CMS 3D CMS Logo

DeepTauIdBase.h
Go to the documentation of this file.
1 #ifndef DeepTauIdBase_H
2 #define DeepTauIdBase_H
3 
4 /*
5  * \class DeepTauIdBase
6  *
7  * Base class for DeepTauId producers: DeepTauId and DeepTauIdSONIC
8  *
9  */
10 
33 #include "tensorflow/core/util/memmapped_file_system.h"
34 
35 #include <Math/VectorUtil.h>
36 #include <map>
37 #include <fstream>
38 #include "oneapi/tbb/concurrent_unordered_set.h"
39 
40 namespace deep_tau {
48  };
49 
51 } // namespace deep_tau
52 
53 namespace {
54 
55  namespace dnn_inputs_v2 {
56  constexpr int number_of_inner_cell = 11;
57  constexpr int number_of_outer_cell = 21;
58  constexpr int number_of_conv_features = 64;
59  namespace TauBlockInputs {
60  enum vars {
61  rho = 0,
62  tau_pt,
63  tau_eta,
64  tau_phi,
65  tau_mass,
66  tau_E_over_pt,
67  tau_charge,
68  tau_n_charged_prongs,
69  tau_n_neutral_prongs,
70  chargedIsoPtSum,
71  chargedIsoPtSumdR03_over_dR05,
72  footprintCorrection,
73  neutralIsoPtSum,
74  neutralIsoPtSumWeight_over_neutralIsoPtSum,
75  neutralIsoPtSumWeightdR03_over_neutralIsoPtSum,
76  neutralIsoPtSumdR03_over_dR05,
77  photonPtSumOutsideSignalCone,
78  puCorrPtSum,
79  tau_dxy_pca_x,
80  tau_dxy_pca_y,
81  tau_dxy_pca_z,
82  tau_dxy_valid,
83  tau_dxy,
84  tau_dxy_sig,
85  tau_ip3d_valid,
86  tau_ip3d,
87  tau_ip3d_sig,
88  tau_dz,
89  tau_dz_sig_valid,
90  tau_dz_sig,
91  tau_flightLength_x,
92  tau_flightLength_y,
93  tau_flightLength_z,
94  tau_flightLength_sig,
95  tau_pt_weighted_deta_strip,
96  tau_pt_weighted_dphi_strip,
97  tau_pt_weighted_dr_signal,
98  tau_pt_weighted_dr_iso,
99  tau_leadingTrackNormChi2,
100  tau_e_ratio_valid,
101  tau_e_ratio,
102  tau_gj_angle_diff_valid,
103  tau_gj_angle_diff,
104  tau_n_photons,
105  tau_emFraction,
106  tau_inside_ecal_crack,
107  leadChargedCand_etaAtEcalEntrance_minus_tau_eta,
108  NumberOfInputs
109  };
110  inline std::vector<int> varsToDrop = {
111  tau_phi, tau_dxy_pca_x, tau_dxy_pca_y, tau_dxy_pca_z}; // indices of vars to be dropped in the full var enum
112  } // namespace TauBlockInputs
113 
114  namespace EgammaBlockInputs {
115  enum vars {
116  rho = 0,
117  tau_pt,
118  tau_eta,
119  tau_inside_ecal_crack,
120  pfCand_ele_valid,
121  pfCand_ele_rel_pt,
122  pfCand_ele_deta,
123  pfCand_ele_dphi,
124  pfCand_ele_pvAssociationQuality,
125  pfCand_ele_puppiWeight,
126  pfCand_ele_charge,
127  pfCand_ele_lostInnerHits,
128  pfCand_ele_numberOfPixelHits,
129  pfCand_ele_vertex_dx,
130  pfCand_ele_vertex_dy,
131  pfCand_ele_vertex_dz,
132  pfCand_ele_vertex_dx_tauFL,
133  pfCand_ele_vertex_dy_tauFL,
134  pfCand_ele_vertex_dz_tauFL,
135  pfCand_ele_hasTrackDetails,
136  pfCand_ele_dxy,
137  pfCand_ele_dxy_sig,
138  pfCand_ele_dz,
139  pfCand_ele_dz_sig,
140  pfCand_ele_track_chi2_ndof,
141  pfCand_ele_track_ndof,
142  ele_valid,
143  ele_rel_pt,
144  ele_deta,
145  ele_dphi,
146  ele_cc_valid,
147  ele_cc_ele_rel_energy,
148  ele_cc_gamma_rel_energy,
149  ele_cc_n_gamma,
150  ele_rel_trackMomentumAtVtx,
151  ele_rel_trackMomentumAtCalo,
152  ele_rel_trackMomentumOut,
153  ele_rel_trackMomentumAtEleClus,
154  ele_rel_trackMomentumAtVtxWithConstraint,
155  ele_rel_ecalEnergy,
156  ele_ecalEnergy_sig,
157  ele_eSuperClusterOverP,
158  ele_eSeedClusterOverP,
159  ele_eSeedClusterOverPout,
160  ele_eEleClusterOverPout,
161  ele_deltaEtaSuperClusterTrackAtVtx,
162  ele_deltaEtaSeedClusterTrackAtCalo,
163  ele_deltaEtaEleClusterTrackAtCalo,
164  ele_deltaPhiEleClusterTrackAtCalo,
165  ele_deltaPhiSuperClusterTrackAtVtx,
166  ele_deltaPhiSeedClusterTrackAtCalo,
167  ele_mvaInput_earlyBrem,
168  ele_mvaInput_lateBrem,
169  ele_mvaInput_sigmaEtaEta,
170  ele_mvaInput_hadEnergy,
171  ele_mvaInput_deltaEta,
172  ele_gsfTrack_normalizedChi2,
173  ele_gsfTrack_numberOfValidHits,
174  ele_rel_gsfTrack_pt,
175  ele_gsfTrack_pt_sig,
176  ele_has_closestCtfTrack,
177  ele_closestCtfTrack_normalizedChi2,
178  ele_closestCtfTrack_numberOfValidHits,
179  pfCand_gamma_valid,
180  pfCand_gamma_rel_pt,
181  pfCand_gamma_deta,
182  pfCand_gamma_dphi,
183  pfCand_gamma_pvAssociationQuality,
184  pfCand_gamma_fromPV,
185  pfCand_gamma_puppiWeight,
186  pfCand_gamma_puppiWeightNoLep,
187  pfCand_gamma_lostInnerHits,
188  pfCand_gamma_numberOfPixelHits,
189  pfCand_gamma_vertex_dx,
190  pfCand_gamma_vertex_dy,
191  pfCand_gamma_vertex_dz,
192  pfCand_gamma_vertex_dx_tauFL,
193  pfCand_gamma_vertex_dy_tauFL,
194  pfCand_gamma_vertex_dz_tauFL,
195  pfCand_gamma_hasTrackDetails,
196  pfCand_gamma_dxy,
197  pfCand_gamma_dxy_sig,
198  pfCand_gamma_dz,
199  pfCand_gamma_dz_sig,
200  pfCand_gamma_track_chi2_ndof,
201  pfCand_gamma_track_ndof,
202  NumberOfInputs
203  };
204  }
205 
206  namespace MuonBlockInputs {
207  enum vars {
208  rho = 0,
209  tau_pt,
210  tau_eta,
211  tau_inside_ecal_crack,
212  pfCand_muon_valid,
213  pfCand_muon_rel_pt,
214  pfCand_muon_deta,
215  pfCand_muon_dphi,
216  pfCand_muon_pvAssociationQuality,
217  pfCand_muon_fromPV,
218  pfCand_muon_puppiWeight,
219  pfCand_muon_charge,
220  pfCand_muon_lostInnerHits,
221  pfCand_muon_numberOfPixelHits,
222  pfCand_muon_vertex_dx,
223  pfCand_muon_vertex_dy,
224  pfCand_muon_vertex_dz,
225  pfCand_muon_vertex_dx_tauFL,
226  pfCand_muon_vertex_dy_tauFL,
227  pfCand_muon_vertex_dz_tauFL,
228  pfCand_muon_hasTrackDetails,
229  pfCand_muon_dxy,
230  pfCand_muon_dxy_sig,
231  pfCand_muon_dz,
232  pfCand_muon_dz_sig,
233  pfCand_muon_track_chi2_ndof,
234  pfCand_muon_track_ndof,
235  muon_valid,
236  muon_rel_pt,
237  muon_deta,
238  muon_dphi,
239  muon_dxy,
240  muon_dxy_sig,
241  muon_normalizedChi2_valid,
242  muon_normalizedChi2,
243  muon_numberOfValidHits,
244  muon_segmentCompatibility,
245  muon_caloCompatibility,
246  muon_pfEcalEnergy_valid,
247  muon_rel_pfEcalEnergy,
248  muon_n_matches_DT_1,
249  muon_n_matches_DT_2,
250  muon_n_matches_DT_3,
251  muon_n_matches_DT_4,
252  muon_n_matches_CSC_1,
253  muon_n_matches_CSC_2,
254  muon_n_matches_CSC_3,
255  muon_n_matches_CSC_4,
256  muon_n_matches_RPC_1,
257  muon_n_matches_RPC_2,
258  muon_n_matches_RPC_3,
259  muon_n_matches_RPC_4,
260  muon_n_hits_DT_1,
261  muon_n_hits_DT_2,
262  muon_n_hits_DT_3,
263  muon_n_hits_DT_4,
264  muon_n_hits_CSC_1,
265  muon_n_hits_CSC_2,
266  muon_n_hits_CSC_3,
267  muon_n_hits_CSC_4,
268  muon_n_hits_RPC_1,
269  muon_n_hits_RPC_2,
270  muon_n_hits_RPC_3,
271  muon_n_hits_RPC_4,
272  NumberOfInputs
273  };
274  }
275 
276  namespace HadronBlockInputs {
277  enum vars {
278  rho = 0,
279  tau_pt,
280  tau_eta,
281  tau_inside_ecal_crack,
282  pfCand_chHad_valid,
283  pfCand_chHad_rel_pt,
284  pfCand_chHad_deta,
285  pfCand_chHad_dphi,
286  pfCand_chHad_leadChargedHadrCand,
287  pfCand_chHad_pvAssociationQuality,
288  pfCand_chHad_fromPV,
289  pfCand_chHad_puppiWeight,
290  pfCand_chHad_puppiWeightNoLep,
291  pfCand_chHad_charge,
292  pfCand_chHad_lostInnerHits,
293  pfCand_chHad_numberOfPixelHits,
294  pfCand_chHad_vertex_dx,
295  pfCand_chHad_vertex_dy,
296  pfCand_chHad_vertex_dz,
297  pfCand_chHad_vertex_dx_tauFL,
298  pfCand_chHad_vertex_dy_tauFL,
299  pfCand_chHad_vertex_dz_tauFL,
300  pfCand_chHad_hasTrackDetails,
301  pfCand_chHad_dxy,
302  pfCand_chHad_dxy_sig,
303  pfCand_chHad_dz,
304  pfCand_chHad_dz_sig,
305  pfCand_chHad_track_chi2_ndof,
306  pfCand_chHad_track_ndof,
307  pfCand_chHad_hcalFraction,
308  pfCand_chHad_rawCaloFraction,
309  pfCand_nHad_valid,
310  pfCand_nHad_rel_pt,
311  pfCand_nHad_deta,
312  pfCand_nHad_dphi,
313  pfCand_nHad_puppiWeight,
314  pfCand_nHad_puppiWeightNoLep,
315  pfCand_nHad_hcalFraction,
316  NumberOfInputs
317  };
318  }
319  } // namespace dnn_inputs_v2
320 
321  inline float getTauID(const pat::Tau& tau,
322  const std::string& tauID,
323  float default_value = -999.,
324  bool assert_input = true) {
325  static tbb::concurrent_unordered_set<std::string> isFirstWarning;
326  if (tau.isTauIDAvailable(tauID)) {
327  return tau.tauID(tauID);
328  } else {
329  if (assert_input) {
330  throw cms::Exception("DeepTauId")
331  << "Exception in <getTauID>: No tauID '" << tauID << "' available in pat::Tau given as function argument.";
332  }
333  if (isFirstWarning.insert(tauID).second) {
334  edm::LogWarning("DeepTauID") << "Warning in <getTauID>: No tauID '" << tauID
335  << "' available in pat::Tau given as function argument."
