TICL/plugins/TrackstersMergeProducer.cc

0001 #include <memory>  // unique_ptr
0002 #include "FWCore/Framework/interface/stream/EDProducer.h"
0003 #include "FWCore/ParameterSet/interface/ParameterSet.h"
0004 #include "FWCore/ParameterSet/interface/ConfigurationDescriptions.h"
0005 #include "FWCore/ParameterSet/interface/PluginDescription.h"
0006 #include "FWCore/MessageLogger/interface/MessageLogger.h"
0007 #include "FWCore/Utilities/interface/ESGetToken.h"
0008 #include "FWCore/Framework/interface/ESHandle.h"
0009 #include "FWCore/Framework/interface/Frameworkfwd.h"
0010 #include "FWCore/Framework/interface/MakerMacros.h"
0011 #include "FWCore/ParameterSet/interface/ParameterSetDescription.h"
0012 #include "FWCore/Framework/interface/ConsumesCollector.h"
0013
0014 #include "DataFormats/CaloRecHit/interface/CaloCluster.h"
0015 #include "DataFormats/HGCalReco/interface/Common.h"
0016 #include "DataFormats/HGCalReco/interface/TICLLayerTile.h"
0017 #include "DataFormats/HGCalReco/interface/Trackster.h"
0018 #include "DataFormats/TrackReco/interface/Track.h"
0019 #include "DataFormats/MuonReco/interface/Muon.h"
0020 #include "DataFormats/GeometrySurface/interface/BoundDisk.h"
0021 #include "DataFormats/HGCalReco/interface/TICLCandidate.h"
0022 #include "DataFormats/TrackReco/interface/TrackFwd.h"
0023 #include "DataFormats/Math/interface/Vector3D.h"
0024
0025 #include "RecoHGCal/TICL/interface/GlobalCache.h"
0026
0027 #include "PhysicsTools/TensorFlow/interface/TfGraphRecord.h"
0028 #include "PhysicsTools/TensorFlow/interface/TensorFlow.h"
0029 #include "PhysicsTools/TensorFlow/interface/TfGraphDefWrapper.h"
0030
0031 #include "RecoHGCal/TICL/plugins/LinkingAlgoBase.h"
0032 #include "RecoHGCal/TICL/plugins/LinkingAlgoFactory.h"
0033 #include "RecoHGCal/TICL/plugins/LinkingAlgoByDirectionGeometric.h"
0034
0035 #include "RecoLocalCalo/HGCalRecAlgos/interface/RecHitTools.h"
0036 #include "PhysicsTools/TensorFlow/interface/TensorFlow.h"
0037
0038 #include "TrackingTools/TrajectoryState/interface/TrajectoryStateTransform.h"
0039 #include "TrackingTools/GeomPropagators/interface/Propagator.h"
0040 #include "TrackingTools/Records/interface/TrackingComponentsRecord.h"
0041
0042 #include "MagneticField/Engine/interface/MagneticField.h"
0043 #include "MagneticField/Records/interface/IdealMagneticFieldRecord.h"
0044
0045 #include "Geometry/HGCalCommonData/interface/HGCalDDDConstants.h"
0046 #include "Geometry/Records/interface/IdealGeometryRecord.h"
0047 #include "Geometry/CommonDetUnit/interface/GeomDet.h"
0048
0049 #include "TrackstersPCA.h"
0050
0051 using namespace ticl;
0052
0053 class TrackstersMergeProducer : public edm::stream::EDProducer<> {
0054 public:
0055   explicit TrackstersMergeProducer(const edm::ParameterSet &ps);
0056   ~TrackstersMergeProducer() override{};
0057   void produce(edm::Event &, const edm::EventSetup &) override;
0058   static void fillDescriptions(edm::ConfigurationDescriptions &descriptions);
0059
0060   // static methods for handling the global cache
0061   static std::unique_ptr<TrackstersCache> initializeGlobalCache(const edm::ParameterSet &);
0062   static void globalEndJob(TrackstersCache *);
0063
0064   void beginJob();
0065   void endJob();
0066
0067   void beginRun(edm::Run const &iEvent, edm::EventSetup const &es) override;
0068
0069 private:
0070   typedef ticl::Trackster::IterationIndex TracksterIterIndex;
0071   typedef math::XYZVector Vector;
0072
0073   void fillTile(TICLTracksterTiles &, const std::vector<Trackster> &, TracksterIterIndex);
0074
0075   void energyRegressionAndID(const std::vector<reco::CaloCluster> &layerClusters,
0076                              const tensorflow::Session *,
0077                              std::vector<Trackster> &result) const;
0078   void printTrackstersDebug(const std::vector<Trackster> &, const char *label) const;
0079   void assignTimeToCandidates(std::vector<TICLCandidate> &resultCandidates) const;
0080   void dumpTrackster(const Trackster &) const;
0081
0082   std::unique_ptr<LinkingAlgoBase> linkingAlgo_;
0083
