

overridedefault

Member Function Documentation

template<typename TILES >

void PatternRecognitionbyFastJet::buildJetAndTracksters	(	std::vector< fastjet::PseudoJet > &	,
		std::vector< ticl::Trackster > &
	)

private

Definition at line 43 of file PatternRecognitionbyFastJet.cc.

References fwrapper::jets.

                                                                                                  {
   if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Basic) {
     edm::LogVerbatim("PatternRecogntionbyFastJet")
         << "Creating FastJet with " << fjInputs.size() << " LayerClusters in input";
   }
   fastjet::ClusterSequence sequence(fjInputs, JetDefinition(antikt_algorithm, antikt_radius_));
   auto jets = fastjet::sorted_by_pt(sequence.inclusive_jets(0));
   if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Basic) {
     edm::LogVerbatim("PatternRecogntionbyFastJet") << "FastJet produced " << jets.size() << " jets/trackster";
   }
 
   auto trackster_idx = result.size();
   result.resize(trackster_idx + jets.size());
   for (const auto &pj : jets) {
     if (pj.constituents().size() > static_cast<unsigned int>(minNumLayerCluster_)) {
       for (const auto &component : pj.constituents()) {
         result[trackster_idx].vertices().push_back(component.user_index());
         result[trackster_idx].vertex_multiplicity().push_back(1);
         if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Basic) {
           edm::LogVerbatim("PatternRecogntionbyFastJet")
               << "Jet has " << pj.constituents().size() << " components that are stored in trackster " << trackster_idx;
         }
       }
       trackster_idx++;
     } else {
       if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Advanced) {
         edm::LogVerbatim("PatternRecogntionbyFastJet")
             << "Jet with " << pj.constituents().size() << " constituents discarded since too small wrt "
             << minNumLayerCluster_;
       }
     }
   }
   fjInputs.clear();
 }

template<typename TILES >

void PatternRecognitionbyFastJet::energyRegressionAndID	(	const std::vector< reco::CaloCluster > &	layerClusters,
		const tensorflow::Session *	eidSession,
		std::vector< Trackster > &	result
	)

Definition at line 158 of file PatternRecognitionbyFastJet.cc.

References a, funct::abs(), b, data, relval_parameters_module::energy, reco::CaloCluster::energy(), reco::CaloCluster::eta(), validate-o2o-wbm::f, lowptgsfeleseed::features(), reco::CaloCluster::hitsAndFractions(), mps_fire::i, input, dqmiolumiharvest::j, isotrackApplyRegressor::k, cmsLHEtoEOSManager::l, reco::CaloCluster::phi(), tensorflow::run(), and beam_dqm_sourceclient-live_cfg::vertices.

                                                                                                  {
   // Energy regression and particle identification strategy:
   //
   // 1. Set default values for regressed energy and particle id for each trackster.
   // 2. Store indices of tracksters whose total sum of cluster energies is above the
   //    eidMinClusterEnergy_ (GeV) treshold. Inference is not applied for soft tracksters.
   // 3. When no trackster passes the selection, return.
   // 4. Create input and output tensors. The batch dimension is determined by the number of
   //    selected tracksters.
   // 5. Fill input tensors with layer cluster features. Per layer, clusters are ordered descending
   //    by energy. Given that tensor data is contiguous in memory, we can use pointer arithmetic to
   //    fill values, even with batching.
   // 6. Zero-fill features for empty clusters in each layer.
   // 7. Batched inference.
   // 8. Assign the regressed energy and id probabilities to each trackster.
   //
   // Indices used throughout this method:
   // i -> batch element / trackster
   // j -> layer
   // k -> cluster
   // l -> feature
 
   // set default values per trackster, determine if the cluster energy threshold is passed,
   // and store indices of hard tracksters
   std::vector<int> tracksterIndices;
   for (int i = 0; i < static_cast<int>(tracksters.size()); i++) {
     // calculate the cluster energy sum (2)
     // note: after the loop, sumClusterEnergy might be just above the threshold which is enough to
     // decide whether to run inference for the trackster or not
     float sumClusterEnergy = 0.;
     for (const unsigned int &vertex : tracksters[i].vertices()) {
       sumClusterEnergy += static_cast<float>(layerClusters[vertex].energy());
       // there might be many clusters, so try to stop early
       if (sumClusterEnergy >= eidMinClusterEnergy_) {
         // set default values (1)
         tracksters[i].setRegressedEnergy(0.f);
         tracksters[i].zeroProbabilities();
         tracksterIndices.push_back(i);
         break;
       }
     }
   }
 
   // do nothing when no trackster passes the selection (3)
   int batchSize = static_cast<int>(tracksterIndices.size());
   if (batchSize == 0) {
     return;
   }
 