336  << " Using default_value = " << default_value << " instead." << std::endl;
337  }
338  return default_value;
339  }
340  }
341 
342  struct TauFunc {
343  const reco::TauDiscriminatorContainer* basicTauDiscriminatorCollection;
344  const reco::TauDiscriminatorContainer* basicTauDiscriminatordR03Collection;
347 
348  using BasicDiscr = deep_tau::BasicDiscriminator;
349  std::map<BasicDiscr, size_t> indexMap;
350  std::map<BasicDiscr, size_t> indexMapdR03;
351 
352  const float getChargedIsoPtSum(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
353  return (*basicTauDiscriminatorCollection)[tau_ref].rawValues.at(indexMap.at(BasicDiscr::ChargedIsoPtSum));
354  }
355  const float getChargedIsoPtSum(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
356  return getTauID(tau, "chargedIsoPtSum");
357  }
358  const float getChargedIsoPtSumdR03(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
359  return (*basicTauDiscriminatordR03Collection)[tau_ref].rawValues.at(indexMapdR03.at(BasicDiscr::ChargedIsoPtSum));
360  }
361  const float getChargedIsoPtSumdR03(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
362  return getTauID(tau, "chargedIsoPtSumdR03");
363  }
364  const float getFootprintCorrectiondR03(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
365  return (*basicTauDiscriminatordR03Collection)[tau_ref].rawValues.at(
366  indexMapdR03.at(BasicDiscr::FootprintCorrection));
367  }
368  const float getFootprintCorrectiondR03(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
369  return getTauID(tau, "footprintCorrectiondR03");
370  }
371  const float getFootprintCorrection(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
372  return getTauID(tau, "footprintCorrection");
373  }
374  const float getNeutralIsoPtSum(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
375  return (*basicTauDiscriminatorCollection)[tau_ref].rawValues.at(indexMap.at(BasicDiscr::NeutralIsoPtSum));
376  }
377  const float getNeutralIsoPtSum(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
378  return getTauID(tau, "neutralIsoPtSum");
379  }
380  const float getNeutralIsoPtSumdR03(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
381  return (*basicTauDiscriminatordR03Collection)[tau_ref].rawValues.at(indexMapdR03.at(BasicDiscr::NeutralIsoPtSum));
382  }
383  const float getNeutralIsoPtSumdR03(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
384  return getTauID(tau, "neutralIsoPtSumdR03");
385  }
386  const float getNeutralIsoPtSumWeight(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
387  return (*basicTauDiscriminatorCollection)[tau_ref].rawValues.at(indexMap.at(BasicDiscr::NeutralIsoPtSumWeight));
388  }
389  const float getNeutralIsoPtSumWeight(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
390  return getTauID(tau, "neutralIsoPtSumWeight");
391  }
392  const float getNeutralIsoPtSumdR03Weight(const reco::PFTau& tau,
393  const edm::RefToBase<reco::BaseTau> tau_ref) const {
394  return (*basicTauDiscriminatordR03Collection)[tau_ref].rawValues.at(
395  indexMapdR03.at(BasicDiscr::NeutralIsoPtSumWeight));
396  }
397  const float getNeutralIsoPtSumdR03Weight(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
398  return getTauID(tau, "neutralIsoPtSumWeightdR03");
399  }
400  const float getPhotonPtSumOutsideSignalCone(const reco::PFTau& tau,
401  const edm::RefToBase<reco::BaseTau> tau_ref) const {
402  return (*basicTauDiscriminatorCollection)[tau_ref].rawValues.at(
404  }
405  const float getPhotonPtSumOutsideSignalCone(const pat::Tau& tau,
406  const edm::RefToBase<reco::BaseTau> tau_ref) const {
407  return getTauID(tau, "photonPtSumOutsideSignalCone");
408  }
409  const float getPhotonPtSumOutsideSignalConedR03(const reco::PFTau& tau,
410  const edm::RefToBase<reco::BaseTau> tau_ref) const {
411  return (*basicTauDiscriminatordR03Collection)[tau_ref].rawValues.at(
413  }
414  const float getPhotonPtSumOutsideSignalConedR03(const pat::Tau& tau,
415  const edm::RefToBase<reco::BaseTau> tau_ref) const {
416  return getTauID(tau, "photonPtSumOutsideSignalConedR03");
417  }
418  const float getPuCorrPtSum(const reco::PFTau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
419  return (*basicTauDiscriminatorCollection)[tau_ref].rawValues.at(indexMap.at(BasicDiscr::PUcorrPtSum));
420  }
421  const float getPuCorrPtSum(const pat::Tau& tau, const edm::RefToBase<reco::BaseTau> tau_ref) const {
422  return getTauID(tau, "puCorrPtSum");
423  }
424 
425  auto getdxyPCA(const reco::PFTau& tau, const size_t tau_index) const {
426  return pfTauTransverseImpactParameters->value(tau_index)->dxy_PCA();
427  }
428  auto getdxyPCA(const pat::Tau& tau, const size_t tau_index) const { return tau.dxy_PCA(); }
429  auto getdxy(const reco::PFTau& tau, const size_t tau_index) const {
430  return pfTauTransverseImpactParameters->value(tau_index)->dxy();
431  }
432  auto getdxy(const pat::Tau& tau, const size_t tau_index) const { return tau.dxy(); }
433  auto getdxyError(const reco::PFTau& tau, const size_t tau_index) const {
434  return pfTauTransverseImpactParameters->value(tau_index)->dxy_error();
435  }
436  auto getdxyError(const pat::Tau& tau, const size_t tau_index) const { return tau.dxy_error(); }
437  auto getdxySig(const reco::PFTau& tau, const size_t tau_index) const {
438  return pfTauTransverseImpactParameters->value(tau_index)->dxy_Sig();
439  }
440  auto getdxySig(const pat::Tau& tau, const size_t tau_index) const { return tau.dxy_Sig(); }
441  auto getip3d(const reco::PFTau& tau, const size_t tau_index) const {
442  return pfTauTransverseImpactParameters->value(tau_index)->ip3d();
443  }
444  auto getip3d(const pat::Tau& tau, const size_t tau_index) const { return tau.ip3d(); }
445  auto getip3dError(const reco::PFTau& tau, const size_t tau_index) const {
446  return pfTauTransverseImpactParameters->value(tau_index)->ip3d_error();
447  }
448  auto getip3dError(const pat::Tau& tau, const size_t tau_index) const { return tau.ip3d_error(); }
449  auto getip3dSig(const reco::PFTau& tau, const size_t tau_index) const {
450  return pfTauTransverseImpactParameters->value(tau_index)->ip3d_Sig();
451  }
452  auto getip3dSig(const pat::Tau& tau, const size_t tau_index) const { return tau.ip3d_Sig(); }
453  auto getHasSecondaryVertex(const reco::PFTau& tau, const size_t tau_index) const {
454  return pfTauTransverseImpactParameters->value(tau_index)->hasSecondaryVertex();
455  }
456  auto getHasSecondaryVertex(const pat::Tau& tau, const size_t tau_index) const { return tau.hasSecondaryVertex(); }
457  auto getFlightLength(const reco::PFTau& tau, const size_t tau_index) const {
458  return pfTauTransverseImpactParameters->value(tau_index)->flightLength();
459  }
460  auto getFlightLength(const pat::Tau& tau, const size_t tau_index) const { return tau.flightLength(); }
461  auto getFlightLengthSig(const reco::PFTau& tau, const size_t tau_index) const {
462  return pfTauTransverseImpactParameters->value(tau_index)->flightLengthSig();
463  }
464  auto getFlightLengthSig(const pat::Tau& tau, const size_t tau_index) const { return tau.flightLengthSig(); }
465 
466  auto getLeadingTrackNormChi2(const reco::PFTau& tau) { return reco::tau::lead_track_chi2(tau); }
467  auto getLeadingTrackNormChi2(const pat::Tau& tau) { return tau.leadingTrackNormChi2(); }
468  auto getEmFraction(const pat::Tau& tau) { return tau.emFraction_MVA(); }
469  auto getEmFraction(const reco::PFTau& tau) { return tau.emFraction(); }
470  auto getEtaAtEcalEntrance(const pat::Tau& tau) { return tau.etaAtEcalEntranceLeadChargedCand(); }
471  auto getEtaAtEcalEntrance(const reco::PFTau& tau) {
472  return tau.leadPFChargedHadrCand()->positionAtECALEntrance().eta();
473  }
474  auto getEcalEnergyLeadingChargedHadr(const reco::PFTau& tau) { return tau.leadPFChargedHadrCand()->ecalEnergy(); }
475  auto getEcalEnergyLeadingChargedHadr(const pat::Tau& tau) { return tau.ecalEnergyLeadChargedHadrCand(); }
476  auto getHcalEnergyLeadingChargedHadr(const reco::PFTau& tau) { return tau.leadPFChargedHadrCand()->hcalEnergy(); }
477  auto getHcalEnergyLeadingChargedHadr(const pat::Tau& tau) { return tau.hcalEnergyLeadChargedHadrCand(); }
478 
479  template <typename PreDiscrType>
480  bool passPrediscriminants(const PreDiscrType prediscriminants,
481  const size_t andPrediscriminants,
482  const edm::RefToBase<reco::BaseTau> tau_ref) {
483  bool passesPrediscriminants = (andPrediscriminants ? 1 : 0);
484  // check tau passes prediscriminants
485  size_t nPrediscriminants = prediscriminants.size();
486  for (size_t iDisc = 0; iDisc < nPrediscriminants; ++iDisc) {
487  // current discriminant result for this tau
488  double discResult = (*prediscriminants[iDisc].handle)[tau_ref];
489  uint8_t thisPasses = (discResult > prediscriminants[iDisc].cut) ? 1 : 0;
490 
491  // if we are using the AND option, as soon as one fails,
492  // the result is FAIL and we can quit looping.
493  // if we are using the OR option as soon as one passes,
494  // the result is pass and we can quit looping
495 
496  // truth table
497  // | result (thisPasses)
498  // | F | T
499  //-----------------------------------
500  // AND(T) | res=fails | continue
501  // | break |
502  //-----------------------------------
503  // OR (F) | continue | res=passes
504  // | | break
505 
506  if (thisPasses ^ andPrediscriminants) //XOR
507  {
508  passesPrediscriminants = (andPrediscriminants ? 0 : 1); //NOR
509  break;
510  }
511  }
512  return passesPrediscriminants;
513  }
514  };
515 
516  namespace candFunc {
517  inline auto getTauDz(const reco::PFCandidate& cand) { return cand.bestTrack()->dz(); }
518  inline auto getTauDz(const pat::PackedCandidate& cand) { return cand.dz(); }
519  inline auto getTauDZSigValid(const reco::PFCandidate& cand) {
520  return cand.bestTrack() != nullptr && std::isnormal(cand.bestTrack()->dz()) && std::isnormal(cand.dzError()) &&
521  cand.dzError() > 0;
522  }
523  inline auto getTauDZSigValid(const pat::PackedCandidate& cand) {
524  return cand.hasTrackDetails() && std::isnormal(cand.dz()) && std::isnormal(cand.dzError()) && cand.dzError() > 0;
525  }
526  inline auto getTauDxy(const reco::PFCandidate& cand) { return cand.bestTrack()->dxy(); }
527  inline auto getTauDxy(const pat::PackedCandidate& cand) { return cand.dxy(); }
528  inline auto getPvAssocationQuality(const reco::PFCandidate& cand) { return 0.7013f; }
529  inline auto getPvAssocationQuality(const pat::PackedCandidate& cand) { return cand.pvAssociationQuality(); }
530  inline auto getPuppiWeight(const reco::PFCandidate& cand, const float aod_value) { return aod_value; }
531  inline auto getPuppiWeight(const pat::PackedCandidate& cand, const float aod_value) { return cand.puppiWeight(); }
532  inline auto getPuppiWeightNoLep(const reco::PFCandidate& cand, const float aod_value) { return aod_value; }
533  inline auto getPuppiWeightNoLep(const pat::PackedCandidate& cand, const float aod_value) {
534  return cand.puppiWeightNoLep();
535  }
536  inline auto getLostInnerHits(const reco::PFCandidate& cand, float default_value) {
537  return cand.bestTrack() != nullptr
538  ? cand.bestTrack()->hitPattern().numberOfLostHits(reco::HitPattern::MISSING_INNER_HITS)
539  : default_value;
540  }
541  inline auto getLostInnerHits(const pat::PackedCandidate& cand, float default_value) { return cand.lostInnerHits(); }
542  inline auto getNumberOfPixelHits(const reco::PFCandidate& cand, float default_value) {
543  return cand.bestTrack() != nullptr
544  ? cand.bestTrack()->hitPattern().numberOfLostHits(reco::HitPattern::MISSING_INNER_HITS)
545  : default_value;
546  }
547  inline auto getNumberOfPixelHits(const pat::PackedCandidate& cand, float default_value) {
548  return cand.numberOfPixelHits();
549  }
550  inline auto getHasTrackDetails(const reco::PFCandidate& cand) { return cand.bestTrack() != nullptr; }
551  inline auto getHasTrackDetails(const pat::PackedCandidate& cand) { return cand.hasTrackDetails(); }
552  inline auto getPseudoTrack(const reco::PFCandidate& cand) { return *cand.bestTrack(); }
553  inline auto getPseudoTrack(const pat::PackedCandidate& cand) { return cand.pseudoTrack(); }
554  inline auto getFromPV(const reco::PFCandidate& cand) { return 0.9994f; }
555  inline auto getFromPV(const pat::PackedCandidate& cand) { return cand.fromPV(); }
556  inline auto getHCalFraction(const reco::PFCandidate& cand, bool disable_hcalFraction_workaround) {
557  return cand.rawHcalEnergy() / (cand.rawHcalEnergy() + cand.rawEcalEnergy());
558  }
559  inline auto getHCalFraction(const pat::PackedCandidate& cand, bool disable_hcalFraction_workaround) {
560  float hcal_fraction = 0.;
562  // CV: use consistent definition for pfCand_chHad_hcalFraction
563  // in DeepTauId.cc code and in TauMLTools/Production/plugins/TauTupleProducer.cc
564  hcal_fraction = cand.hcalFraction();
565  } else {
566  // CV: backwards compatibility with DeepTau training v2p1 used during Run 2
567  if (cand.pdgId() == 1 || cand.pdgId() == 130) {
568  hcal_fraction = cand.hcalFraction();
569  } else if (cand.isIsolatedChargedHadron()) {
570  hcal_fraction = cand.rawHcalFraction();
571  }
572  }
573  return hcal_fraction;
574  }
575  inline auto getRawCaloFraction(const reco::PFCandidate& cand) {
576  return (cand.rawEcalEnergy() + cand.rawHcalEnergy()) / cand.energy();
577  }
578  inline auto getRawCaloFraction(const pat::PackedCandidate& cand) { return cand.rawCaloFraction(); }
579  }; // namespace candFunc
580 
581  template <typename LVector1, typename LVector2>
582  float dEta(const LVector1& p4, const LVector2& tau_p4) {
583  return static_cast<float>(p4.eta() - tau_p4.eta());
584  }
585 
586  template <typename LVector1, typename LVector2>
587  float dPhi(const LVector1& p4_1, const LVector2& p4_2) {
588  return static_cast<float>(reco::deltaPhi(p4_2.phi(), p4_1.phi()));
589  }
590 
591  struct MuonHitMatchV2 {
592  static constexpr size_t n_muon_stations = 4;
593  static constexpr int first_station_id = 1;
594  static constexpr int last_station_id = first_station_id + n_muon_stations - 1;
595  using CountArray = std::array<unsigned, n_muon_stations>;
596  using CountMap = std::map<int, CountArray>;
597 
598  const std::vector<int>& consideredSubdets() {
599  static const std::vector<int> subdets = {MuonSubdetId::DT, MuonSubdetId::CSC, MuonSubdetId::RPC};
600  return subdets;
601  }
602 
603  const std::string& subdetName(int subdet) {
604  static const std::map<int, std::string> subdet_names = {
605  {MuonSubdetId::DT, "DT"}, {MuonSubdetId::CSC, "CSC"}, {MuonSubdetId::RPC, "RPC"}};
606  if (!subdet_names.count(subdet))
607  throw cms::Exception("MuonHitMatch") << "Subdet name for subdet id " << subdet << " not found.";
608  return subdet_names.at(subdet);
609  }
610 
611  size_t getStationIndex(int station, bool throw_exception) const {
612  if (station < first_station_id || station > last_station_id) {
613  if (throw_exception)
614  throw cms::Exception("MuonHitMatch") << "Station id is out of range";
616  }