0084   const edm::EDGetTokenT<std::vector<Trackster>> tracksters_clue3d_token_;
0085   const edm::EDGetTokenT<std::vector<reco::CaloCluster>> clusters_token_;
0086   const edm::EDGetTokenT<edm::ValueMap<std::pair<float, float>>> clustersTime_token_;
0087   const edm::EDGetTokenT<std::vector<reco::Track>> tracks_token_;
0088   const edm::EDGetTokenT<edm::ValueMap<float>> tracks_time_token_;
0089   const edm::EDGetTokenT<edm::ValueMap<float>> tracks_time_quality_token_;
0090   const edm::EDGetTokenT<edm::ValueMap<float>> tracks_time_err_token_;
0091   const edm::EDGetTokenT<std::vector<reco::Muon>> muons_token_;
0092   const std::string tfDnnLabel_;
0093   const edm::ESGetToken<TfGraphDefWrapper, TfGraphRecord> tfDnnToken_;
0094   const tensorflow::Session *tfSession_;
0095
0096   const edm::ESGetToken<CaloGeometry, CaloGeometryRecord> geometry_token_;
0097   const std::string detector_;
0098   const std::string propName_;
0099
0100   const edm::ESGetToken<MagneticField, IdealMagneticFieldRecord> bfield_token_;
0101   const edm::ESGetToken<Propagator, TrackingComponentsRecord> propagator_token_;
0102   const bool optimiseAcrossTracksters_;
0103   const int eta_bin_window_;
0104   const int phi_bin_window_;
0105   const double pt_sigma_high_;
0106   const double pt_sigma_low_;
0107   const double halo_max_distance2_;
0108   const double track_min_pt_;
0109   const double track_min_eta_;
0110   const double track_max_eta_;
0111   const int track_max_missing_outerhits_;
0112   const double cosangle_align_;
0113   const double e_over_h_threshold_;
0114   const double pt_neutral_threshold_;
0115   const double resol_calo_offset_had_;
0116   const double resol_calo_scale_had_;
0117   const double resol_calo_offset_em_;
0118   const double resol_calo_scale_em_;
0119   const std::string eidInputName_;
0120   const std::string eidOutputNameEnergy_;
0121   const std::string eidOutputNameId_;
0122   const float eidMinClusterEnergy_;
0123   const int eidNLayers_;
0124   const int eidNClusters_;
0125   std::once_flag initializeGeometry_;
0126
0127   const HGCalDDDConstants *hgcons_;
0128
0129   std::unique_ptr<GeomDet> firstDisk_[2];
0130
0131   tensorflow::Session *eidSession_;
0132   hgcal::RecHitTools rhtools_;
0133
0134   static constexpr int eidNFeatures_ = 3;
0135
0136   edm::ESGetToken<HGCalDDDConstants, IdealGeometryRecord> hdc_token_;
0137 };
0138
0139 TrackstersMergeProducer::TrackstersMergeProducer(const edm::ParameterSet &ps)
0140     : tracksters_clue3d_token_(consumes<std::vector<Trackster>>(ps.getParameter<edm::InputTag>("trackstersclue3d"))),
0141       clusters_token_(consumes<std::vector<reco::CaloCluster>>(ps.getParameter<edm::InputTag>("layer_clusters"))),
0142       clustersTime_token_(
0143           consumes<edm::ValueMap<std::pair<float, float>>>(ps.getParameter<edm::InputTag>("layer_clustersTime"))),
0144       tracks_token_(consumes<std::vector<reco::Track>>(ps.getParameter<edm::InputTag>("tracks"))),
0145       tracks_time_token_(consumes<edm::ValueMap<float>>(ps.getParameter<edm::InputTag>("tracksTime"))),
0146       tracks_time_quality_token_(consumes<edm::ValueMap<float>>(ps.getParameter<edm::InputTag>("tracksTimeQual"))),
0147       tracks_time_err_token_(consumes<edm::ValueMap<float>>(ps.getParameter<edm::InputTag>("tracksTimeErr"))),
0148       muons_token_(consumes<std::vector<reco::Muon>>(ps.getParameter<edm::InputTag>("muons"))),
0149       tfDnnLabel_(ps.getParameter<std::string>("tfDnnLabel")),
0150       tfDnnToken_(esConsumes(edm::ESInputTag("", tfDnnLabel_))),
0151       tfSession_(nullptr),
0152       geometry_token_(esConsumes<CaloGeometry, CaloGeometryRecord, edm::Transition::BeginRun>()),
0153       detector_(ps.getParameter<std::string>("detector")),
0154       propName_(ps.getParameter<std::string>("propagator")),
0155       bfield_token_(esConsumes<MagneticField, IdealMagneticFieldRecord, edm::Transition::BeginRun>()),
0156       propagator_token_(
0157           esConsumes<Propagator, TrackingComponentsRecord, edm::Transition::BeginRun>(edm::ESInputTag("", propName_))),