   // create input and output tensors (4)
   tensorflow::TensorShape shape({batchSize, eidNLayers_, eidNClusters_, eidNFeatures_});
   tensorflow::Tensor input(tensorflow::DT_FLOAT, shape);
   tensorflow::NamedTensorList inputList = {{eidInputName_, input}};
 
   std::vector<tensorflow::Tensor> outputs;
   std::vector<std::string> outputNames;
   if (!eidOutputNameEnergy_.empty()) {
     outputNames.push_back(eidOutputNameEnergy_);
   }
   if (!eidOutputNameId_.empty()) {
     outputNames.push_back(eidOutputNameId_);
   }
 
   // fill input tensor (5)
   for (int i = 0; i < batchSize; i++) {
     const Trackster &trackster = tracksters[tracksterIndices[i]];
 
     // per layer, we only consider the first eidNClusters_ clusters in terms of energy, so in order
     // to avoid creating large / nested structures to do the sorting for an unknown number of total
     // clusters, create a sorted list of layer cluster indices to keep track of the filled clusters
     std::vector<int> clusterIndices(trackster.vertices().size());
     for (int k = 0; k < (int)trackster.vertices().size(); k++) {
       clusterIndices[k] = k;
     }
     sort(clusterIndices.begin(), clusterIndices.end(), [&layerClusters, &trackster](const int &a, const int &b) {
       return layerClusters[trackster.vertices(a)].energy() > layerClusters[trackster.vertices(b)].energy();
     });
 
     // keep track of the number of seen clusters per layer
     std::vector<int> seenClusters(eidNLayers_);
 
     // loop through clusters by descending energy
     for (const int &k : clusterIndices) {
       // get features per layer and cluster and store the values directly in the input tensor
       const reco::CaloCluster &cluster = layerClusters[trackster.vertices(k)];
       int j = rhtools_.getLayerWithOffset(cluster.hitsAndFractions()[0].first) - 1;
       if (j < eidNLayers_ && seenClusters[j] < eidNClusters_) {
         // get the pointer to the first feature value for the current batch, layer and cluster
         float *features = &input.tensor<float, 4>()(i, j, seenClusters[j], 0);
 
         // fill features
         *(features++) = float(cluster.energy() / float(trackster.vertex_multiplicity(k)));
         *(features++) = float(std::abs(cluster.eta()));
         *(features) = float(cluster.phi());
 
         // increment seen clusters
         seenClusters[j]++;
       }
     }
 
     // zero-fill features of empty clusters in each layer (6)
     for (int j = 0; j < eidNLayers_; j++) {
       for (int k = seenClusters[j]; k < eidNClusters_; k++) {
         float *features = &input.tensor<float, 4>()(i, j, k, 0);
         for (int l = 0; l < eidNFeatures_; l++) {
           *(features++) = 0.f;
         }
       }
     }
   }
 
   // run the inference (7)
   tensorflow::run(const_cast<tensorflow::Session *>(eidSession), inputList, outputNames, &outputs);
 
   // store regressed energy per trackster (8)
   if (!eidOutputNameEnergy_.empty()) {
     // get the pointer to the energy tensor, dimension is batch x 1
     float *energy = outputs[0].flat<float>().data();
 
     for (const int &i : tracksterIndices) {
       tracksters[i].setRegressedEnergy(*(energy++));
     }
   }
 
   // store id probabilities per trackster (8)
   if (!eidOutputNameId_.empty()) {
     // get the pointer to the id probability tensor, dimension is batch x id_probabilities.size()
     int probsIdx = eidOutputNameEnergy_.empty() ? 0 : 1;
     float *probs = outputs[probsIdx].flat<float>().data();
 
     for (const int &i : tracksterIndices) {
       tracksters[i].setProbabilities(probs);
       probs += tracksters[i].id_probabilities().size();
     }
   }
 }

template<typename TILES >

void PatternRecognitionbyFastJet::fillPSetDescription ( edm::ParameterSetDescription & iDesc )

static

Definition at line 298 of file PatternRecognitionbyFastJet.cc.

References edm::ParameterSetDescription::add(), edm::ParameterDescriptionNode::setComment(), and AlCaHLTBitMon_QueryRunRegistry::string.

                                                                                               {
   iDesc.add<int>("algo_verbosity", 0);
   iDesc.add<double>("antikt_radius", 0.09)->setComment("Radius to be used while running the Anti-kt clustering");
   iDesc.add<int>("minNumLayerCluster", 5)->setComment("Not Inclusive");
   iDesc.add<std::string>("eid_input_name", "input");
   iDesc.add<std::string>("eid_output_name_energy", "output/regressed_energy");
   iDesc.add<std::string>("eid_output_name_id", "output/id_probabilities");
   iDesc.add<double>("eid_min_cluster_energy", 1.);
   iDesc.add<int>("eid_n_layers", 50);
   iDesc.add<int>("eid_n_clusters", 10);
 }

template<typename TILES >

void PatternRecognitionbyFastJet::makeTracksters	(	const typename PatternRecognitionAlgoBaseT< TILES >::Inputs &	input,
		std::vector< Trackster > &	result,
		std::unordered_map< int, std::vector< int >> &	seedToTracksterAssociation
	)

override

Definition at line 80 of file PatternRecognitionbyFastJet.cc.