617  return static_cast<size_t>(station - 1);
618  }
619 
620  MuonHitMatchV2(const pat::Muon& muon) {
621  for (int subdet : consideredSubdets()) {
622  n_matches[subdet].fill(0);
623  n_hits[subdet].fill(0);
624  }
625 
626  countMatches(muon, n_matches);
627  countHits(muon, n_hits);
628  }
629 
630  void countMatches(const pat::Muon& muon, CountMap& n_matches) {
631  for (const auto& segment : muon.matches()) {
632  if (segment.segmentMatches.empty() && segment.rpcMatches.empty())
633  continue;
634  if (n_matches.count(segment.detector())) {
635  const size_t station_index = getStationIndex(segment.station(), true);
636  ++n_matches.at(segment.detector()).at(station_index);
637  }
638  }
639  }
640 
641  void countHits(const pat::Muon& muon, CountMap& n_hits) {
642  if (muon.outerTrack().isNonnull()) {
643  const auto& hit_pattern = muon.outerTrack()->hitPattern();
644  for (int hit_index = 0; hit_index < hit_pattern.numberOfAllHits(reco::HitPattern::TRACK_HITS); ++hit_index) {
645  auto hit_id = hit_pattern.getHitPattern(reco::HitPattern::TRACK_HITS, hit_index);
646  if (hit_id == 0)
647  break;
648  if (hit_pattern.muonHitFilter(hit_id) && (hit_pattern.getHitType(hit_id) == TrackingRecHit::valid ||
649  hit_pattern.getHitType(hit_id) == TrackingRecHit::bad)) {
650  const size_t station_index = getStationIndex(hit_pattern.getMuonStation(hit_id), false);
651  if (station_index < n_muon_stations) {
652  CountArray* muon_n_hits = nullptr;
653  if (hit_pattern.muonDTHitFilter(hit_id))
654  muon_n_hits = &n_hits.at(MuonSubdetId::DT);
655  else if (hit_pattern.muonCSCHitFilter(hit_id))
656  muon_n_hits = &n_hits.at(MuonSubdetId::CSC);
657  else if (hit_pattern.muonRPCHitFilter(hit_id))
658  muon_n_hits = &n_hits.at(MuonSubdetId::RPC);
659 
660  if (muon_n_hits)
661  ++muon_n_hits->at(station_index);
662  }
663  }
664  }
665  }
666  }
667 
668  unsigned nMatches(int subdet, int station) const {
669  if (!n_matches.count(subdet))
670  throw cms::Exception("MuonHitMatch") << "Subdet " << subdet << " not found.";
671  const size_t station_index = getStationIndex(station, true);
672  return n_matches.at(subdet).at(station_index);
673  }
674 
675  unsigned nHits(int subdet, int station) const {
676  if (!n_hits.count(subdet))
677  throw cms::Exception("MuonHitMatch") << "Subdet " << subdet << " not found.";
678  const size_t station_index = getStationIndex(station, true);
679  return n_hits.at(subdet).at(station_index);
680  }
681 
682  unsigned countMuonStationsWithMatches(int first_station, int last_station) const {
683  static const std::map<int, std::vector<bool>> masks = {
684  {MuonSubdetId::DT, {false, false, false, false}},
685  {MuonSubdetId::CSC, {true, false, false, false}},
686  {MuonSubdetId::RPC, {false, false, false, false}},
687  };
688  const size_t first_station_index = getStationIndex(first_station, true);
689  const size_t last_station_index = getStationIndex(last_station, true);
690  unsigned cnt = 0;
691  for (size_t n = first_station_index; n <= last_station_index; ++n) {
692  for (const auto& match : n_matches) {
693  if (!masks.at(match.first).at(n) && match.second.at(n) > 0)
694  ++cnt;
695  }
696  }
697  return cnt;
698  }
699 
700  unsigned countMuonStationsWithHits(int first_station, int last_station) const {
701  static const std::map<int, std::vector<bool>> masks = {
702  {MuonSubdetId::DT, {false, false, false, false}},
703  {MuonSubdetId::CSC, {false, false, false, false}},
704  {MuonSubdetId::RPC, {false, false, false, false}},
705  };
706 
707  const size_t first_station_index = getStationIndex(first_station, true);
708  const size_t last_station_index = getStationIndex(last_station, true);
709  unsigned cnt = 0;
710  for (size_t n = first_station_index; n <= last_station_index; ++n) {
711  for (const auto& hit : n_hits) {
712  if (!masks.at(hit.first).at(n) && hit.second.at(n) > 0)
713  ++cnt;
714  }
715  }
716  return cnt;
717  }
718 
719  private:
720  CountMap n_matches, n_hits;
721  };
722 
723  enum class CellObjectType {
724  PfCand_electron,
725  PfCand_muon,
726  PfCand_chargedHadron,
727  PfCand_neutralHadron,
728  PfCand_gamma,
729  Electron,
730  Muon,
731  Other
732  };
733 
734  template <typename Object>
735  inline CellObjectType GetCellObjectType(const Object&);
736  template <>
737  inline CellObjectType GetCellObjectType(const pat::Electron&) {
739  }
740  template <>
741  inline CellObjectType GetCellObjectType(const pat::Muon&) {
742  return CellObjectType::Muon;
743  }
744 
745  template <>
746  inline CellObjectType GetCellObjectType(reco::Candidate const& cand) {
747  static const std::map<int, CellObjectType> obj_types = {{11, CellObjectType::PfCand_electron},
748  {13, CellObjectType::PfCand_muon},
749  {22, CellObjectType::PfCand_gamma},
750  {130, CellObjectType::PfCand_neutralHadron},
751  {211, CellObjectType::PfCand_chargedHadron}};
752 
753  auto iter = obj_types.find(std::abs(cand.pdgId()));
754  if (iter == obj_types.end())
755  return CellObjectType::Other;
756  return iter->second;
757  }
758 
759  using Cell = std::map<CellObjectType, size_t>;
760  struct CellIndex {
761  int eta, phi;
762 
763  bool operator<(const CellIndex& other) const {
764  if (eta != other.eta)
765  return eta < other.eta;
766  return phi < other.phi;
767  }
768  };
769 
770  class CellGrid {
771  public:
772  using Map = std::map<CellIndex, Cell>;
773  using const_iterator = Map::const_iterator;
774 
775  CellGrid(unsigned n_cells_eta,
776  unsigned n_cells_phi,
777  double cell_size_eta,
778  double cell_size_phi,
780  : nCellsEta(n_cells_eta),
781  nCellsPhi(n_cells_phi),
782  nTotal(nCellsEta * nCellsPhi),
783  cellSizeEta(cell_size_eta),
784  cellSizePhi(cell_size_phi),
785  disable_CellIndex_workaround_(disable_CellIndex_workaround) {
786  if (nCellsEta % 2 != 1 || nCellsEta < 1)
787  throw cms::Exception("DeepTauId") << "Invalid number of eta cells.";
788  if (nCellsPhi % 2 != 1 || nCellsPhi < 1)
789  throw cms::Exception("DeepTauId") << "Invalid number of phi cells.";
790  if (cellSizeEta <= 0 || cellSizePhi <= 0)
791  throw cms::Exception("DeepTauId") << "Invalid cell size.";
792  }
793 
794  int maxEtaIndex() const { return static_cast<int>((nCellsEta - 1) / 2); }
795  int maxPhiIndex() const { return static_cast<int>((nCellsPhi - 1) / 2); }
796  double maxDeltaEta() const { return cellSizeEta * (0.5 + maxEtaIndex()); }
797  double maxDeltaPhi() const { return cellSizePhi * (0.5 + maxPhiIndex()); }
798  int getEtaTensorIndex(const CellIndex& cellIndex) const { return cellIndex.eta + maxEtaIndex(); }
799  int getPhiTensorIndex(const CellIndex& cellIndex) const { return cellIndex.phi + maxPhiIndex(); }
800 
801  bool tryGetCellIndex(double deltaEta, double deltaPhi, CellIndex& cellIndex) const {
802  const auto getCellIndex = [this](double x, double maxX, double size, int& index) {
803  const double absX = std::abs(x);
804  if (absX > maxX)
805  return false;
806  double absIndex;
807  if (disable_CellIndex_workaround_) {
808  // CV: use consistent definition for CellIndex
809  // in DeepTauId.cc code and new DeepTau trainings
810  absIndex = std::floor(absX / size + 0.5);
811  } else {
812  // CV: backwards compatibility with DeepTau training v2p1 used during Run 2
813  absIndex = std::floor(std::abs(absX / size - 0.5));
814  }
815  index = static_cast<int>(std::copysign(absIndex, x));
816  return true;
817  };
818 
819  return getCellIndex(deltaEta, maxDeltaEta(), cellSizeEta, cellIndex.eta) &&
820  getCellIndex(deltaPhi, maxDeltaPhi(), cellSizePhi, cellIndex.phi);
821  }
822 
823  size_t num_valid_cells() const { return cells.size(); }
824  Cell& operator[](const CellIndex& cellIndex) { return cells[cellIndex]; }
825  const Cell& at(const CellIndex& cellIndex) const { return cells.at(cellIndex); }
826  size_t count(const CellIndex& cellIndex) const { return cells.count(cellIndex); }
827  const_iterator find(const CellIndex& cellIndex) const { return cells.find(cellIndex); }
828  const_iterator begin() const { return cells.begin(); }
829  const_iterator end() const { return cells.end(); }
830 
831  public:
832  const unsigned nCellsEta, nCellsPhi, nTotal;
833  const double cellSizeEta, cellSizePhi;
834 
835  private:
836  std::map<CellIndex, Cell> cells;
837  const bool disable_CellIndex_workaround_;
838  };
839 } // anonymous namespace
840 
841 template <class Producer>
842 class DeepTauIdBase : public Producer {
843 public:
851  using LorentzVectorXYZ = ROOT::Math::LorentzVector<ROOT::Math::PxPyPzE4D<double>>;
853  using CutterPtr = std::unique_ptr<Cutter>;
854  using WPList = std::vector<CutterPtr>;
855 
856  struct IDOutput {
857  std::vector<size_t> num_, den_;
858 
859  IDOutput(const std::vector<size_t>& num, const std::vector<size_t>& den) : num_(num), den_(den) {}
860 
861  // for direct inference, read the output tensorflow::Tensor
862  float read_value(const tensorflow::Tensor& pred, size_t tau_index, size_t elem) const {
863  return pred.matrix<float>()(tau_index, elem);
864  }
865  // for SONIC, read the output vector
866  float read_value(const std::vector<std::vector<float>>& pred, size_t tau_index, size_t elem) const {
867  return pred.at(tau_index).at(elem);
868  }
869 
870  template <typename PredType>
871  std::unique_ptr<TauDiscriminator> get_value(const edm::Handle<TauCollection>& taus,
872  const PredType& pred,
873  const WPList* working_points,
874  bool is_online) const {
875  std::vector<reco::SingleTauDiscriminatorContainer> outputbuffer(taus->size());
876 
877  for (size_t tau_index = 0; tau_index < taus->size(); ++tau_index) {
878  float x = 0;
879  for (size_t num_elem : num_)
880  x += read_value(pred, tau_index, num_elem);
881  if (x != 0 && !den_.empty()) {
882  float den_val = 0;
883  for (size_t den_elem : den_)
884  den_val += read_value(pred, tau_index, den_elem);
885  x = den_val != 0 ? x / den_val : std::numeric_limits<float>::max();
886  }
887  outputbuffer[tau_index].rawValues.push_back(x);
888  if (working_points) {
889  for (const auto& wp : *working_points) {
890  const bool pass = x > (*wp)(taus->at(tau_index), is_online);
891  outputbuffer[tau_index].workingPoints.push_back(pass);
892  }
893  }
894  }
895  std::unique_ptr<TauDiscriminator> output = std::make_unique<TauDiscriminator>();
897  filler.insert(taus, outputbuffer.begin(), outputbuffer.end());
898  filler.fill();
899  return output;
900  }
901  };
902 
903  using IDOutputCollection = std::map<std::string, IDOutput>;
904 
905  static constexpr float default_value = -999.;
906 
908  static constexpr size_t e_index = 0, mu_index = 1, tau_index = 2, jet_index = 3;
909  static const IDOutputCollection idoutputs_ = {
910  {"VSe", IDOutput({tau_index}, {e_index, tau_index})},
911  {"VSmu", IDOutput({tau_index}, {mu_index, tau_index})},
912  {"VSjet", IDOutput({tau_index}, {jet_index, tau_index})},
913  };
914  return idoutputs_;
915  }
916 
918 
919  const std::map<BasicDiscriminator, size_t> matchDiscriminatorIndices(
920  edm::Event const& event,
921  edm::EDGetTokenT<reco::TauDiscriminatorContainer> discriminatorContainerToken,
922  std::vector<BasicDiscriminator> requiredDiscr) {
923  std::map<std::string, size_t> discrIndexMapStr;
924  auto const aHandle = event.getHandle(discriminatorContainerToken);
925  auto const aProv = aHandle.provenance();
926  if (aProv == nullptr)
927  aHandle.whyFailed()->raise();
928  const auto& psetsFromProvenance = edm::parameterSet(aProv->stable(), event.processHistory());
929  auto const idlist = psetsFromProvenance.getParameter<std::vector<edm::ParameterSet>>("IDdefinitions");
930  for (size_t j = 0; j < idlist.size(); ++j) {
931  std::string idname = idlist[j].getParameter<std::string>("IDname");
932  if (discrIndexMapStr.count(idname)) {
933  throw cms::Exception("DeepTauId")
934  << "basic discriminator " << idname << " appears more than once in the input.";
935  }
936  discrIndexMapStr[idname] = j;
937  }
938 
939  //translate to a map of <BasicDiscriminator, index> and check if all discriminators are present
940  std::map<BasicDiscriminator, size_t> discrIndexMap;
941  for (size_t i = 0; i < requiredDiscr.size(); i++) {
942  if (discrIndexMapStr.find(stringFromDiscriminator_.at(requiredDiscr[i])) == discrIndexMapStr.end())
943  throw cms::Exception("DeepTauId") << "Basic Discriminator " << stringFromDiscriminator_.at(requiredDiscr[i])
944  << " was not provided in the config file.";
945  else
946  discrIndexMap[requiredDiscr[i]] = discrIndexMapStr[stringFromDiscriminator_.at(requiredDiscr[i])];
947  }
948  return discrIndexMap;
949  }
950 
952  desc.add<edm::InputTag>("electrons", edm::InputTag("slimmedElectrons"));
953  desc.add<edm::InputTag>("muons", edm::InputTag("slimmedMuons"));
954  desc.add<edm::InputTag>("taus", edm::InputTag("slimmedTaus"));
955  desc.add<edm::InputTag>("pfcands", edm::InputTag("packedPFCandidates"));
956  desc.add<edm::InputTag>("vertices", edm::InputTag("offlineSlimmedPrimaryVertices"));
957  desc.add<edm::InputTag>("rho", edm::InputTag("fixedGridRhoAll"));
958  desc.add<bool>("mem_mapped", false);
959  desc.add<unsigned>("year", 2017);
960  desc.add<unsigned>("version", 2);
961  desc.add<unsigned>("sub_version", 1);
962  desc.add<int>("debug_level", 0);
963  desc.add<bool>("disable_dxy_pca", false);
964  desc.add<bool>("disable_hcalFraction_workaround", false);
965  desc.add<bool>("disable_CellIndex_workaround", false);
966  desc.add<bool>("save_inputs", false);
967  desc.add<bool>("is_online", false);
968 
969  desc.add<std::vector<std::string>>("VSeWP", {"-1."});
970  desc.add<std::vector<std::string>>("VSmuWP", {"-1."});
971  desc.add<std::vector<std::string>>("VSjetWP", {"-1."});
972 
973  desc.addUntracked<edm::InputTag>("basicTauDiscriminators", edm::InputTag("basicTauDiscriminators"));
974  desc.addUntracked<edm::InputTag>("basicTauDiscriminatorsdR03", edm::InputTag("basicTauDiscriminatorsdR03"));
975  desc.add<edm::InputTag>("pfTauTransverseImpactParameters", edm::InputTag("hpsPFTauTransverseImpactParameters"));
976 
977  {
978  edm::ParameterSetDescription pset_Prediscriminants;
979  pset_Prediscriminants.add<std::string>("BooleanOperator", "and");
980  {
982  psd1.add<double>("cut");
983  psd1.add<edm::InputTag>("Producer");
984  pset_Prediscriminants.addOptional<edm::ParameterSetDescription>("decayMode", psd1);
985  }
986  desc.add<edm::ParameterSetDescription>("Prediscriminants", pset_Prediscriminants);
987  }
988  }
989 
990 public:
992  : Producer(cfg),
993  tausToken_(this->template consumes<TauCollection>(cfg.getParameter<edm::InputTag>("taus"))),
994  pfcandToken_(this->template consumes<CandidateCollection>(cfg.getParameter<edm::InputTag>("pfcands"))),
995  vtxToken_(this->template consumes<reco::VertexCollection>(cfg.getParameter<edm::InputTag>("vertices"))),
996  is_online_(cfg.getParameter<bool>("is_online")),
999  this->template consumes<std::vector<pat::Electron>>(cfg.getParameter<edm::InputTag>("electrons"))),
1000  muons_token_(this->template consumes<std::vector<pat::Muon>>(cfg.getParameter<edm::InputTag>("muons"))),