0158       optimiseAcrossTracksters_(ps.getParameter<bool>("optimiseAcrossTracksters")),
0159       eta_bin_window_(ps.getParameter<int>("eta_bin_window")),
0160       phi_bin_window_(ps.getParameter<int>("phi_bin_window")),
0161       pt_sigma_high_(ps.getParameter<double>("pt_sigma_high")),
0162       pt_sigma_low_(ps.getParameter<double>("pt_sigma_low")),
0163       halo_max_distance2_(ps.getParameter<double>("halo_max_distance2")),
0164       track_min_pt_(ps.getParameter<double>("track_min_pt")),
0165       track_min_eta_(ps.getParameter<double>("track_min_eta")),
0166       track_max_eta_(ps.getParameter<double>("track_max_eta")),
0167       track_max_missing_outerhits_(ps.getParameter<int>("track_max_missing_outerhits")),
0168       cosangle_align_(ps.getParameter<double>("cosangle_align")),
0169       e_over_h_threshold_(ps.getParameter<double>("e_over_h_threshold")),
0170       pt_neutral_threshold_(ps.getParameter<double>("pt_neutral_threshold")),
0171       resol_calo_offset_had_(ps.getParameter<double>("resol_calo_offset_had")),
0172       resol_calo_scale_had_(ps.getParameter<double>("resol_calo_scale_had")),
0173       resol_calo_offset_em_(ps.getParameter<double>("resol_calo_offset_em")),
0174       resol_calo_scale_em_(ps.getParameter<double>("resol_calo_scale_em")),
0175       eidInputName_(ps.getParameter<std::string>("eid_input_name")),
0176       eidOutputNameEnergy_(ps.getParameter<std::string>("eid_output_name_energy")),
0177       eidOutputNameId_(ps.getParameter<std::string>("eid_output_name_id")),
0178       eidMinClusterEnergy_(ps.getParameter<double>("eid_min_cluster_energy")),
0179       eidNLayers_(ps.getParameter<int>("eid_n_layers")),
0180       eidNClusters_(ps.getParameter<int>("eid_n_clusters")),
0181       eidSession_(nullptr) {
0182   produces<std::vector<Trackster>>();
0183   produces<std::vector<TICLCandidate>>();
0184
0185   std::string detectorName_ = (detector_ == "HFNose") ? "HGCalHFNoseSensitive" : "HGCalEESensitive";
0186   hdc_token_ =
0187       esConsumes<HGCalDDDConstants, IdealGeometryRecord, edm::Transition::BeginRun>(edm::ESInputTag("", detectorName_));
0188
0189   auto linkingPSet = ps.getParameter<edm::ParameterSet>("linkingPSet");
0190   auto algoType = linkingPSet.getParameter<std::string>("type");
0191   linkingAlgo_ = LinkingAlgoFactory::get()->create(algoType, linkingPSet);
0192 }
0193
0194 void TrackstersMergeProducer::beginJob() {}
0195
0196 void TrackstersMergeProducer::endJob(){};
0197
0198 void TrackstersMergeProducer::beginRun(edm::Run const &iEvent, edm::EventSetup const &es) {
0199   edm::ESHandle<HGCalDDDConstants> hdc = es.getHandle(hdc_token_);
0200   hgcons_ = hdc.product();
0201
0202   edm::ESHandle<CaloGeometry> geom = es.getHandle(geometry_token_);
0203   rhtools_.setGeometry(*geom);
0204
0205   edm::ESHandle<MagneticField> bfield = es.getHandle(bfield_token_);
0206   edm::ESHandle<Propagator> propagator = es.getHandle(propagator_token_);
0207
0208   linkingAlgo_->initialize(hgcons_, rhtools_, bfield, propagator);
0209 };
0210
0211 void TrackstersMergeProducer::fillTile(TICLTracksterTiles &tracksterTile,
0212                                        const std::vector<Trackster> &tracksters,
0213                                        TracksterIterIndex tracksterIteration) {
0214   int tracksterId = 0;
0215   for (auto const &t : tracksters) {
0216     tracksterTile.fill(tracksterIteration, t.barycenter().eta(), t.barycenter().phi(), tracksterId);
0217     LogDebug("TrackstersMergeProducer") << "Adding tracksterId: " << tracksterId << " into bin [eta,phi]: [ "
0218                                         << tracksterTile[tracksterIteration].etaBin(t.barycenter().eta()) << ", "
0219                                         << tracksterTile[tracksterIteration].phiBin(t.barycenter().phi())
0220                                         << "] for iteration: " << tracksterIteration << std::endl;
0221
0222     tracksterId++;
0223   }
0224 }
0225
0226 void TrackstersMergeProducer::dumpTrackster(const Trackster &t) const {