References ticl::assignPCAtoTracksters(), caloGeomToken_, haddnano::cl, ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::es, relativeConstraints::geom, edm::EventSetup::getData(), ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::layerClusters, ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::layerClustersTime, ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::mask, L1TMuonDQMOffline_cfi::nEtaBins, ecaldqm::binning::nPhiBins, hltrates_dqm_sourceclient-live_cfg::offset, ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::regions, submitPVValidationJobs::t, ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::tfSession, ticl::PatternRecognitionAlgoBaseT< TILES >::Inputs::tiles, and relativeConstraints::value.

                                                                        {
   // Protect from events with no seeding regions
   if (input.regions.empty())
     return;
 
   edm::EventSetup const &es = input.es;
   const CaloGeometry &geom = es.getData(caloGeomToken_);
   rhtools_.setGeometry(geom);
 
   constexpr auto isHFnose = std::is_same<TILES, TICLLayerTilesHFNose>::value;
   constexpr int nEtaBin = TILES::constants_type_t::nEtaBins;
   constexpr int nPhiBin = TILES::constants_type_t::nPhiBins;
 
   // We need to partition the two sides of the HGCAL detector
   auto lastLayerPerSide = static_cast<unsigned int>(rhtools_.lastLayer(isHFnose)) - 1;
   unsigned int maxLayer = 2 * lastLayerPerSide - 1;
   std::vector<fastjet::PseudoJet> fjInputs;
   fjInputs.clear();
   for (unsigned int currentLayer = 0; currentLayer <= maxLayer; ++currentLayer) {
     if (currentLayer == lastLayerPerSide) {
       buildJetAndTracksters(fjInputs, result);
     }
     const auto &tileOnLayer = input.tiles[currentLayer];
     for (int ieta = 0; ieta <= nEtaBin; ++ieta) {
       auto offset = ieta * nPhiBin;
       if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Advanced) {
         edm::LogVerbatim("PatternRecogntionbyFastJet") << "offset: " << offset;
       }
       for (int iphi = 0; iphi <= nPhiBin; ++iphi) {
         if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Advanced) {
           edm::LogVerbatim("PatternRecogntionbyFastJet") << "iphi: " << iphi;
           edm::LogVerbatim("PatternRecogntionbyFastJet") << "Entries in tileBin: " << tileOnLayer[offset + iphi].size();
         }
         for (auto clusterIdx : tileOnLayer[offset + iphi]) {
           // Skip masked layer clusters
           if (input.mask[clusterIdx] == 0.) {
             if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Advanced) {
               edm::LogVerbatim("PatternRecogntionbyFastJet") << "Skipping masked layerIdx " << clusterIdx;
             }
             continue;
           }
           // Should we correct for the position of the PV?
           auto const &cl = input.layerClusters[clusterIdx];
           math::XYZVector direction(cl.x(), cl.y(), cl.z());
           direction = direction.Unit();
           direction *= cl.energy();
           auto fpj = fastjet::PseudoJet(direction.X(), direction.Y(), direction.Z(), cl.energy());
           fpj.set_user_index(clusterIdx);
           fjInputs.push_back(fpj);
         }  // End of loop on the clusters on currentLayer
       }    // End of loop over phi-bin region
     }      // End of loop over eta-bin region
   }        // End of loop over layers
 
   // Collect the jet from the other side wrt to the one taken care of inside the main loop above.
   buildJetAndTracksters(fjInputs, result);
 
   ticl::assignPCAtoTracksters(result,
                               input.layerClusters,
                               input.layerClustersTime,
                               rhtools_.getPositionLayer(rhtools_.lastLayerEE(isHFnose), isHFnose).z());
 
   // run energy regression and ID
   energyRegressionAndID(input.layerClusters, input.tfSession, result);
   if (PatternRecognitionAlgoBaseT<TILES>::algo_verbosity_ > PatternRecognitionAlgoBaseT<TILES>::Basic) {
     for (auto const &t : result) {
       edm::LogVerbatim("PatternRecogntionbyFastJet") << "Barycenter: " << t.barycenter();
       edm::LogVerbatim("PatternRecogntionbyFastJet") << "LCs: " << t.vertices().size();
       edm::LogVerbatim("PatternRecogntionbyFastJet") << "Energy: " << t.raw_energy();
       edm::LogVerbatim("PatternRecogntionbyFastJet") << "Regressed: " << t.regressed_energy();
     }
   }
 }