1001  rho_token_(this->template consumes<double>(cfg.getParameter<edm::InputTag>("rho"))),
1003  cfg.getUntrackedParameter<edm::InputTag>("basicTauDiscriminators"))),
1005  cfg.getUntrackedParameter<edm::InputTag>("basicTauDiscriminatorsdR03"))),
1007  this->template consumes<
1009  cfg.getParameter<edm::InputTag>("pfTauTransverseImpactParameters"))),
1010  year_(cfg.getParameter<unsigned>("year")),
1011  version_(cfg.getParameter<unsigned>("version")),
1012  sub_version_(cfg.getParameter<unsigned>("sub_version")),
1013  debug_level(cfg.getParameter<int>("debug_level")),
1014  disable_dxy_pca_(cfg.getParameter<bool>("disable_dxy_pca")),
1015  disable_hcalFraction_workaround_(cfg.getParameter<bool>("disable_hcalFraction_workaround")),
1016  disable_CellIndex_workaround_(cfg.getParameter<bool>("disable_CellIndex_workaround")),
1017  save_inputs_(cfg.getParameter<bool>("save_inputs")),
1018  json_file_(nullptr),
1019  file_counter_(0) {
1020  for (const auto& output_desc : idoutputs_) {
1021  this->template produces<TauDiscriminator>(output_desc.first);
1022  const auto& cut_list = cfg.getParameter<std::vector<std::string>>(output_desc.first + "WP");
1023  for (const std::string& cut_str : cut_list) {
1024  workingPoints_[output_desc.first].push_back(std::make_unique<Cutter>(cut_str));
1025  }
1026  }
1027 
1028  // prediscriminant operator
1029  // require the tau to pass the following prediscriminants
1030  const edm::ParameterSet& prediscriminantConfig = cfg.getParameter<edm::ParameterSet>("Prediscriminants");
1031 
1032  // determine boolean operator used on the prediscriminants
1033  std::string pdBoolOperator = prediscriminantConfig.getParameter<std::string>("BooleanOperator");
1034  // convert string to lowercase
1035  transform(pdBoolOperator.begin(), pdBoolOperator.end(), pdBoolOperator.begin(), ::tolower);
1036 
1037  if (pdBoolOperator == "and") {
1038  andPrediscriminants_ = 0x1; //use chars instead of bools so we can do a bitwise trick later
1039  } else if (pdBoolOperator == "or") {
1040  andPrediscriminants_ = 0x0;
1041  } else {
1042  throw cms::Exception("TauDiscriminationProducerBase")
1043  << "PrediscriminantBooleanOperator defined incorrectly, options are: AND,OR";
1044  }
1045 
1046  // get the list of prediscriminants
1047  std::vector<std::string> prediscriminantsNames =
1048  prediscriminantConfig.getParameterNamesForType<edm::ParameterSet>();
1049 
1050  for (auto const& iDisc : prediscriminantsNames) {
1051  const edm::ParameterSet& iPredisc = prediscriminantConfig.getParameter<edm::ParameterSet>(iDisc);
1052  const edm::InputTag& label = iPredisc.getParameter<edm::InputTag>("Producer");
1053  double cut = iPredisc.getParameter<double>("cut");
1054 
1055  if (is_online_) {
1056  TauDiscInfo<reco::PFTauDiscriminator> thisDiscriminator;
1057  thisDiscriminator.label = label;
1058  thisDiscriminator.cut = cut;
1059  thisDiscriminator.disc_token = this->template consumes<reco::PFTauDiscriminator>(label);
1060  recoPrediscriminants_.push_back(thisDiscriminator);
1061  } else {
1062  TauDiscInfo<pat::PATTauDiscriminator> thisDiscriminator;
1063  thisDiscriminator.label = label;
1064  thisDiscriminator.cut = cut;
1065  thisDiscriminator.disc_token = this->template consumes<pat::PATTauDiscriminator>(label);
1066  patPrediscriminants_.push_back(thisDiscriminator);
1067  }
1068  }
1069  if (version_ == 2) {
1070  using namespace dnn_inputs_v2;
1071  namespace sc = deep_tau::Scaling;
1072  tauInputs_indices_.resize(TauBlockInputs::NumberOfInputs);
1073  std::iota(std::begin(tauInputs_indices_), std::end(tauInputs_indices_), 0);
1074 
1075  if (sub_version_ == 1) {
1077  } else if (sub_version_ == 5) {
1078  std::sort(TauBlockInputs::varsToDrop.begin(), TauBlockInputs::varsToDrop.end());
1079  for (auto v : TauBlockInputs::varsToDrop) {
1080  tauInputs_indices_.at(v) = TauBlockInputs::NumberOfInputs - TauBlockInputs::varsToDrop.size();
1081  for (std::size_t i = v + 1; i < TauBlockInputs::NumberOfInputs; ++i)
1082  tauInputs_indices_.at(i) -= 1; // shift all the following indices by 1
1083  }
1084  if (year_ == 2026) {
1086  } else {
1088  }
1089  } else
1090  throw cms::Exception("DeepTauId") << "subversion " << sub_version_ << " is not supported.";
1091 
1092  std::map<std::vector<bool>, std::vector<sc::FeatureT>> GridFeatureTypes_map = {
1093  {{false}, {sc::FeatureT::TauFlat, sc::FeatureT::GridGlobal}}, // feature types without inner/outer grid split
1094  {{false, true},
1095  {sc::FeatureT::PfCand_electron,
1096  sc::FeatureT::PfCand_muon, // feature types with inner/outer grid split
1097  sc::FeatureT::PfCand_chHad,
1098  sc::FeatureT::PfCand_nHad,
1099  sc::FeatureT::PfCand_gamma,
1101  sc::FeatureT::Muon}}};
1102 
1103  // check that sizes of mean/std/lim_min/lim_max vectors are equal between each other
1104  for (const auto& p : GridFeatureTypes_map) {
1105  for (auto is_inner : p.first) {
1106  for (auto featureType : p.second) {
1107  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(featureType, is_inner));
1108  if (!(sp.mean_.size() == sp.std_.size() && sp.mean_.size() == sp.lim_min_.size() &&
1109  sp.mean_.size() == sp.lim_max_.size()))
1110  throw cms::Exception("DeepTauId") << "sizes of scaling parameter vectors do not match between each other";
1111  }
1112  }
1113  }
1114  } else {
1115  throw cms::Exception("DeepTauId") << "version " << version_ << " is not supported.";
1116  }
1117  }
1118 
1119  template <typename ConsumeType>
1120  struct TauDiscInfo {
1124  double cut;
1125  void fill(const edm::Event& evt) { evt.getByToken(disc_token, handle); }
1126  };
1127 
1128  // select boolean operation on prediscriminants (and = 0x01, or = 0x00)
1130  std::vector<TauDiscInfo<pat::PATTauDiscriminator>> patPrediscriminants_;
1131  std::vector<TauDiscInfo<reco::PFTauDiscriminator>> recoPrediscriminants_;
1132 
1133 protected:
1134  static constexpr float pi = M_PI;
1135 
1136  template <typename PredType>
1138  for (const auto& output_desc : idoutputs_) {
1139  const WPList* working_points = nullptr;
1140  if (workingPoints_.find(output_desc.first) != workingPoints_.end()) {
1141  working_points = &workingPoints_.at(output_desc.first);
1142  }
1143  auto result = output_desc.second.get_value(taus, pred, working_points, is_online_);
1144  event.put(std::move(result), output_desc.first);
1145  }
1146  }
1147 
1148  template <typename T>
1149  static float getValue(T value) {
1150  return std::isnormal(value) ? static_cast<float>(value) : 0.f;
1151  }
1152 
1153  template <typename T>
1154  static float getValueLinear(T value, float min_value, float max_value, bool positive) {
1155  const float fixed_value = getValue(value);
1156  const float clamped_value = std::clamp(fixed_value, min_value, max_value);
1157  float transformed_value = (clamped_value - min_value) / (max_value - min_value);
1158  if (!positive)
1159  transformed_value = transformed_value * 2 - 1;
1160  return transformed_value;
1161  }
1162 
1163  template <typename T>
1164  static float getValueNorm(T value, float mean, float sigma, float n_sigmas_max = 5) {
1165  const float fixed_value = getValue(value);
1166  const float norm_value = (fixed_value - mean) / sigma;
1167  return std::clamp(norm_value, -n_sigmas_max, n_sigmas_max);
1168  }
1169 
1170  static bool isAbove(double value, double min) { return std::isnormal(value) && value > min; }
1171 
1173  float& cc_ele_energy,
1174  float& cc_gamma_energy,
1175  int& cc_n_gamma) {
1176  cc_ele_energy = cc_gamma_energy = 0;
1177  cc_n_gamma = 0;
1178  const auto& superCluster = ele.superCluster();
1179  if (superCluster.isNonnull() && superCluster.isAvailable() && superCluster->clusters().isNonnull() &&
1180  superCluster->clusters().isAvailable()) {
1181  for (auto iter = superCluster->clustersBegin(); iter != superCluster->clustersEnd(); ++iter) {
1182  const float energy = static_cast<float>((*iter)->energy());
1183  if (iter == superCluster->clustersBegin())
1184  cc_ele_energy += energy;
1185  else {
1186  cc_gamma_energy += energy;
1187  ++cc_n_gamma;
1188  }
1189  }
1190  return true;
1191  } else
1192  return false;
1193  }
1194 
1195 protected:
1196  // load prediscriminators
1198  edm::ProductID tauProductID = taus.id();
1199  size_t nPrediscriminants =
1201  for (size_t iDisc = 0; iDisc < nPrediscriminants; ++iDisc) {
1202  edm::ProductID discKeyId;
1203  if (is_online_) {
1204  recoPrediscriminants_[iDisc].fill(event);
1205  discKeyId = recoPrediscriminants_[iDisc].handle->keyProduct().id();
1206  } else {
1207  patPrediscriminants_[iDisc].fill(event);
1208  discKeyId = patPrediscriminants_[iDisc].handle->keyProduct().id();
1209  }
1210 
1211  // Check to make sure the product is correct for the discriminator.
1212  // If not, throw a more informative exception.
1213  if (tauProductID != discKeyId) {
1214  throw cms::Exception("MisconfiguredPrediscriminant")
1215  << "The tau collection has product ID: " << tauProductID
1216  << " but the pre-discriminator is keyed with product ID: " << discKeyId << std::endl;
1217  }
1218  }
1219  }
1220 
1221  template <typename Collection, typename TauCastType>
1222  void fillGrids(const TauCastType& tau, const Collection& objects, CellGrid& inner_grid, CellGrid& outer_grid) {
1223  static constexpr double outer_dR2 = 0.25; //0.5^2
1224  const double inner_radius = getInnerSignalConeRadius(tau.polarP4().pt());
1225  const double inner_dR2 = std::pow(inner_radius, 2);
1226 
1227  const auto addObject = [&](size_t n, double deta, double dphi, CellGrid& grid) {
1228  const auto& obj = objects.at(n);
1229  const CellObjectType obj_type = GetCellObjectType(obj);
1230  if (obj_type == CellObjectType::Other)
1231  return;
1232  CellIndex cell_index;
1233  if (grid.tryGetCellIndex(deta, dphi, cell_index)) {
1234  Cell& cell = grid[cell_index];
1235  auto iter = cell.find(obj_type);
1236  if (iter != cell.end()) {
1237  const auto& prev_obj = objects.at(iter->second);
1238  if (obj.polarP4().pt() > prev_obj.polarP4().pt())
1239  iter->second = n;
1240  } else {
1241  cell[obj_type] = n;
1242  }
1243  }
1244  };
1245 
1246  for (size_t n = 0; n < objects.size(); ++n) {
1247  const auto& obj = objects.at(n);
1248  const double deta = obj.polarP4().eta() - tau.polarP4().eta();
1249  const double dphi = reco::deltaPhi(obj.polarP4().phi(), tau.polarP4().phi());
1250  const double dR2 = std::pow(deta, 2) + std::pow(dphi, 2);
1251  if (dR2 < inner_dR2)
1252  addObject(n, deta, dphi, inner_grid);
1253  if (dR2 < outer_dR2)
1254  addObject(n, deta, dphi, outer_grid);
1255  }
1256  }
1257 
1258  template <typename CandidateCastType, typename TauCastType, typename TauBlockType>
1259  void createTauBlockInputs(const TauCastType& tau,
1260  const size_t& tau_index,
1261  const edm::RefToBase<reco::BaseTau> tau_ref,
1262  const reco::Vertex& pv,
1263  double rho,
1264  TauFunc tau_funcs,
1265  TauBlockType& tauBlockInputs) {
1266  namespace dnn = dnn_inputs_v2::TauBlockInputs;
1267  namespace sc = deep_tau::Scaling;
1268  namespace candFunc = candFunc;
1269  sc::FeatureT ft = sc::FeatureT::TauFlat;
1270  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft, false));
1271 
1272  const auto& get = [&](int var_index) -> float& {
1273  if constexpr (std::is_same_v<TauBlockType, std::vector<float>::iterator>) {
1274  return *(tauBlockInputs + tauInputs_indices_.at(var_index));
1275  } else {
1276  return ((tensorflow::Tensor)tauBlockInputs).matrix<float>()(0, tauInputs_indices_.at(var_index));
1277  }
1278  };
1279 
1280  auto leadChargedHadrCand = dynamic_cast<const CandidateCastType*>(tau.leadChargedHadrCand().get());
1281 
1282  get(dnn::rho) = sp.scale(rho, tauInputs_indices_[dnn::rho]);
1283  get(dnn::tau_pt) = sp.scale(tau.polarP4().pt(), tauInputs_indices_[dnn::tau_pt]);
1284  get(dnn::tau_eta) = sp.scale(tau.polarP4().eta(), tauInputs_indices_[dnn::tau_eta]);
1285  if (sub_version_ == 1) {
1286  get(dnn::tau_phi) = getValueLinear(tau.polarP4().phi(), -pi, pi, false);
1287  }
1288  get(dnn::tau_mass) = sp.scale(tau.polarP4().mass(), tauInputs_indices_[dnn::tau_mass]);
1289  get(dnn::tau_E_over_pt) = sp.scale(tau.p4().energy() / tau.p4().pt(), tauInputs_indices_[dnn::tau_E_over_pt]);
1290  get(dnn::tau_charge) = sp.scale(tau.charge(), tauInputs_indices_[dnn::tau_charge]);
1291  get(dnn::tau_n_charged_prongs) = sp.scale(tau.decayMode() / 5 + 1, tauInputs_indices_[dnn::tau_n_charged_prongs]);
1292  get(dnn::tau_n_neutral_prongs) = sp.scale(tau.decayMode() % 5, tauInputs_indices_[dnn::tau_n_neutral_prongs]);
1293  get(dnn::chargedIsoPtSum) =
1294  sp.scale(tau_funcs.getChargedIsoPtSum(tau, tau_ref), tauInputs_indices_[dnn::chargedIsoPtSum]);
1295  get(dnn::chargedIsoPtSumdR03_over_dR05) =
1296  sp.scale(tau_funcs.getChargedIsoPtSumdR03(tau, tau_ref) / tau_funcs.getChargedIsoPtSum(tau, tau_ref),
1297  tauInputs_indices_[dnn::chargedIsoPtSumdR03_over_dR05]);
1298  if (sub_version_ == 1)
1299  get(dnn::footprintCorrection) =
1300  sp.scale(tau_funcs.getFootprintCorrectiondR03(tau, tau_ref), tauInputs_indices_[dnn::footprintCorrection]);
1301  else if (sub_version_ == 5) {
1302  if (is_online_)
1303  get(dnn::footprintCorrection) =
1304  sp.scale(tau_funcs.getFootprintCorrectiondR03(tau, tau_ref), tauInputs_indices_[dnn::footprintCorrection]);
1305  else
1306  get(dnn::footprintCorrection) =
1307  sp.scale(tau_funcs.getFootprintCorrection(tau, tau_ref), tauInputs_indices_[dnn::footprintCorrection]);
1308  }
1309  get(dnn::neutralIsoPtSum) =
1310  sp.scale(tau_funcs.getNeutralIsoPtSum(tau, tau_ref), tauInputs_indices_[dnn::neutralIsoPtSum]);
1311  get(dnn::neutralIsoPtSumWeight_over_neutralIsoPtSum) =
1312  sp.scale(tau_funcs.getNeutralIsoPtSumWeight(tau, tau_ref) / tau_funcs.getNeutralIsoPtSum(tau, tau_ref),
1313  tauInputs_indices_[dnn::neutralIsoPtSumWeight_over_neutralIsoPtSum]);
1314  get(dnn::neutralIsoPtSumWeightdR03_over_neutralIsoPtSum) =
1315  sp.scale(tau_funcs.getNeutralIsoPtSumdR03Weight(tau, tau_ref) / tau_funcs.getNeutralIsoPtSum(tau, tau_ref),
1316  tauInputs_indices_[dnn::neutralIsoPtSumWeightdR03_over_neutralIsoPtSum]);
1317  get(dnn::neutralIsoPtSumdR03_over_dR05) =
1318  sp.scale(tau_funcs.getNeutralIsoPtSumdR03(tau, tau_ref) / tau_funcs.getNeutralIsoPtSum(tau, tau_ref),
1319  tauInputs_indices_[dnn::neutralIsoPtSumdR03_over_dR05]);
1320  get(dnn::photonPtSumOutsideSignalCone) = sp.scale(tau_funcs.getPhotonPtSumOutsideSignalCone(tau, tau_ref),
1321  tauInputs_indices_[dnn::photonPtSumOutsideSignalCone]);
1322  get(dnn::puCorrPtSum) = sp.scale(tau_funcs.getPuCorrPtSum(tau, tau_ref), tauInputs_indices_[dnn::puCorrPtSum]);
1323  // The global PCA coordinates were used as inputs during the NN training, but it was decided to disable
1324  // them for the inference, because modeling of dxy_PCA in MC poorly describes the data, and x and y coordinates
1325  // in data results outside of the expected 5 std. dev. input validity range. On the other hand,
1326  // these coordinates are strongly era-dependent. Kept as comment to document what NN expects.