0227   auto e_over_h = (t.raw_em_pt() / ((t.raw_pt() - t.raw_em_pt()) != 0. ? (t.raw_pt() - t.raw_em_pt()) : 1.));
0228   LogDebug("TrackstersMergeProducer")
0229       << "\nTrackster raw_pt: " << t.raw_pt() << " raw_em_pt: " << t.raw_em_pt() << " eoh: " << e_over_h
0230       << " barycenter: " << t.barycenter() << " eta,phi (baricenter): " << t.barycenter().eta() << ", "
0231       << t.barycenter().phi() << " eta,phi (eigen): " << t.eigenvectors(0).eta() << ", " << t.eigenvectors(0).phi()
0232       << " pt(eigen): " << std::sqrt(t.eigenvectors(0).Unit().perp2()) * t.raw_energy() << " seedID: " << t.seedID()
0233       << " seedIndex: " << t.seedIndex() << " size: " << t.vertices().size() << " average usage: "
0234       << (std::accumulate(std::begin(t.vertex_multiplicity()), std::end(t.vertex_multiplicity()), 0.) /
0235           (float)t.vertex_multiplicity().size())
0236       << " raw_energy: " << t.raw_energy() << " regressed energy: " << t.regressed_energy()
0237       << " probs(ga/e/mu/np/cp/nh/am/unk): ";
0238   for (auto const &p : t.id_probabilities()) {
0239     LogDebug("TrackstersMergeProducer") << std::fixed << p << " ";
0240   }
0241   LogDebug("TrackstersMergeProducer") << " sigmas: ";
0242   for (auto const &s : t.sigmas()) {
0243     LogDebug("TrackstersMergeProducer") << s << " ";
0244   }
0245   LogDebug("TrackstersMergeProducer") << std::endl;
0246 }
0247
0248 void TrackstersMergeProducer::produce(edm::Event &evt, const edm::EventSetup &es) {
0249   auto resultTrackstersMerged = std::make_unique<std::vector<Trackster>>();
0250   auto resultCandidates = std::make_unique<std::vector<TICLCandidate>>();
0251   auto resultFromTracks = std::make_unique<std::vector<TICLCandidate>>();
0252   tfSession_ = es.getData(tfDnnToken_).getSession();
0253
0254   edm::Handle<std::vector<Trackster>> trackstersclue3d_h;
0255   evt.getByToken(tracksters_clue3d_token_, trackstersclue3d_h);
0256
0257   edm::Handle<std::vector<reco::Track>> track_h;
0258   evt.getByToken(tracks_token_, track_h);
0259   const auto &tracks = *track_h;
0260
0261   const auto &layerClusters = evt.get(clusters_token_);
0262   const auto &layerClustersTimes = evt.get(clustersTime_token_);
0263   const auto &muons = evt.get(muons_token_);
0264   const auto &trackTime = evt.get(tracks_time_token_);
0265   const auto &trackTimeErr = evt.get(tracks_time_err_token_);
0266   const auto &trackTimeQual = evt.get(tracks_time_quality_token_);
0267
0268   // Linking
0269   linkingAlgo_->linkTracksters(
0270       track_h, trackTime, trackTimeErr, trackTimeQual, muons, trackstersclue3d_h, *resultCandidates, *resultFromTracks);
0271
0272   // Print debug info
0273   LogDebug("TrackstersMergeProducer") << "Results from the linking step : " << std::endl
0274                                       << "No. of Tracks : " << tracks.size()
0275                                       << "  No. of Tracksters : " << (*trackstersclue3d_h).size() << std::endl
0276                                       << "(neutral candidates have track id -1)" << std::endl;
0277
0278   std::vector<TICLCandidate> &candidates = *resultCandidates;
0279   for (const auto &cand : candidates) {
0280     auto track_ptr = cand.trackPtr();
0281     auto trackster_ptrs = cand.tracksters();
0282
0283 #ifdef EDM_ML_DEBUG
0284     auto track_idx = track_ptr.get() - (edm::Ptr<reco::Track>(track_h, 0)).get();
0285     track_idx = (track_ptr.isNull()) ? -1 : track_idx;
0286     LogDebug("TrackstersMergeProducer") << "PDG ID " << cand.pdgId() << " charge " << cand.charge() << " p " << cand.p()
0287                                         << std::endl;
0288     LogDebug("TrackstersMergeProducer") << "track id (p) : " << track_idx << " ("
0289                                         << (track_ptr.isNull() ? -1 : track_ptr->p()) << ") "
0290                                         << " trackster ids (E) : ";
0291 #endif
0292
0293     // Merge included tracksters
0294     ticl::Trackster outTrackster;
0295     auto updated_size = 0;
0296     for (const auto &ts_ptr : trackster_ptrs) {
0297 #ifdef EDM_ML_DEBUG
0298       auto ts_idx = ts_ptr.get() - (edm::Ptr<ticl::Trackster>(trackstersclue3d_h, 0)).get();
0299       LogDebug("TrackstersMergeProducer") << ts_idx << " (" << ts_ptr->raw_energy() << ") ";
0300 #endif
0301
0302       auto &thisTrackster = *ts_ptr;
0303       updated_size += thisTrackster.vertices().size();
0304       outTrackster.vertices().reserve(updated_size);
0305       outTrackster.vertex_multiplicity().reserve(updated_size);
0306       std::copy(std::begin(thisTrackster.vertices()),
0307                 std::end(thisTrackster.vertices()),
0308                 std::back_inserter(outTrackster.vertices()));
0309       std::copy(std::begin(thisTrackster.vertex_multiplicity()),
0310                 std::end(thisTrackster.vertex_multiplicity()),
0311                 std::back_inserter(outTrackster.vertex_multiplicity()));
0312     }
0313
0314     LogDebug("TrackstersMergeProducer") << std::endl;
0315
0316     // Find duplicate LCs
0317     auto &orig_vtx = outTrackster.vertices();
0318     auto vtx_sorted{orig_vtx};
0319     std::sort(std::begin(vtx_sorted), std::end(vtx_sorted));
0320     for (unsigned int iLC = 1; iLC < vtx_sorted.size(); ++iLC) {
0321       if (vtx_sorted[iLC] == vtx_sorted[iLC - 1]) {
0322         // Clean up duplicate LCs
0323         const auto lcIdx = vtx_sorted[iLC];
0324         const auto firstEl = std::find(orig_vtx.begin(), orig_vtx.end(), lcIdx);
0325         const auto firstPos = std::distance(std::begin(orig_vtx), firstEl);
0326         auto iDup = std::find(std::next(firstEl), orig_vtx.end(), lcIdx);
0327         while (iDup != orig_vtx.end()) {
0328           orig_vtx.erase(iDup);
0329           outTrackster.vertex_multiplicity().erase(outTrackster.vertex_multiplicity().begin() +
0330                                                    std::distance(std::begin(orig_vtx), iDup));
0331           outTrackster.vertex_multiplicity()[firstPos] -= 1;
0332           iDup = std::find(std::next(firstEl), orig_vtx.end(), lcIdx);
0333         };
0334       }
0335     }
0336
0337     outTrackster.zeroProbabilities();
0338     if (!track_ptr.isNull())
0339       outTrackster.setSeed(track_h.id(), track_ptr.get() - (edm::Ptr<reco::Track>(track_h, 0)).get());
0340     if (!outTrackster.vertices().empty()) {
0341       resultTrackstersMerged->push_back(outTrackster);
0342     }
0343   }
0344
0345   assignPCAtoTracksters(*resultTrackstersMerged,
0346                         layerClusters,
0347                         layerClustersTimes,
0348                         rhtools_.getPositionLayer(rhtools_.lastLayerEE()).z());
0349   energyRegressionAndID(layerClusters, tfSession_, *resultTrackstersMerged);
0350
0351   //filling the TICLCandidates information
0352   assert(resultTrackstersMerged->size() == resultCandidates->size());
0353
0354   auto isHad = [](const Trackster &tracksterMerge) {
0355     return tracksterMerge.id_probability(Trackster::ParticleType::photon) +
0356                tracksterMerge.id_probability(Trackster::ParticleType::electron) <
0357            0.5;
0358   };
0359   for (size_t i = 0; i < resultTrackstersMerged->size(); i++) {
0360     auto const &tm = (*resultTrackstersMerged)[i];
0361     auto &cand = (*resultCandidates)[i];
0362     //common properties
0363     cand.setIdProbabilities(tm.id_probabilities());
0364     //charged candidates
0365     if (!cand.trackPtr().isNull()) {
0366       auto pdgId = isHad(tm) ? 211 : 11;
0367       auto const &tk = cand.trackPtr().get();
0368       cand.setPdgId(pdgId * tk->charge());
0369       cand.setCharge(tk->charge());
0370       cand.setRawEnergy(tm.raw_energy());
0371       auto const &regrE = tm.regressed_energy();
0372       math::XYZTLorentzVector p4(regrE * tk->momentum().unit().x(),
0373                                  regrE * tk->momentum().unit().y(),
0374                                  regrE * tk->momentum().unit().z(),
0375                                  regrE);
0376       cand.setP4(p4);
0377     } else {  // neutral candidates