1327  if (sub_version_ == 1) {
1328  if (!disable_dxy_pca_) {
1329  auto const pca = tau_funcs.getdxyPCA(tau, tau_index);
1330  get(dnn::tau_dxy_pca_x) = sp.scale(pca.x(), tauInputs_indices_[dnn::tau_dxy_pca_x]);
1331  get(dnn::tau_dxy_pca_y) = sp.scale(pca.y(), tauInputs_indices_[dnn::tau_dxy_pca_y]);
1332  get(dnn::tau_dxy_pca_z) = sp.scale(pca.z(), tauInputs_indices_[dnn::tau_dxy_pca_z]);
1333  } else {
1334  get(dnn::tau_dxy_pca_x) = 0;
1335  get(dnn::tau_dxy_pca_y) = 0;
1336  get(dnn::tau_dxy_pca_z) = 0;
1337  }
1338  }
1339 
1340  const bool tau_dxy_valid =
1341  isAbove(tau_funcs.getdxy(tau, tau_index), -10) && isAbove(tau_funcs.getdxyError(tau, tau_index), 0);
1342  if (tau_dxy_valid) {
1343  get(dnn::tau_dxy_valid) = sp.scale(tau_dxy_valid, tauInputs_indices_[dnn::tau_dxy_valid]);
1344  get(dnn::tau_dxy) = sp.scale(tau_funcs.getdxy(tau, tau_index), tauInputs_indices_[dnn::tau_dxy]);
1345  get(dnn::tau_dxy_sig) =
1346  sp.scale(std::abs(tau_funcs.getdxy(tau, tau_index)) / tau_funcs.getdxyError(tau, tau_index),
1347  tauInputs_indices_[dnn::tau_dxy_sig]);
1348  }
1349  const bool tau_ip3d_valid =
1350  isAbove(tau_funcs.getip3d(tau, tau_index), -10) && isAbove(tau_funcs.getip3dError(tau, tau_index), 0);
1351  if (tau_ip3d_valid) {
1352  get(dnn::tau_ip3d_valid) = sp.scale(tau_ip3d_valid, tauInputs_indices_[dnn::tau_ip3d_valid]);
1353  get(dnn::tau_ip3d) = sp.scale(tau_funcs.getip3d(tau, tau_index), tauInputs_indices_[dnn::tau_ip3d]);
1354  get(dnn::tau_ip3d_sig) =
1355  sp.scale(std::abs(tau_funcs.getip3d(tau, tau_index)) / tau_funcs.getip3dError(tau, tau_index),
1356  tauInputs_indices_[dnn::tau_ip3d_sig]);
1357  }
1358  if (leadChargedHadrCand) {
1359  const bool hasTrackDetails = candFunc::getHasTrackDetails(*leadChargedHadrCand);
1360  const float tau_dz = (is_online_ && !hasTrackDetails) ? 0 : candFunc::getTauDz(*leadChargedHadrCand);
1361  get(dnn::tau_dz) = sp.scale(tau_dz, tauInputs_indices_[dnn::tau_dz]);
1362  get(dnn::tau_dz_sig_valid) =
1363  sp.scale(candFunc::getTauDZSigValid(*leadChargedHadrCand), tauInputs_indices_[dnn::tau_dz_sig_valid]);
1364  const double dzError = hasTrackDetails ? leadChargedHadrCand->dzError() : -999.;
1365  get(dnn::tau_dz_sig) = sp.scale(std::abs(tau_dz) / dzError, tauInputs_indices_[dnn::tau_dz_sig]);
1366  }
1367  get(dnn::tau_flightLength_x) =
1368  sp.scale(tau_funcs.getFlightLength(tau, tau_index).x(), tauInputs_indices_[dnn::tau_flightLength_x]);
1369  get(dnn::tau_flightLength_y) =
1370  sp.scale(tau_funcs.getFlightLength(tau, tau_index).y(), tauInputs_indices_[dnn::tau_flightLength_y]);
1371  get(dnn::tau_flightLength_z) =
1372  sp.scale(tau_funcs.getFlightLength(tau, tau_index).z(), tauInputs_indices_[dnn::tau_flightLength_z]);
1373  if (sub_version_ == 1)
1374  get(dnn::tau_flightLength_sig) = 0.55756444; //This value is set due to a bug in the training
1375  else if (sub_version_ == 5)
1376  get(dnn::tau_flightLength_sig) =
1377  sp.scale(tau_funcs.getFlightLengthSig(tau, tau_index), tauInputs_indices_[dnn::tau_flightLength_sig]);
1378 
1379  get(dnn::tau_pt_weighted_deta_strip) = sp.scale(reco::tau::pt_weighted_deta_strip(tau, tau.decayMode()),
1380  tauInputs_indices_[dnn::tau_pt_weighted_deta_strip]);
1381 
1382  get(dnn::tau_pt_weighted_dphi_strip) = sp.scale(reco::tau::pt_weighted_dphi_strip(tau, tau.decayMode()),
1383  tauInputs_indices_[dnn::tau_pt_weighted_dphi_strip]);
1384  get(dnn::tau_pt_weighted_dr_signal) = sp.scale(reco::tau::pt_weighted_dr_signal(tau, tau.decayMode()),
1385  tauInputs_indices_[dnn::tau_pt_weighted_dr_signal]);
1386  get(dnn::tau_pt_weighted_dr_iso) =
1387  sp.scale(reco::tau::pt_weighted_dr_iso(tau, tau.decayMode()), tauInputs_indices_[dnn::tau_pt_weighted_dr_iso]);
1388  get(dnn::tau_leadingTrackNormChi2) =
1389  sp.scale(tau_funcs.getLeadingTrackNormChi2(tau), tauInputs_indices_[dnn::tau_leadingTrackNormChi2]);
1390  const auto eratio = reco::tau::eratio(tau);
1391  const bool tau_e_ratio_valid = std::isnormal(eratio) && eratio > 0.f;
1392  get(dnn::tau_e_ratio_valid) = sp.scale(tau_e_ratio_valid, tauInputs_indices_[dnn::tau_e_ratio_valid]);
1393  get(dnn::tau_e_ratio) = tau_e_ratio_valid ? sp.scale(eratio, tauInputs_indices_[dnn::tau_e_ratio]) : 0.f;
1394  const double gj_angle_diff = calculateGottfriedJacksonAngleDifference(tau, tau_index, tau_funcs);
1395  const bool tau_gj_angle_diff_valid = (std::isnormal(gj_angle_diff) || gj_angle_diff == 0) && gj_angle_diff >= 0;
1396  get(dnn::tau_gj_angle_diff_valid) =
1397  sp.scale(tau_gj_angle_diff_valid, tauInputs_indices_[dnn::tau_gj_angle_diff_valid]);
1398  get(dnn::tau_gj_angle_diff) =
1399  tau_gj_angle_diff_valid ? sp.scale(gj_angle_diff, tauInputs_indices_[dnn::tau_gj_angle_diff]) : 0;
1400  get(dnn::tau_n_photons) = sp.scale(reco::tau::n_photons_total(tau), tauInputs_indices_[dnn::tau_n_photons]);
1401  get(dnn::tau_emFraction) = sp.scale(tau_funcs.getEmFraction(tau), tauInputs_indices_[dnn::tau_emFraction]);
1402 
1403  get(dnn::tau_inside_ecal_crack) =
1404  sp.scale(isInEcalCrack(tau.p4().eta()), tauInputs_indices_[dnn::tau_inside_ecal_crack]);
1405  get(dnn::leadChargedCand_etaAtEcalEntrance_minus_tau_eta) =
1406  sp.scale(tau_funcs.getEtaAtEcalEntrance(tau) - tau.p4().eta(),
1407  tauInputs_indices_[dnn::leadChargedCand_etaAtEcalEntrance_minus_tau_eta]);
1408  }
1409 
1410  template <typename CandidateCastType, typename TauCastType, typename EgammaBlockType>
1412  const TauCastType& tau,
1413  const size_t tau_index,
1414  const edm::RefToBase<reco::BaseTau> tau_ref,
1415  const reco::Vertex& pv,
1416  double rho,
1417  const std::vector<pat::Electron>* electrons,
1418  const edm::View<reco::Candidate>& pfCands,
1419  const Cell& cell_map,
1420  TauFunc tau_funcs,
1421  bool is_inner,
1422  EgammaBlockType& egammaBlockInputs) {
1423  namespace dnn = dnn_inputs_v2::EgammaBlockInputs;
1424  namespace sc = deep_tau::Scaling;
1425  namespace candFunc = candFunc;
1426  sc::FeatureT ft_global = sc::FeatureT::GridGlobal;
1427  sc::FeatureT ft_PFe = sc::FeatureT::PfCand_electron;
1428  sc::FeatureT ft_PFg = sc::FeatureT::PfCand_gamma;
1430 
1431  // needed to remap indices from scaling vectors to those from dnn_inputs_v2::EgammaBlockInputs
1432  int PFe_index_offset = scalingParamsMap_->at(std::make_pair(ft_global, false)).mean_.size();
1433  int e_index_offset = PFe_index_offset + scalingParamsMap_->at(std::make_pair(ft_PFe, false)).mean_.size();
1434  int PFg_index_offset = e_index_offset + scalingParamsMap_->at(std::make_pair(ft_e, false)).mean_.size();
1435 
1436  // to account for swapped order of PfCand_gamma and Electron blocks for v2p5 training w.r.t. v2p1
1437  int fill_index_offset_e = 0;
1438  int fill_index_offset_PFg = 0;
1439  if (sub_version_ == 5) {
1440  fill_index_offset_e =
1441  scalingParamsMap_->at(std::make_pair(ft_PFg, false)).mean_.size(); // size of PF gamma features
1442  fill_index_offset_PFg =
1443  -scalingParamsMap_->at(std::make_pair(ft_e, false)).mean_.size(); // size of Electron features
1444  }
1445 
1446  const auto& get = [&](int var_index) -> float& {
1447  if constexpr (std::is_same_v<EgammaBlockType, std::vector<float>::iterator>) {
1448  return *(egammaBlockInputs + var_index);
1449  } else {
1450  return ((tensorflow::Tensor)egammaBlockInputs).tensor<float, 4>()(idx, 0, 0, var_index);
1451  }
1452  };
1453 
1454  const bool valid_index_pf_ele = cell_map.count(CellObjectType::PfCand_electron);
1455  const bool valid_index_pf_gamma = cell_map.count(CellObjectType::PfCand_gamma);
1456  const bool valid_index_ele = cell_map.count(CellObjectType::Electron);
1457 
1458  if (!cell_map.empty()) {
1459  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_global, false));
1460  get(dnn::rho) = sp.scale(rho, dnn::rho);
1461  get(dnn::tau_pt) = sp.scale(tau.polarP4().pt(), dnn::tau_pt);
1462  get(dnn::tau_eta) = sp.scale(tau.polarP4().eta(), dnn::tau_eta);
1463  get(dnn::tau_inside_ecal_crack) = sp.scale(isInEcalCrack(tau.polarP4().eta()), dnn::tau_inside_ecal_crack);
1464  }
1465  if (valid_index_pf_ele) {
1466  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_PFe, is_inner));
1467  size_t index_pf_ele = cell_map.at(CellObjectType::PfCand_electron);
1468  const auto& ele_cand = dynamic_cast<const CandidateCastType&>(pfCands.at(index_pf_ele));
1469 
1470  get(dnn::pfCand_ele_valid) = sp.scale(valid_index_pf_ele, dnn::pfCand_ele_valid - PFe_index_offset);
1471  get(dnn::pfCand_ele_rel_pt) =
1472  sp.scale(ele_cand.polarP4().pt() / tau.polarP4().pt(), dnn::pfCand_ele_rel_pt - PFe_index_offset);
1473  get(dnn::pfCand_ele_deta) =
1474  sp.scale(ele_cand.polarP4().eta() - tau.polarP4().eta(), dnn::pfCand_ele_deta - PFe_index_offset);
1475  get(dnn::pfCand_ele_dphi) =
1476  sp.scale(dPhi(tau.polarP4(), ele_cand.polarP4()), dnn::pfCand_ele_dphi - PFe_index_offset);
1477  get(dnn::pfCand_ele_pvAssociationQuality) = sp.scale<int>(
1478  candFunc::getPvAssocationQuality(ele_cand), dnn::pfCand_ele_pvAssociationQuality - PFe_index_offset);
1479  get(dnn::pfCand_ele_puppiWeight) = is_inner ? sp.scale(candFunc::getPuppiWeight(ele_cand, 0.9906834f),
1480  dnn::pfCand_ele_puppiWeight - PFe_index_offset)
1481  : sp.scale(candFunc::getPuppiWeight(ele_cand, 0.9669586f),
1482  dnn::pfCand_ele_puppiWeight - PFe_index_offset);
1483  get(dnn::pfCand_ele_charge) = sp.scale(ele_cand.charge(), dnn::pfCand_ele_charge - PFe_index_offset);
1484  get(dnn::pfCand_ele_lostInnerHits) =
1485  sp.scale<int>(candFunc::getLostInnerHits(ele_cand, 0), dnn::pfCand_ele_lostInnerHits - PFe_index_offset);
1486  get(dnn::pfCand_ele_numberOfPixelHits) =
1487  sp.scale(candFunc::getNumberOfPixelHits(ele_cand, 0), dnn::pfCand_ele_numberOfPixelHits - PFe_index_offset);
1488  get(dnn::pfCand_ele_vertex_dx) =
1489  sp.scale(ele_cand.vertex().x() - pv.position().x(), dnn::pfCand_ele_vertex_dx - PFe_index_offset);
1490  get(dnn::pfCand_ele_vertex_dy) =
1491  sp.scale(ele_cand.vertex().y() - pv.position().y(), dnn::pfCand_ele_vertex_dy - PFe_index_offset);
1492  get(dnn::pfCand_ele_vertex_dz) =
1493  sp.scale(ele_cand.vertex().z() - pv.position().z(), dnn::pfCand_ele_vertex_dz - PFe_index_offset);
1494  get(dnn::pfCand_ele_vertex_dx_tauFL) =
1495  sp.scale(ele_cand.vertex().x() - pv.position().x() - tau_funcs.getFlightLength(tau, tau_index).x(),
1496  dnn::pfCand_ele_vertex_dx_tauFL - PFe_index_offset);
1497  get(dnn::pfCand_ele_vertex_dy_tauFL) =
1498  sp.scale(ele_cand.vertex().y() - pv.position().y() - tau_funcs.getFlightLength(tau, tau_index).y(),
1499  dnn::pfCand_ele_vertex_dy_tauFL - PFe_index_offset);
1500  get(dnn::pfCand_ele_vertex_dz_tauFL) =
1501  sp.scale(ele_cand.vertex().z() - pv.position().z() - tau_funcs.getFlightLength(tau, tau_index).z(),
1502  dnn::pfCand_ele_vertex_dz_tauFL - PFe_index_offset);
1503 
1504  const bool hasTrackDetails = candFunc::getHasTrackDetails(ele_cand);
1505  if (hasTrackDetails) {
1506  get(dnn::pfCand_ele_hasTrackDetails) =
1507  sp.scale(hasTrackDetails, dnn::pfCand_ele_hasTrackDetails - PFe_index_offset);
1508  get(dnn::pfCand_ele_dxy) = sp.scale(candFunc::getTauDxy(ele_cand), dnn::pfCand_ele_dxy - PFe_index_offset);
1509  get(dnn::pfCand_ele_dxy_sig) = sp.scale(std::abs(candFunc::getTauDxy(ele_cand)) / ele_cand.dxyError(),
1510  dnn::pfCand_ele_dxy_sig - PFe_index_offset);
1511  get(dnn::pfCand_ele_dz) = sp.scale(candFunc::getTauDz(ele_cand), dnn::pfCand_ele_dz - PFe_index_offset);
1512  get(dnn::pfCand_ele_dz_sig) = sp.scale(std::abs(candFunc::getTauDz(ele_cand)) / ele_cand.dzError(),
1513  dnn::pfCand_ele_dz_sig - PFe_index_offset);
1514  get(dnn::pfCand_ele_track_chi2_ndof) =
1515  candFunc::getPseudoTrack(ele_cand).ndof() > 0
1516  ? sp.scale(candFunc::getPseudoTrack(ele_cand).chi2() / candFunc::getPseudoTrack(ele_cand).ndof(),
1517  dnn::pfCand_ele_track_chi2_ndof - PFe_index_offset)
1518  : 0;
1519  get(dnn::pfCand_ele_track_ndof) =
1520  candFunc::getPseudoTrack(ele_cand).ndof() > 0
1521  ? sp.scale(candFunc::getPseudoTrack(ele_cand).ndof(), dnn::pfCand_ele_track_ndof - PFe_index_offset)
1522  : 0;
1523  }
1524  }
1525  if (valid_index_pf_gamma) {
1526  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_PFg, is_inner));
1527  size_t index_pf_gamma = cell_map.at(CellObjectType::PfCand_gamma);
1528  const auto& gamma_cand = dynamic_cast<const CandidateCastType&>(pfCands.at(index_pf_gamma));
1529 
1530  get(dnn::pfCand_gamma_valid + fill_index_offset_PFg) =
1531  sp.scale(valid_index_pf_gamma, dnn::pfCand_gamma_valid - PFg_index_offset);
1532  get(dnn::pfCand_gamma_rel_pt + fill_index_offset_PFg) =
1533  sp.scale(gamma_cand.polarP4().pt() / tau.polarP4().pt(), dnn::pfCand_gamma_rel_pt - PFg_index_offset);
1534  get(dnn::pfCand_gamma_deta + fill_index_offset_PFg) =
1535  sp.scale(gamma_cand.polarP4().eta() - tau.polarP4().eta(), dnn::pfCand_gamma_deta - PFg_index_offset);
1536  get(dnn::pfCand_gamma_dphi + fill_index_offset_PFg) =
1537  sp.scale(dPhi(tau.polarP4(), gamma_cand.polarP4()), dnn::pfCand_gamma_dphi - PFg_index_offset);
1538  get(dnn::pfCand_gamma_pvAssociationQuality + fill_index_offset_PFg) = sp.scale<int>(
1539  candFunc::getPvAssocationQuality(gamma_cand), dnn::pfCand_gamma_pvAssociationQuality - PFg_index_offset);
1540  get(dnn::pfCand_gamma_fromPV + fill_index_offset_PFg) =
1541  sp.scale<int>(candFunc::getFromPV(gamma_cand), dnn::pfCand_gamma_fromPV - PFg_index_offset);
1542  get(dnn::pfCand_gamma_puppiWeight + fill_index_offset_PFg) =
1543  is_inner ? sp.scale(candFunc::getPuppiWeight(gamma_cand, 0.9084110f),
1544  dnn::pfCand_gamma_puppiWeight - PFg_index_offset)
1545  : sp.scale(candFunc::getPuppiWeight(gamma_cand, 0.4211567f),
1546  dnn::pfCand_gamma_puppiWeight - PFg_index_offset);
1547  get(dnn::pfCand_gamma_puppiWeightNoLep + fill_index_offset_PFg) =
1548  is_inner ? sp.scale(candFunc::getPuppiWeightNoLep(gamma_cand, 0.8857716f),
1549  dnn::pfCand_gamma_puppiWeightNoLep - PFg_index_offset)
1550  : sp.scale(candFunc::getPuppiWeightNoLep(gamma_cand, 0.3822604f),
1551  dnn::pfCand_gamma_puppiWeightNoLep - PFg_index_offset);
1552  get(dnn::pfCand_gamma_lostInnerHits + fill_index_offset_PFg) =
1553  sp.scale<int>(candFunc::getLostInnerHits(gamma_cand, 0), dnn::pfCand_gamma_lostInnerHits - PFg_index_offset);
1554  get(dnn::pfCand_gamma_numberOfPixelHits + fill_index_offset_PFg) = sp.scale(
1555  candFunc::getNumberOfPixelHits(gamma_cand, 0), dnn::pfCand_gamma_numberOfPixelHits - PFg_index_offset);
1556  get(dnn::pfCand_gamma_vertex_dx + fill_index_offset_PFg) =
1557  sp.scale(gamma_cand.vertex().x() - pv.position().x(), dnn::pfCand_gamma_vertex_dx - PFg_index_offset);
1558  get(dnn::pfCand_gamma_vertex_dy + fill_index_offset_PFg) =
1559  sp.scale(gamma_cand.vertex().y() - pv.position().y(), dnn::pfCand_gamma_vertex_dy - PFg_index_offset);
1560  get(dnn::pfCand_gamma_vertex_dz + fill_index_offset_PFg) =
1561  sp.scale(gamma_cand.vertex().z() - pv.position().z(), dnn::pfCand_gamma_vertex_dz - PFg_index_offset);
1562  get(dnn::pfCand_gamma_vertex_dx_tauFL + fill_index_offset_PFg) =
1563  sp.scale(gamma_cand.vertex().x() - pv.position().x() - tau_funcs.getFlightLength(tau, tau_index).x(),
1564  dnn::pfCand_gamma_vertex_dx_tauFL - PFg_index_offset);
1565  get(dnn::pfCand_gamma_vertex_dy_tauFL + fill_index_offset_PFg) =
1566  sp.scale(gamma_cand.vertex().y() - pv.position().y() - tau_funcs.getFlightLength(tau, tau_index).y(),
1567  dnn::pfCand_gamma_vertex_dy_tauFL - PFg_index_offset);
1568  get(dnn::pfCand_gamma_vertex_dz_tauFL + fill_index_offset_PFg) =
1569  sp.scale(gamma_cand.vertex().z() - pv.position().z() - tau_funcs.getFlightLength(tau, tau_index).z(),
1570  dnn::pfCand_gamma_vertex_dz_tauFL - PFg_index_offset);
1571  const bool hasTrackDetails = candFunc::getHasTrackDetails(gamma_cand);
1572  if (hasTrackDetails) {
1573  get(dnn::pfCand_gamma_hasTrackDetails + fill_index_offset_PFg) =
1574  sp.scale(hasTrackDetails, dnn::pfCand_gamma_hasTrackDetails - PFg_index_offset);
1575  get(dnn::pfCand_gamma_dxy + fill_index_offset_PFg) =
1576  sp.scale(candFunc::getTauDxy(gamma_cand), dnn::pfCand_gamma_dxy - PFg_index_offset);
1577  get(dnn::pfCand_gamma_dxy_sig + fill_index_offset_PFg) =
1578  sp.scale(std::abs(candFunc::getTauDxy(gamma_cand)) / gamma_cand.dxyError(),
1579  dnn::pfCand_gamma_dxy_sig - PFg_index_offset);
1580  get(dnn::pfCand_gamma_dz + fill_index_offset_PFg) =
1581  sp.scale(candFunc::getTauDz(gamma_cand), dnn::pfCand_gamma_dz - PFg_index_offset);
1582  get(dnn::pfCand_gamma_dz_sig + fill_index_offset_PFg) =
1583  sp.scale(std::abs(candFunc::getTauDz(gamma_cand)) / gamma_cand.dzError(),
1584  dnn::pfCand_gamma_dz_sig - PFg_index_offset);
1585  get(dnn::pfCand_gamma_track_chi2_ndof + fill_index_offset_PFg) =
1586  candFunc::getPseudoTrack(gamma_cand).ndof() > 0
1587  ? sp.scale(candFunc::getPseudoTrack(gamma_cand).chi2() / candFunc::getPseudoTrack(gamma_cand).ndof(),
1588  dnn::pfCand_gamma_track_chi2_ndof - PFg_index_offset)
1589  : 0;
1590  get(dnn::pfCand_gamma_track_ndof + fill_index_offset_PFg) =
1591  candFunc::getPseudoTrack(gamma_cand).ndof() > 0
1592  ? sp.scale(candFunc::getPseudoTrack(gamma_cand).ndof(), dnn::pfCand_gamma_track_ndof - PFg_index_offset)
1593  : 0;
1594  }
1595  }
1596  if (valid_index_ele) {
1597  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_e, is_inner));
1598  size_t index_ele = cell_map.at(CellObjectType::Electron);
1599  const auto& ele = electrons->at(index_ele);
1600 
1601  get(dnn::ele_valid + fill_index_offset_e) = sp.scale(valid_index_ele, dnn::ele_valid - e_index_offset);
1602  get(dnn::ele_rel_pt + fill_index_offset_e) =
1603  sp.scale(ele.polarP4().pt() / tau.polarP4().pt(), dnn::ele_rel_pt - e_index_offset);
1604  get(dnn::ele_deta + fill_index_offset_e) =
1605  sp.scale(ele.polarP4().eta() - tau.polarP4().eta(), dnn::ele_deta - e_index_offset);
1606  get(dnn::ele_dphi + fill_index_offset_e) =
1607  sp.scale(dPhi(tau.polarP4(), ele.polarP4()), dnn::ele_dphi - e_index_offset);
1608 
1609  float cc_ele_energy, cc_gamma_energy;
1610  int cc_n_gamma;
1611  const bool cc_valid = calculateElectronClusterVarsV2(ele, cc_ele_energy, cc_gamma_energy, cc_n_gamma);
1612  if (cc_valid) {
1613  get(dnn::ele_cc_valid + fill_index_offset_e) = sp.scale(cc_valid, dnn::ele_cc_valid - e_index_offset);
1614  get(dnn::ele_cc_ele_rel_energy + fill_index_offset_e) =
1615  sp.scale(cc_ele_energy / ele.polarP4().pt(), dnn::ele_cc_ele_rel_energy - e_index_offset);
1616  get(dnn::ele_cc_gamma_rel_energy + fill_index_offset_e) =
1617  sp.scale(cc_gamma_energy / cc_ele_energy, dnn::ele_cc_gamma_rel_energy - e_index_offset);
1618  get(dnn::ele_cc_n_gamma + fill_index_offset_e) = sp.scale(cc_n_gamma, dnn::ele_cc_n_gamma - e_index_offset);
1619  }
1620  get(dnn::ele_rel_trackMomentumAtVtx + fill_index_offset_e) =
1621  sp.scale(ele.trackMomentumAtVtx().R() / ele.polarP4().pt(), dnn::ele_rel_trackMomentumAtVtx - e_index_offset);
1622  get(dnn::ele_rel_trackMomentumAtCalo + fill_index_offset_e) = sp.scale(
1623  ele.trackMomentumAtCalo().R() / ele.polarP4().pt(), dnn::ele_rel_trackMomentumAtCalo - e_index_offset);
1624  get(dnn::ele_rel_trackMomentumOut + fill_index_offset_e) =
1625  sp.scale(ele.trackMomentumOut().R() / ele.polarP4().pt(), dnn::ele_rel_trackMomentumOut - e_index_offset);
1626  get(dnn::ele_rel_trackMomentumAtEleClus + fill_index_offset_e) = sp.scale(
1627  ele.trackMomentumAtEleClus().R() / ele.polarP4().pt(), dnn::ele_rel_trackMomentumAtEleClus - e_index_offset);
1628  get(dnn::ele_rel_trackMomentumAtVtxWithConstraint + fill_index_offset_e) =
1629  sp.scale(ele.trackMomentumAtVtxWithConstraint().R() / ele.polarP4().pt(),
1630  dnn::ele_rel_trackMomentumAtVtxWithConstraint - e_index_offset);
1631  get(dnn::ele_rel_ecalEnergy + fill_index_offset_e) =
1632  sp.scale(ele.ecalEnergy() / ele.polarP4().pt(), dnn::ele_rel_ecalEnergy - e_index_offset);
1633  get(dnn::ele_ecalEnergy_sig + fill_index_offset_e) =
1634  sp.scale(ele.ecalEnergy() / ele.ecalEnergyError(), dnn::ele_ecalEnergy_sig - e_index_offset);
1635  get(dnn::ele_eSuperClusterOverP + fill_index_offset_e) =
1636  sp.scale(ele.eSuperClusterOverP(), dnn::ele_eSuperClusterOverP - e_index_offset);
1637  get(dnn::ele_eSeedClusterOverP + fill_index_offset_e) =
1638  sp.scale(ele.eSeedClusterOverP(), dnn::ele_eSeedClusterOverP - e_index_offset);
1639  get(dnn::ele_eSeedClusterOverPout + fill_index_offset_e) =
1640  sp.scale(ele.eSeedClusterOverPout(), dnn::ele_eSeedClusterOverPout - e_index_offset);
1641  get(dnn::ele_eEleClusterOverPout + fill_index_offset_e) =
1642  sp.scale(ele.eEleClusterOverPout(), dnn::ele_eEleClusterOverPout - e_index_offset);
1643  get(dnn::ele_deltaEtaSuperClusterTrackAtVtx + fill_index_offset_e) =
1644  sp.scale(ele.deltaEtaSuperClusterTrackAtVtx(), dnn::ele_deltaEtaSuperClusterTrackAtVtx - e_index_offset);
1645  get(dnn::ele_deltaEtaSeedClusterTrackAtCalo + fill_index_offset_e) =
1646  sp.scale(ele.deltaEtaSeedClusterTrackAtCalo(), dnn::ele_deltaEtaSeedClusterTrackAtCalo - e_index_offset);
1647  get(dnn::ele_deltaEtaEleClusterTrackAtCalo + fill_index_offset_e) =
1648  sp.scale(ele.deltaEtaEleClusterTrackAtCalo(), dnn::ele_deltaEtaEleClusterTrackAtCalo - e_index_offset);
1649  get(dnn::ele_deltaPhiEleClusterTrackAtCalo + fill_index_offset_e) =
1650  sp.scale(ele.deltaPhiEleClusterTrackAtCalo(), dnn::ele_deltaPhiEleClusterTrackAtCalo - e_index_offset);
1651  get(dnn::ele_deltaPhiSuperClusterTrackAtVtx + fill_index_offset_e) =
1652  sp.scale(ele.deltaPhiSuperClusterTrackAtVtx(), dnn::ele_deltaPhiSuperClusterTrackAtVtx - e_index_offset);
1653  get(dnn::ele_deltaPhiSeedClusterTrackAtCalo + fill_index_offset_e) =
1654  sp.scale(ele.deltaPhiSeedClusterTrackAtCalo(), dnn::ele_deltaPhiSeedClusterTrackAtCalo - e_index_offset);
1655  const bool mva_valid =
1656  (ele.mvaInput().earlyBrem > -2) ||
1657  (year_ !=
1658  2026); // Known issue that input can be invalid in Phase2 samples (early/lateBrem==-2, hadEnergy==0, sigmaEtaEta/deltaEta==3.40282e+38). Unknown if also in Run2/3, so don't change there
1659  if (mva_valid) {
1660  get(dnn::ele_mvaInput_earlyBrem + fill_index_offset_e) =
1661  sp.scale(ele.mvaInput().earlyBrem, dnn::ele_mvaInput_earlyBrem - e_index_offset);
1662  get(dnn::ele_mvaInput_lateBrem + fill_index_offset_e) =
1663  sp.scale(ele.mvaInput().lateBrem, dnn::ele_mvaInput_lateBrem - e_index_offset);
1664  get(dnn::ele_mvaInput_sigmaEtaEta + fill_index_offset_e) =
1665  sp.scale(ele.mvaInput().sigmaEtaEta, dnn::ele_mvaInput_sigmaEtaEta - e_index_offset);
1666  get(dnn::ele_mvaInput_hadEnergy + fill_index_offset_e) =
1667  sp.scale(ele.mvaInput().hadEnergy, dnn::ele_mvaInput_hadEnergy - e_index_offset);
1668  get(dnn::ele_mvaInput_deltaEta + fill_index_offset_e) =
1669  sp.scale(ele.mvaInput().deltaEta, dnn::ele_mvaInput_deltaEta - e_index_offset);
1670  }
1671  const auto& gsfTrack = ele.gsfTrack();
1672  if (gsfTrack.isNonnull()) {
1673  get(dnn::ele_gsfTrack_normalizedChi2 + fill_index_offset_e) =
1674  sp.scale(gsfTrack->normalizedChi2(), dnn::ele_gsfTrack_normalizedChi2 - e_index_offset);
1675  get(dnn::ele_gsfTrack_numberOfValidHits + fill_index_offset_e) =
1676  sp.scale(gsfTrack->numberOfValidHits(), dnn::ele_gsfTrack_numberOfValidHits - e_index_offset);
1677  get(dnn::ele_rel_gsfTrack_pt + fill_index_offset_e) =
1678  sp.scale(gsfTrack->pt() / ele.polarP4().pt(), dnn::ele_rel_gsfTrack_pt - e_index_offset);
1679  get(dnn::ele_gsfTrack_pt_sig + fill_index_offset_e) =
1680  sp.scale(gsfTrack->pt() / gsfTrack->ptError(), dnn::ele_gsfTrack_pt_sig - e_index_offset);
1681  }
1682  const auto& closestCtfTrack = ele.closestCtfTrackRef();
1683  const bool has_closestCtfTrack = closestCtfTrack.isNonnull();
1684  if (has_closestCtfTrack) {
1685  get(dnn::ele_has_closestCtfTrack + fill_index_offset_e) =
1686  sp.scale(has_closestCtfTrack, dnn::ele_has_closestCtfTrack - e_index_offset);
1687  get(dnn::ele_closestCtfTrack_normalizedChi2 + fill_index_offset_e) =
1688  sp.scale(closestCtfTrack->normalizedChi2(), dnn::ele_closestCtfTrack_normalizedChi2 - e_index_offset);
1689  get(dnn::ele_closestCtfTrack_numberOfValidHits + fill_index_offset_e) =
1690  sp.scale(closestCtfTrack->numberOfValidHits(), dnn::ele_closestCtfTrack_numberOfValidHits - e_index_offset);
1691  }
1692  }
1693  }
1694 
1695  template <typename CandidateCastType, typename TauCastType, typename MuonBlockType>
1697  const TauCastType& tau,
1698  const size_t tau_index,
1699  const edm::RefToBase<reco::BaseTau> tau_ref,
1700  const reco::Vertex& pv,
1701  double rho,
1702  const std::vector<pat::Muon>* muons,
1703  const edm::View<reco::Candidate>& pfCands,
1704  const Cell& cell_map,
1705  TauFunc tau_funcs,
1706  bool is_inner,
1707  MuonBlockType& muonBlockInputs) {
1708  namespace dnn = dnn_inputs_v2::MuonBlockInputs;
1709  namespace sc = deep_tau::Scaling;
1710  namespace candFunc = candFunc;
1711  using MuonHitMatchV2 = MuonHitMatchV2;
1712  sc::FeatureT ft_global = sc::FeatureT::GridGlobal;
1713  sc::FeatureT ft_PFmu = sc::FeatureT::PfCand_muon;
1715 
1716  // needed to remap indices from scaling vectors to those from dnn_inputs_v2::MuonBlockInputs
1717  int PFmu_index_offset = scalingParamsMap_->at(std::make_pair(ft_global, false)).mean_.size();
1718  int mu_index_offset = PFmu_index_offset + scalingParamsMap_->at(std::make_pair(ft_PFmu, false)).mean_.size();
1719 
1720  const auto& get = [&](int var_index) -> float& {
1721  if constexpr (std::is_same_v<MuonBlockType, std::vector<float>::iterator>) {
1722  return *(muonBlockInputs + var_index);
1723  } else {
1724  return ((tensorflow::Tensor)muonBlockInputs).tensor<float, 4>()(idx, 0, 0, var_index);