0378       auto pdgId = isHad(tm) ? 130 : 22;
0379       cand.setPdgId(pdgId);
0380       cand.setCharge(0);
0381       cand.setRawEnergy(tm.raw_energy());
0382       const float &regrE = tm.regressed_energy();
0383       math::XYZTLorentzVector p4(regrE * tm.barycenter().unit().x(),
0384                                  regrE * tm.barycenter().unit().y(),
0385                                  regrE * tm.barycenter().unit().z(),
0386                                  regrE);
0387       cand.setP4(p4);
0388     }
0389   }
0390   for (auto &cand : *resultFromTracks) {  //Tracks with no linked tracksters are promoted to charged hadron candidates
0391     auto const &tk = cand.trackPtr().get();
0392     cand.setPdgId(211 * tk->charge());
0393     cand.setCharge(tk->charge());
0394     const float energy = std::sqrt(tk->p() * tk->p() + ticl::mpion2);
0395     cand.setRawEnergy(energy);
0396     math::PtEtaPhiMLorentzVector p4Polar(tk->pt(), tk->eta(), tk->phi(), ticl::mpion);
0397     cand.setP4(p4Polar);
0398   }
0399   // Compute timing
0400   resultCandidates->insert(resultCandidates->end(), resultFromTracks->begin(), resultFromTracks->end());
0401   assignTimeToCandidates(*resultCandidates);
0402
0403   evt.put(std::move(resultTrackstersMerged));
0404   evt.put(std::move(resultCandidates));
0405 }
0406
0407 void TrackstersMergeProducer::energyRegressionAndID(const std::vector<reco::CaloCluster> &layerClusters,
0408                                                     const tensorflow::Session *eidSession,
0409                                                     std::vector<Trackster> &tracksters) const {
0410   // Energy regression and particle identification strategy:
0411   //
0412   // 1. Set default values for regressed energy and particle id for each trackster.
0413   // 2. Store indices of tracksters whose total sum of cluster energies is above the
0414   //    eidMinClusterEnergy_ (GeV) threshold. Inference is not applied for soft tracksters.
0415   // 3. When no trackster passes the selection, return.
0416   // 4. Create input and output tensors. The batch dimension is determined by the number of
0417   //    selected tracksters.
0418   // 5. Fill input tensors with layer cluster features. Per layer, clusters are ordered descending
0419   //    by energy. Given that tensor data is contiguous in memory, we can use pointer arithmetic to
0420   //    fill values, even with batching.
0421   // 6. Zero-fill features for empty clusters in each layer.
0422   // 7. Batched inference.
0423   // 8. Assign the regressed energy and id probabilities to each trackster.
0424   //
0425   // Indices used throughout this method:
0426   // i -> batch element / trackster
0427   // j -> layer
0428   // k -> cluster
0429   // l -> feature
0430
0431   // do nothing when no trackster passes the selection (3)
0432   int batchSize = (int)tracksters.size();
0433   if (batchSize == 0) {
0434     return;
0435   }
0436
0437   for (auto &t : tracksters) {
0438     t.setRegressedEnergy(0.f);
0439     t.zeroProbabilities();
0440   }
0441
0442   // create input and output tensors (4)
0443   tensorflow::TensorShape shape({batchSize, eidNLayers_, eidNClusters_, eidNFeatures_});
0444   tensorflow::Tensor input(tensorflow::DT_FLOAT, shape);
0445   tensorflow::NamedTensorList inputList = {{eidInputName_, input}};
0446   static constexpr int inputDimension = 4;
0447
0448   std::vector<tensorflow::Tensor> outputs;
0449   std::vector<std::string> outputNames;
0450   if (!eidOutputNameEnergy_.empty()) {
0451     outputNames.push_back(eidOutputNameEnergy_);
0452   }
0453   if (!eidOutputNameId_.empty()) {
0454     outputNames.push_back(eidOutputNameId_);
0455   }
0456
0457   // fill input tensor (5)
0458   for (int i = 0; i < batchSize; i++) {
0459     const Trackster &trackster = tracksters[i];
0460
0461     // per layer, we only consider the first eidNClusters_ clusters in terms of
0462     // energy, so in order to avoid creating large / nested structures to do
0463     // the sorting for an unknown number of total clusters, create a sorted