1725  }
1726  };
1727 
1728  const bool valid_index_pf_muon = cell_map.count(CellObjectType::PfCand_muon);
1729  const bool valid_index_muon = cell_map.count(CellObjectType::Muon);
1730 
1731  if (!cell_map.empty()) {
1732  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_global, false));
1733  get(dnn::rho) = sp.scale(rho, dnn::rho);
1734  get(dnn::tau_pt) = sp.scale(tau.polarP4().pt(), dnn::tau_pt);
1735  get(dnn::tau_eta) = sp.scale(tau.polarP4().eta(), dnn::tau_eta);
1736  get(dnn::tau_inside_ecal_crack) = sp.scale(isInEcalCrack(tau.polarP4().eta()), dnn::tau_inside_ecal_crack);
1737  }
1738  if (valid_index_pf_muon) {
1739  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_PFmu, is_inner));
1740  size_t index_pf_muon = cell_map.at(CellObjectType::PfCand_muon);
1741  const auto& muon_cand = dynamic_cast<const CandidateCastType&>(pfCands.at(index_pf_muon));
1742 
1743  get(dnn::pfCand_muon_valid) = sp.scale(valid_index_pf_muon, dnn::pfCand_muon_valid - PFmu_index_offset);
1744  get(dnn::pfCand_muon_rel_pt) =
1745  sp.scale(muon_cand.polarP4().pt() / tau.polarP4().pt(), dnn::pfCand_muon_rel_pt - PFmu_index_offset);
1746  get(dnn::pfCand_muon_deta) =
1747  sp.scale(muon_cand.polarP4().eta() - tau.polarP4().eta(), dnn::pfCand_muon_deta - PFmu_index_offset);
1748  get(dnn::pfCand_muon_dphi) =
1749  sp.scale(dPhi(tau.polarP4(), muon_cand.polarP4()), dnn::pfCand_muon_dphi - PFmu_index_offset);
1750  get(dnn::pfCand_muon_pvAssociationQuality) = sp.scale<int>(
1751  candFunc::getPvAssocationQuality(muon_cand), dnn::pfCand_muon_pvAssociationQuality - PFmu_index_offset);
1752  get(dnn::pfCand_muon_fromPV) =
1753  sp.scale<int>(candFunc::getFromPV(muon_cand), dnn::pfCand_muon_fromPV - PFmu_index_offset);
1754  get(dnn::pfCand_muon_puppiWeight) = is_inner ? sp.scale(candFunc::getPuppiWeight(muon_cand, 0.9786588f),
1755  dnn::pfCand_muon_puppiWeight - PFmu_index_offset)
1756  : sp.scale(candFunc::getPuppiWeight(muon_cand, 0.8132477f),
1757  dnn::pfCand_muon_puppiWeight - PFmu_index_offset);
1758  get(dnn::pfCand_muon_charge) = sp.scale(muon_cand.charge(), dnn::pfCand_muon_charge - PFmu_index_offset);
1759  get(dnn::pfCand_muon_lostInnerHits) =
1760  sp.scale<int>(candFunc::getLostInnerHits(muon_cand, 0), dnn::pfCand_muon_lostInnerHits - PFmu_index_offset);
1761  get(dnn::pfCand_muon_numberOfPixelHits) = sp.scale(candFunc::getNumberOfPixelHits(muon_cand, 0),
1762  dnn::pfCand_muon_numberOfPixelHits - PFmu_index_offset);
1763  get(dnn::pfCand_muon_vertex_dx) =
1764  sp.scale(muon_cand.vertex().x() - pv.position().x(), dnn::pfCand_muon_vertex_dx - PFmu_index_offset);
1765  get(dnn::pfCand_muon_vertex_dy) =
1766  sp.scale(muon_cand.vertex().y() - pv.position().y(), dnn::pfCand_muon_vertex_dy - PFmu_index_offset);
1767  get(dnn::pfCand_muon_vertex_dz) =
1768  sp.scale(muon_cand.vertex().z() - pv.position().z(), dnn::pfCand_muon_vertex_dz - PFmu_index_offset);
1769  get(dnn::pfCand_muon_vertex_dx_tauFL) =
1770  sp.scale(muon_cand.vertex().x() - pv.position().x() - tau_funcs.getFlightLength(tau, tau_index).x(),
1771  dnn::pfCand_muon_vertex_dx_tauFL - PFmu_index_offset);
1772  get(dnn::pfCand_muon_vertex_dy_tauFL) =
1773  sp.scale(muon_cand.vertex().y() - pv.position().y() - tau_funcs.getFlightLength(tau, tau_index).y(),
1774  dnn::pfCand_muon_vertex_dy_tauFL - PFmu_index_offset);
1775  get(dnn::pfCand_muon_vertex_dz_tauFL) =
1776  sp.scale(muon_cand.vertex().z() - pv.position().z() - tau_funcs.getFlightLength(tau, tau_index).z(),
1777  dnn::pfCand_muon_vertex_dz_tauFL - PFmu_index_offset);
1778 
1779  const bool hasTrackDetails = candFunc::getHasTrackDetails(muon_cand);
1780  if (hasTrackDetails) {
1781  get(dnn::pfCand_muon_hasTrackDetails) =
1782  sp.scale(hasTrackDetails, dnn::pfCand_muon_hasTrackDetails - PFmu_index_offset);
1783  get(dnn::pfCand_muon_dxy) = sp.scale(candFunc::getTauDxy(muon_cand), dnn::pfCand_muon_dxy - PFmu_index_offset);
1784  get(dnn::pfCand_muon_dxy_sig) = sp.scale(std::abs(candFunc::getTauDxy(muon_cand)) / muon_cand.dxyError(),
1785  dnn::pfCand_muon_dxy_sig - PFmu_index_offset);
1786  get(dnn::pfCand_muon_dz) = sp.scale(candFunc::getTauDz(muon_cand), dnn::pfCand_muon_dz - PFmu_index_offset);
1787  get(dnn::pfCand_muon_dz_sig) = sp.scale(std::abs(candFunc::getTauDz(muon_cand)) / muon_cand.dzError(),
1788  dnn::pfCand_muon_dz_sig - PFmu_index_offset);
1789  get(dnn::pfCand_muon_track_chi2_ndof) =
1790  candFunc::getPseudoTrack(muon_cand).ndof() > 0
1791  ? sp.scale(candFunc::getPseudoTrack(muon_cand).chi2() / candFunc::getPseudoTrack(muon_cand).ndof(),
1792  dnn::pfCand_muon_track_chi2_ndof - PFmu_index_offset)
1793  : 0;
1794  get(dnn::pfCand_muon_track_ndof) =
1795  candFunc::getPseudoTrack(muon_cand).ndof() > 0
1796  ? sp.scale(candFunc::getPseudoTrack(muon_cand).ndof(), dnn::pfCand_muon_track_ndof - PFmu_index_offset)
1797  : 0;
1798  }
1799  }
1800  if (valid_index_muon) {
1801  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_mu, is_inner));
1802  size_t index_muon = cell_map.at(CellObjectType::Muon);
1803  const auto& muon = muons->at(index_muon);
1804 
1805  get(dnn::muon_valid) = sp.scale(valid_index_muon, dnn::muon_valid - mu_index_offset);
1806  get(dnn::muon_rel_pt) = sp.scale(muon.polarP4().pt() / tau.polarP4().pt(), dnn::muon_rel_pt - mu_index_offset);
1807  get(dnn::muon_deta) = sp.scale(muon.polarP4().eta() - tau.polarP4().eta(), dnn::muon_deta - mu_index_offset);
1808  get(dnn::muon_dphi) = sp.scale(dPhi(tau.polarP4(), muon.polarP4()), dnn::muon_dphi - mu_index_offset);
1809  get(dnn::muon_dxy) = sp.scale(muon.dB(pat::Muon::PV2D), dnn::muon_dxy - mu_index_offset);
1810  get(dnn::muon_dxy_sig) =
1811  sp.scale(std::abs(muon.dB(pat::Muon::PV2D)) / muon.edB(pat::Muon::PV2D), dnn::muon_dxy_sig - mu_index_offset);
1812 
1813  const bool normalizedChi2_valid = muon.globalTrack().isNonnull() && muon.normChi2() >= 0;
1814  if (normalizedChi2_valid) {
1815  get(dnn::muon_normalizedChi2_valid) =
1816  sp.scale(normalizedChi2_valid, dnn::muon_normalizedChi2_valid - mu_index_offset);
1817  get(dnn::muon_normalizedChi2) = sp.scale(muon.normChi2(), dnn::muon_normalizedChi2 - mu_index_offset);
1818  if (muon.innerTrack().isNonnull())
1819  get(dnn::muon_numberOfValidHits) =
1820  sp.scale(muon.numberOfValidHits(), dnn::muon_numberOfValidHits - mu_index_offset);
1821  }
1822  get(dnn::muon_segmentCompatibility) =
1823  sp.scale(muon.segmentCompatibility(), dnn::muon_segmentCompatibility - mu_index_offset);
1824  get(dnn::muon_caloCompatibility) =
1825  sp.scale(muon.caloCompatibility(), dnn::muon_caloCompatibility - mu_index_offset);
1826 
1827  const bool pfEcalEnergy_valid = muon.pfEcalEnergy() >= 0;
1828  if (pfEcalEnergy_valid) {
1829  get(dnn::muon_pfEcalEnergy_valid) =
1830  sp.scale(pfEcalEnergy_valid, dnn::muon_pfEcalEnergy_valid - mu_index_offset);
1831  get(dnn::muon_rel_pfEcalEnergy) =
1832  sp.scale(muon.pfEcalEnergy() / muon.polarP4().pt(), dnn::muon_rel_pfEcalEnergy - mu_index_offset);
1833  }
1834 
1835  MuonHitMatchV2 hit_match(muon);
1836  static const std::map<int, std::pair<int, int>> muonMatchHitVars = {
1837  {MuonSubdetId::DT, {dnn::muon_n_matches_DT_1, dnn::muon_n_hits_DT_1}},
1838  {MuonSubdetId::CSC, {dnn::muon_n_matches_CSC_1, dnn::muon_n_hits_CSC_1}},
1839  {MuonSubdetId::RPC, {dnn::muon_n_matches_RPC_1, dnn::muon_n_hits_RPC_1}}};
1840 
1841  for (int subdet : hit_match.MuonHitMatchV2::consideredSubdets()) {
1842  const auto& matchHitVar = muonMatchHitVars.at(subdet);
1843  for (int station = MuonHitMatchV2::first_station_id; station <= MuonHitMatchV2::last_station_id; ++station) {
1844  const unsigned n_matches = hit_match.nMatches(subdet, station);
1845  const unsigned n_hits = hit_match.nHits(subdet, station);
1846  get(matchHitVar.first + station - 1) = sp.scale(n_matches, matchHitVar.first + station - 1 - mu_index_offset);
1847  get(matchHitVar.second + station - 1) = sp.scale(n_hits, matchHitVar.second + station - 1 - mu_index_offset);
1848  }
1849  }
1850  }
1851  }
1852 
1853  template <typename CandidateCastType, typename TauCastType, typename HadronBlockType>
1855  const TauCastType& tau,
1856  const size_t tau_index,
1857  const edm::RefToBase<reco::BaseTau> tau_ref,
1858  const reco::Vertex& pv,
1859  double rho,
1860  const edm::View<reco::Candidate>& pfCands,
1861  const Cell& cell_map,
1862  TauFunc tau_funcs,
1863  bool is_inner,
1864  HadronBlockType& hadronBlockInputs) {
1865  namespace dnn = dnn_inputs_v2::HadronBlockInputs;
1866  namespace sc = deep_tau::Scaling;
1867  namespace candFunc = candFunc;
1868  sc::FeatureT ft_global = sc::FeatureT::GridGlobal;
1869  sc::FeatureT ft_PFchH = sc::FeatureT::PfCand_chHad;
1870  sc::FeatureT ft_PFnH = sc::FeatureT::PfCand_nHad;
1871 
1872  // needed to remap indices from scaling vectors to those from dnn_inputs_v2::HadronBlockInputs
1873  int PFchH_index_offset = scalingParamsMap_->at(std::make_pair(ft_global, false)).mean_.size();
1874  int PFnH_index_offset = PFchH_index_offset + scalingParamsMap_->at(std::make_pair(ft_PFchH, false)).mean_.size();
1875 
1876  const auto& get = [&](int var_index) -> float& {
1877  if constexpr (std::is_same_v<HadronBlockType, std::vector<float>::iterator>) {
1878  return *(hadronBlockInputs + var_index);
1879  } else {
1880  return ((tensorflow::Tensor)hadronBlockInputs).tensor<float, 4>()(idx, 0, 0, var_index);
1881  }
1882  };
1883 
1884  const bool valid_chH = cell_map.count(CellObjectType::PfCand_chargedHadron);
1885  const bool valid_nH = cell_map.count(CellObjectType::PfCand_neutralHadron);
1886 
1887  if (!cell_map.empty()) {
1888  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_global, false));
1889  get(dnn::rho) = sp.scale(rho, dnn::rho);
1890  get(dnn::tau_pt) = sp.scale(tau.polarP4().pt(), dnn::tau_pt);
1891  get(dnn::tau_eta) = sp.scale(tau.polarP4().eta(), dnn::tau_eta);
1892  get(dnn::tau_inside_ecal_crack) = sp.scale(isInEcalCrack(tau.polarP4().eta()), dnn::tau_inside_ecal_crack);
1893  }
1894  if (valid_chH) {
1895  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_PFchH, is_inner));
1896  size_t index_chH = cell_map.at(CellObjectType::PfCand_chargedHadron);
1897  const auto& chH_cand = dynamic_cast<const CandidateCastType&>(pfCands.at(index_chH));
1898 
1899  get(dnn::pfCand_chHad_valid) = sp.scale(valid_chH, dnn::pfCand_chHad_valid - PFchH_index_offset);
1900  get(dnn::pfCand_chHad_rel_pt) =
1901  sp.scale(chH_cand.polarP4().pt() / tau.polarP4().pt(), dnn::pfCand_chHad_rel_pt - PFchH_index_offset);
1902  get(dnn::pfCand_chHad_deta) =
1903  sp.scale(chH_cand.polarP4().eta() - tau.polarP4().eta(), dnn::pfCand_chHad_deta - PFchH_index_offset);
1904  get(dnn::pfCand_chHad_dphi) =
1905  sp.scale(dPhi(tau.polarP4(), chH_cand.polarP4()), dnn::pfCand_chHad_dphi - PFchH_index_offset);
1906  get(dnn::pfCand_chHad_leadChargedHadrCand) =
1907  sp.scale(&chH_cand == dynamic_cast<const CandidateCastType*>(tau.leadChargedHadrCand().get()),
1908  dnn::pfCand_chHad_leadChargedHadrCand - PFchH_index_offset);
1909  get(dnn::pfCand_chHad_pvAssociationQuality) = sp.scale<int>(
1910  candFunc::getPvAssocationQuality(chH_cand), dnn::pfCand_chHad_pvAssociationQuality - PFchH_index_offset);
1911  get(dnn::pfCand_chHad_fromPV) =
1912  sp.scale<int>(candFunc::getFromPV(chH_cand), dnn::pfCand_chHad_fromPV - PFchH_index_offset);
1913  const float default_chH_pw_inner = 0.7614090f;
1914  const float default_chH_pw_outer = 0.1974930f;
1915  get(dnn::pfCand_chHad_puppiWeight) = is_inner ? sp.scale(candFunc::getPuppiWeight(chH_cand, default_chH_pw_inner),
1916  dnn::pfCand_chHad_puppiWeight - PFchH_index_offset)
1917  : sp.scale(candFunc::getPuppiWeight(chH_cand, default_chH_pw_outer),
1918  dnn::pfCand_chHad_puppiWeight - PFchH_index_offset);
1919  get(dnn::pfCand_chHad_puppiWeightNoLep) =
1920  is_inner ? sp.scale(candFunc::getPuppiWeightNoLep(chH_cand, default_chH_pw_inner),
1921  dnn::pfCand_chHad_puppiWeightNoLep - PFchH_index_offset)
1922  : sp.scale(candFunc::getPuppiWeightNoLep(chH_cand, default_chH_pw_outer),
1923  dnn::pfCand_chHad_puppiWeightNoLep - PFchH_index_offset);
1924  get(dnn::pfCand_chHad_charge) = sp.scale(chH_cand.charge(), dnn::pfCand_chHad_charge - PFchH_index_offset);
1925  get(dnn::pfCand_chHad_lostInnerHits) =
1926  sp.scale<int>(candFunc::getLostInnerHits(chH_cand, 0), dnn::pfCand_chHad_lostInnerHits - PFchH_index_offset);
1927  get(dnn::pfCand_chHad_numberOfPixelHits) = sp.scale(candFunc::getNumberOfPixelHits(chH_cand, 0),
1928  dnn::pfCand_chHad_numberOfPixelHits - PFchH_index_offset);
1929  get(dnn::pfCand_chHad_vertex_dx) =
1930  sp.scale(chH_cand.vertex().x() - pv.position().x(), dnn::pfCand_chHad_vertex_dx - PFchH_index_offset);
1931  get(dnn::pfCand_chHad_vertex_dy) =
1932  sp.scale(chH_cand.vertex().y() - pv.position().y(), dnn::pfCand_chHad_vertex_dy - PFchH_index_offset);