0464     // list of layer cluster indices to keep track of the filled clusters
0465     std::vector<int> clusterIndices(trackster.vertices().size());
0466     for (int k = 0; k < (int)trackster.vertices().size(); k++) {
0467       clusterIndices[k] = k;
0468     }
0469     sort(clusterIndices.begin(), clusterIndices.end(), [&layerClusters, &trackster](const int &a, const int &b) {
0470       return layerClusters[trackster.vertices(a)].energy() > layerClusters[trackster.vertices(b)].energy();
0471     });
0472
0473     // keep track of the number of seen clusters per layer
0474     std::vector<int> seenClusters(eidNLayers_);
0475
0476     // loop through clusters by descending energy
0477     for (const int &k : clusterIndices) {
0478       // get features per layer and cluster and store the values directly in the input tensor
0479       const reco::CaloCluster &cluster = layerClusters[trackster.vertices(k)];
0480       int j = rhtools_.getLayerWithOffset(cluster.hitsAndFractions()[0].first) - 1;
0481       if (j < eidNLayers_ && seenClusters[j] < eidNClusters_) {
0482         // get the pointer to the first feature value for the current batch, layer and cluster
0483         float *features = &input.tensor<float, inputDimension>()(i, j, seenClusters[j], 0);
0484
0485         // fill features
0486         *(features++) = float(cluster.energy() / float(trackster.vertex_multiplicity(k)));
0487         *(features++) = float(std::abs(cluster.eta()));
0488         *(features) = float(cluster.phi());
0489
0490         // increment seen clusters
0491         seenClusters[j]++;
0492       }
0493     }
0494
0495     // zero-fill features of empty clusters in each layer (6)
0496     for (int j = 0; j < eidNLayers_; j++) {
0497       for (int k = seenClusters[j]; k < eidNClusters_; k++) {
0498         float *features = &input.tensor<float, inputDimension>()(i, j, k, 0);
0499         for (int l = 0; l < eidNFeatures_; l++) {
0500           *(features++) = 0.f;
0501         }
0502       }
0503     }
0504   }
0505
0506   // run the inference (7)
0507   tensorflow::run(eidSession, inputList, outputNames, &outputs);
0508
0509   // store regressed energy per trackster (8)
0510   if (!eidOutputNameEnergy_.empty()) {
0511     // get the pointer to the energy tensor, dimension is batch x 1
0512     float *energy = outputs[0].flat<float>().data();
0513
0514     for (int i = 0; i < batchSize; ++i) {
0515       float regressedEnergy =
0516           tracksters[i].raw_energy() > eidMinClusterEnergy_ ? energy[i] : tracksters[i].raw_energy();
0517       tracksters[i].setRegressedEnergy(regressedEnergy);
0518     }
0519   }
0520
0521   // store id probabilities per trackster (8)
0522   if (!eidOutputNameId_.empty()) {
0523     // get the pointer to the id probability tensor, dimension is batch x id_probabilities.size()
0524     int probsIdx = !eidOutputNameEnergy_.empty();
0525     float *probs = outputs[probsIdx].flat<float>().data();
0526     int probsNumber = tracksters[0].id_probabilities().size();
0527     for (int i = 0; i < batchSize; ++i) {
0528       tracksters[i].setProbabilities(&probs[i * probsNumber]);
0529     }
0530   }
0531 }
0532
0533 void TrackstersMergeProducer::assignTimeToCandidates(std::vector<TICLCandidate> &resultCandidates) const {
0534   for (auto &cand : resultCandidates) {
0535     if (cand.tracksters().size() > 1) {  // For single-trackster candidates the timing is already set
0536       float time = 0.f;
0537       float timeErr = 0.f;
0538       for (const auto &tr : cand.tracksters()) {
0539         if (tr->timeError() > 0) {
0540           auto invTimeESq = pow(tr->timeError(), -2);
0541           time += tr->time() * invTimeESq;
0542           timeErr += invTimeESq;
0543         }
0544       }
0545       if (timeErr > 0) {
0546         timeErr = 1. / timeErr;
0547
0548         cand.setTime(time * timeErr);
0549         cand.setTimeError(sqrt(timeErr));
0550       }
0551     }
0552   }
0553 }
0554
0555 void TrackstersMergeProducer::printTrackstersDebug(const std::vector<Trackster> &tracksters, const char *label) const {
0556 #ifdef EDM_ML_DEBUG
0557   int counter = 0;
0558   for (auto const &t : tracksters) {
0559     LogDebug("TrackstersMergeProducer")
0560         << counter++ << " TrackstersMergeProducer (" << label << ") obj barycenter: " << t.barycenter()
0561         << " eta,phi (baricenter): " << t.barycenter().eta() << ", " << t.barycenter().phi()
0562         << " eta,phi (eigen): " << t.eigenvectors(0).eta() << ", " << t.eigenvectors(0).phi()
0563         << " pt(eigen): " << std::sqrt(t.eigenvectors(0).Unit().perp2()) * t.raw_energy() << " seedID: " << t.seedID()
0564         << " seedIndex: " << t.seedIndex() << " size: " << t.vertices().size() << " average usage: "
0565         << (std::accumulate(std::begin(t.vertex_multiplicity()), std::end(t.vertex_multiplicity()), 0.) /
0566             (float)t.vertex_multiplicity().size())
0567         << " raw_energy: " << t.raw_energy() << " regressed energy: " << t.regressed_energy()
0568         << " probs(ga/e/mu/np/cp/nh/am/unk): ";
0569     for (auto const &p : t.id_probabilities()) {
0570       LogDebug("TrackstersMergeProducer") << std::fixed << p << " ";
0571     }
0572     LogDebug("TrackstersMergeProducer") << " sigmas: ";
0573     for (auto const &s : t.sigmas()) {
0574       LogDebug("TrackstersMergeProducer") << s << " ";
0575     }
0576     LogDebug("TrackstersMergeProducer") << std::endl;
0577   }
0578 #endif
0579 }
0580
0581 void TrackstersMergeProducer::fillDescriptions(edm::ConfigurationDescriptions &descriptions) {
0582   edm::ParameterSetDescription desc;
0583
0584   edm::ParameterSetDescription linkingDesc;
0585   linkingDesc.addNode(edm::PluginDescription<LinkingAlgoFactory>("type", "LinkingAlgoByDirectionGeometric", true));
0586   desc.add<edm::ParameterSetDescription>("linkingPSet", linkingDesc);
0587
0588   desc.add<edm::InputTag>("trackstersclue3d", edm::InputTag("ticlTrackstersCLUE3DHigh"));
0589   desc.add<edm::InputTag>("layer_clusters", edm::InputTag("hgcalMergeLayerClusters"));
0590   desc.add<edm::InputTag>("layer_clustersTime", edm::InputTag("hgcalMergeLayerClusters", "timeLayerCluster"));
0591   desc.add<edm::InputTag>("tracks", edm::InputTag("generalTracks"));
0592   desc.add<edm::InputTag>("tracksTime", edm::InputTag("tofPID:t0"));
0593   desc.add<edm::InputTag>("tracksTimeQual", edm::InputTag("mtdTrackQualityMVA:mtdQualMVA"));
0594   desc.add<edm::InputTag>("tracksTimeErr", edm::InputTag("tofPID:sigmat0"));
0595   desc.add<edm::InputTag>("muons", edm::InputTag("muons1stStep"));
0596   desc.add<std::string>("detector", "HGCAL");
0597   desc.add<std::string>("propagator", "PropagatorWithMaterial");
0598   desc.add<bool>("optimiseAcrossTracksters", true);
0599   desc.add<int>("eta_bin_window", 1);
0600   desc.add<int>("phi_bin_window", 1);
0601   desc.add<double>("pt_sigma_high", 2.);
0602   desc.add<double>("pt_sigma_low", 2.);
0603   desc.add<double>("halo_max_distance2", 4.);
0604   desc.add<double>("track_min_pt", 1.);
0605   desc.add<double>("track_min_eta", 1.48);
0606   desc.add<double>("track_max_eta", 3.);
0607   desc.add<int>("track_max_missing_outerhits", 5);
0608   desc.add<double>("cosangle_align", 0.9945);
0609   desc.add<double>("e_over_h_threshold", 1.);
0610   desc.add<double>("pt_neutral_threshold", 2.);
0611   desc.add<double>("resol_calo_offset_had", 1.5);
0612   desc.add<double>("resol_calo_scale_had", 0.15);
0613   desc.add<double>("resol_calo_offset_em", 1.5);
0614   desc.add<double>("resol_calo_scale_em", 0.15);
0615   desc.add<std::string>("tfDnnLabel", "tracksterSelectionTf");
0616   desc.add<std::string>("eid_input_name", "input");
0617   desc.add<std::string>("eid_output_name_energy", "output/regressed_energy");
0618   desc.add<std::string>("eid_output_name_id", "output/id_probabilities");
0619   desc.add<double>("eid_min_cluster_energy", 2.5);
0620   desc.add<int>("eid_n_layers", 50);
0621   desc.add<int>("eid_n_clusters", 10);
0622   descriptions.add("trackstersMergeProducer", desc);
0623 }
0624
0625 DEFINE_FWK_MODULE(TrackstersMergeProducer);