1933  get(dnn::pfCand_chHad_vertex_dz) =
1934  sp.scale(chH_cand.vertex().z() - pv.position().z(), dnn::pfCand_chHad_vertex_dz - PFchH_index_offset);
1935  get(dnn::pfCand_chHad_vertex_dx_tauFL) =
1936  sp.scale(chH_cand.vertex().x() - pv.position().x() - tau_funcs.getFlightLength(tau, tau_index).x(),
1937  dnn::pfCand_chHad_vertex_dx_tauFL - PFchH_index_offset);
1938  get(dnn::pfCand_chHad_vertex_dy_tauFL) =
1939  sp.scale(chH_cand.vertex().y() - pv.position().y() - tau_funcs.getFlightLength(tau, tau_index).y(),
1940  dnn::pfCand_chHad_vertex_dy_tauFL - PFchH_index_offset);
1941  get(dnn::pfCand_chHad_vertex_dz_tauFL) =
1942  sp.scale(chH_cand.vertex().z() - pv.position().z() - tau_funcs.getFlightLength(tau, tau_index).z(),
1943  dnn::pfCand_chHad_vertex_dz_tauFL - PFchH_index_offset);
1944 
1945  const bool hasTrackDetails = candFunc::getHasTrackDetails(chH_cand);
1946  if (hasTrackDetails) {
1947  get(dnn::pfCand_chHad_hasTrackDetails) =
1948  sp.scale(hasTrackDetails, dnn::pfCand_chHad_hasTrackDetails - PFchH_index_offset);
1949  get(dnn::pfCand_chHad_dxy) =
1950  sp.scale(candFunc::getTauDxy(chH_cand), dnn::pfCand_chHad_dxy - PFchH_index_offset);
1951  get(dnn::pfCand_chHad_dxy_sig) = sp.scale(std::abs(candFunc::getTauDxy(chH_cand)) / chH_cand.dxyError(),
1952  dnn::pfCand_chHad_dxy_sig - PFchH_index_offset);
1953  get(dnn::pfCand_chHad_dz) = sp.scale(candFunc::getTauDz(chH_cand), dnn::pfCand_chHad_dz - PFchH_index_offset);
1954  get(dnn::pfCand_chHad_dz_sig) = sp.scale(std::abs(candFunc::getTauDz(chH_cand)) / chH_cand.dzError(),
1955  dnn::pfCand_chHad_dz_sig - PFchH_index_offset);
1956  get(dnn::pfCand_chHad_track_chi2_ndof) =
1957  candFunc::getPseudoTrack(chH_cand).ndof() > 0
1958  ? sp.scale(candFunc::getPseudoTrack(chH_cand).chi2() / candFunc::getPseudoTrack(chH_cand).ndof(),
1959  dnn::pfCand_chHad_track_chi2_ndof - PFchH_index_offset)
1960  : 0;
1961  get(dnn::pfCand_chHad_track_ndof) =
1962  candFunc::getPseudoTrack(chH_cand).ndof() > 0
1963  ? sp.scale(candFunc::getPseudoTrack(chH_cand).ndof(), dnn::pfCand_chHad_track_ndof - PFchH_index_offset)
1964  : 0;
1965  }
1966  float hcal_fraction = candFunc::getHCalFraction(chH_cand, disable_hcalFraction_workaround_);
1967  get(dnn::pfCand_chHad_hcalFraction) =
1968  sp.scale(hcal_fraction, dnn::pfCand_chHad_hcalFraction - PFchH_index_offset);
1969  get(dnn::pfCand_chHad_rawCaloFraction) =
1970  sp.scale(candFunc::getRawCaloFraction(chH_cand), dnn::pfCand_chHad_rawCaloFraction - PFchH_index_offset);
1971  }
1972  if (valid_nH) {
1973  const sc::ScalingParams& sp = scalingParamsMap_->at(std::make_pair(ft_PFnH, is_inner));
1974  size_t index_nH = cell_map.at(CellObjectType::PfCand_neutralHadron);
1975  const auto& nH_cand = dynamic_cast<const CandidateCastType&>(pfCands.at(index_nH));
1976 
1977  get(dnn::pfCand_nHad_valid) = sp.scale(valid_nH, dnn::pfCand_nHad_valid - PFnH_index_offset);
1978  get(dnn::pfCand_nHad_rel_pt) =
1979  sp.scale(nH_cand.polarP4().pt() / tau.polarP4().pt(), dnn::pfCand_nHad_rel_pt - PFnH_index_offset);
1980  get(dnn::pfCand_nHad_deta) =
1981  sp.scale(nH_cand.polarP4().eta() - tau.polarP4().eta(), dnn::pfCand_nHad_deta - PFnH_index_offset);
1982  get(dnn::pfCand_nHad_dphi) =
1983  sp.scale(dPhi(tau.polarP4(), nH_cand.polarP4()), dnn::pfCand_nHad_dphi - PFnH_index_offset);
1984  get(dnn::pfCand_nHad_puppiWeight) = is_inner ? sp.scale(candFunc::getPuppiWeight(nH_cand, 0.9798355f),
1985  dnn::pfCand_nHad_puppiWeight - PFnH_index_offset)
1986  : sp.scale(candFunc::getPuppiWeight(nH_cand, 0.7813260f),
1987  dnn::pfCand_nHad_puppiWeight - PFnH_index_offset);
1988  get(dnn::pfCand_nHad_puppiWeightNoLep) = is_inner
1989  ? sp.scale(candFunc::getPuppiWeightNoLep(nH_cand, 0.9046796f),
1990  dnn::pfCand_nHad_puppiWeightNoLep - PFnH_index_offset)
1991  : sp.scale(candFunc::getPuppiWeightNoLep(nH_cand, 0.6554860f),
1992  dnn::pfCand_nHad_puppiWeightNoLep - PFnH_index_offset);
1993  float hcal_fraction = candFunc::getHCalFraction(nH_cand, disable_hcalFraction_workaround_);
1994  get(dnn::pfCand_nHad_hcalFraction) = sp.scale(hcal_fraction, dnn::pfCand_nHad_hcalFraction - PFnH_index_offset);
1995  }
1996  }
1997 
1998  static void calculateElectronClusterVars(const pat::Electron* ele, float& elecEe, float& elecEgamma) {
1999  if (ele) {
2000  elecEe = elecEgamma = 0;
2001  auto superCluster = ele->superCluster();
2002  if (superCluster.isNonnull() && superCluster.isAvailable() && superCluster->clusters().isNonnull() &&
2003  superCluster->clusters().isAvailable()) {
2004  for (auto iter = superCluster->clustersBegin(); iter != superCluster->clustersEnd(); ++iter) {
2005  const double energy = (*iter)->energy();
2006  if (iter == superCluster->clustersBegin())
2007  elecEe += energy;
2008  else
2009  elecEgamma += energy;
2010  }
2011  }
2012  } else {
2013  elecEe = elecEgamma = default_value;
2014  }
2015  }
2016 
2017  template <typename CandidateCollection, typename TauCastType>
2018  static void processSignalPFComponents(const TauCastType& tau,
2020  LorentzVectorXYZ& p4_inner,
2021  LorentzVectorXYZ& p4_outer,
2022  float& pt_inner,
2023  float& dEta_inner,
2024  float& dPhi_inner,
2025  float& m_inner,
2026  float& pt_outer,
2027  float& dEta_outer,
2028  float& dPhi_outer,
2029  float& m_outer,
2030  float& n_inner,
2031  float& n_outer) {
2032  p4_inner = LorentzVectorXYZ(0, 0, 0, 0);
2033  p4_outer = LorentzVectorXYZ(0, 0, 0, 0);
2034  n_inner = 0;
2035  n_outer = 0;
2036 
2037  const double innerSigCone_radius = getInnerSignalConeRadius(tau.pt());
2038  for (const auto& cand : candidates) {
2039  const double dR = reco::deltaR(cand->p4(), tau.leadChargedHadrCand()->p4());
2040  const bool isInside_innerSigCone = dR < innerSigCone_radius;
2041  if (isInside_innerSigCone) {
2042  p4_inner += cand->p4();
2043  ++n_inner;
2044  } else {
2045  p4_outer += cand->p4();
2046  ++n_outer;
2047  }
2048  }
2049 
2050  pt_inner = n_inner != 0 ? p4_inner.Pt() : default_value;
2051  dEta_inner = n_inner != 0 ? dEta(p4_inner, tau.p4()) : default_value;
2052  dPhi_inner = n_inner != 0 ? dPhi(p4_inner, tau.p4()) : default_value;
2053  m_inner = n_inner != 0 ? p4_inner.mass() : default_value;
2054 
2055  pt_outer = n_outer != 0 ? p4_outer.Pt() : default_value;
2056  dEta_outer = n_outer != 0 ? dEta(p4_outer, tau.p4()) : default_value;
2057  dPhi_outer = n_outer != 0 ? dPhi(p4_outer, tau.p4()) : default_value;
2058  m_outer = n_outer != 0 ? p4_outer.mass() : default_value;
2059  }
2060 
2061  template <typename CandidateCollection, typename TauCastType>
2062  static void processIsolationPFComponents(const TauCastType& tau,
2064  LorentzVectorXYZ& p4,
2065  float& pt,
2066  float& d_eta,
2067  float& d_phi,
2068  float& m,
2069  float& n) {
2070  p4 = LorentzVectorXYZ(0, 0, 0, 0);
2071  n = 0;
2072 
2073  for (const auto& cand : candidates) {
2074  p4 += cand->p4();
2075  ++n;
2076  }
2077 
2078  pt = n != 0 ? p4.Pt() : default_value;
2079  d_eta = n != 0 ? dEta(p4, tau.p4()) : default_value;
2080  d_phi = n != 0 ? dPhi(p4, tau.p4()) : default_value;
2081  m = n != 0 ? p4.mass() : default_value;
2082  }
2083 
2084  static double getInnerSignalConeRadius(double pt) {
2085  static constexpr double min_pt = 30., min_radius = 0.05, cone_opening_coef = 3.;
2086  // This is equivalent of the original formula (std::max(std::min(0.1, 3.0/pt), 0.05)
2087  return std::max(cone_opening_coef / std::max(pt, min_pt), min_radius);
2088  }
2089 
2090  // Copied from https://github.com/cms-sw/cmssw/blob/CMSSW_9_4_X/RecoTauTag/RecoTau/plugins/PATTauDiscriminationByMVAIsolationRun2.cc#L218
2091  template <typename TauCastType>
2092  static bool calculateGottfriedJacksonAngleDifference(const TauCastType& tau,
2093  const size_t tau_index,
2094  double& gj_diff,
2095  TauFunc tau_funcs) {
2096  if (tau_funcs.getHasSecondaryVertex(tau, tau_index)) {
2097  static constexpr double mTau = 1.77682;
2098  const double mAOne = tau.p4().M();
2099  const double pAOneMag = tau.p();
2100  const double argumentThetaGJmax = (std::pow(mTau, 2) - std::pow(mAOne, 2)) / (2 * mTau * pAOneMag);
2101  const double argumentThetaGJmeasured = tau.p4().Vect().Dot(tau_funcs.getFlightLength(tau, tau_index)) /
2102  (pAOneMag * tau_funcs.getFlightLength(tau, tau_index).R());
2103  if (std::abs(argumentThetaGJmax) <= 1. && std::abs(argumentThetaGJmeasured) <= 1.) {
2104  double thetaGJmax = std::asin(argumentThetaGJmax);
2105  double thetaGJmeasured = std::acos(argumentThetaGJmeasured);
2106  gj_diff = thetaGJmeasured - thetaGJmax;
2107  return true;
2108  }
2109  }
2110  return false;
2111  }
2112 
2113  template <typename TauCastType>
2114  static float calculateGottfriedJacksonAngleDifference(const TauCastType& tau,
2115  const size_t tau_index,
2116  TauFunc tau_funcs) {
2117  double gj_diff;
2118  if (calculateGottfriedJacksonAngleDifference(tau, tau_index, gj_diff, tau_funcs))
2119  return static_cast<float>(gj_diff);
2120  return default_value;
2121  }
2122 
2123  static bool isInEcalCrack(double eta) {
2124  const double abs_eta = std::abs(eta);
2125  return abs_eta > 1.46 && abs_eta < 1.558;
2126  }
2127 
2128  template <typename TauCastType>
2129  static const pat::Electron* findMatchedElectron(const TauCastType& tau,
2130  const std::vector<pat::Electron>* electrons,
2131  double deltaR) {
2132  const double dR2 = deltaR * deltaR;
2133  const pat::Electron* matched_ele = nullptr;
2134  for (const auto& ele : *electrons) {
2135  if (reco::deltaR2(tau.p4(), ele.p4()) < dR2 && (!matched_ele || matched_ele->pt() < ele.pt())) {
2136  matched_ele = &ele;
2137  }
2138  }
2139  return matched_ele;
2140  }
2141 
2142 protected:
2146  std::map<std::string, WPList> workingPoints_;
2147  const bool is_online_;
2149 
2150  const std::map<BasicDiscriminator, std::string> stringFromDiscriminator_{
2151  {BasicDiscriminator::ChargedIsoPtSum, "ChargedIsoPtSum"},
2152  {BasicDiscriminator::NeutralIsoPtSum, "NeutralIsoPtSum"},
2153  {BasicDiscriminator::NeutralIsoPtSumWeight, "NeutralIsoPtSumWeight"},
2154  {BasicDiscriminator::FootprintCorrection, "TauFootprintCorrection"},
2155  {BasicDiscriminator::PhotonPtSumOutsideSignalCone, "PhotonPtSumOutsideSignalCone"},
2156  {BasicDiscriminator::PUcorrPtSum, "PUcorrPtSum"}};
2162  const std::vector<BasicDiscriminator> requiredBasicDiscriminatorsdR03_{
2168 
2177  const unsigned year_;
2178  const unsigned version_;
2179  const unsigned sub_version_;
2180  const int debug_level;
2181  const bool disable_dxy_pca_;
2184  const std::map<std::pair<deep_tau::Scaling::FeatureT, bool>, deep_tau::Scaling::ScalingParams>* scalingParamsMap_;
2185  const bool save_inputs_;
2186  std::ofstream* json_file_;
2189  std::vector<int> tauInputs_indices_;
2190 
2191  //boolean to check if discriminator indices are already mapped
2192  bool discrIndicesMapped_ = false;
2193  std::map<BasicDiscriminator, size_t> basicDiscrIndexMap_;
2194  std::map<BasicDiscriminator, size_t> basicDiscrdR03IndexMap_;
2195 };
2196 
2197 #endif
edm::EDGetTokenT< std::vector< pat::Electron > > electrons_token_
size
Write out results.
std::vector< TauDiscInfo< pat::PATTauDiscriminator > > patPrediscriminants_
constexpr double deltaPhi(double phi1, double phi2)
Definition: deltaPhi.h:26
const std::map< std::pair< deep_tau::Scaling::FeatureT, bool >, deep_tau::Scaling::ScalingParams > * scalingParamsMap_
unsigned int n_photons_total(const reco::PFTau &tau)
return total number of pf photon candidates with pT>500 MeV, which are associated to signal ...
T getParameter(std::string const &) const
Definition: ParameterSet.h:307
std::string input_layer_
std::vector< int > tauInputs_indices_
ParameterDescriptionBase * addOptional(U const &iLabel, T const &value)
std::vector< size_t > num_
edm::EDGetTokenT< ConsumeType > disc_token
double pt() const final
transverse momentum
static bool isInEcalCrack(double eta)
edm::EDGetTokenT< double > rho_token_
float pt_weighted_dr_signal(const reco::PFTau &tau, int dm)
bool operator<(IOVSyncValue const &iLHS, IOVSyncValue const &iRHS)
Definition: IOVSyncValue.h:34
static const IDOutputCollection & GetIDOutputs()
vars
Definition: DeepTauIdBase.h:60
std::map< BasicDiscriminator, size_t > basicDiscrIndexMap_
float read_value(const std::vector< std::vector< float >> &pred, size_t tau_index, size_t elem) const
std::unique_ptr< Cutter > CutterPtr
static void processSignalPFComponents(const TauCastType &tau, const CandidateCollection &candidates, LorentzVectorXYZ &p4_inner, LorentzVectorXYZ &p4_outer, float &pt_inner, float &dEta_inner, float &dPhi_inner, float &m_inner, float &pt_outer, float &dEta_outer, float &dPhi_outer, float &m_outer, float &n_inner, float &n_outer)
working_points
Definition: FWLite.py:140
const std::vector< BasicDiscriminator > requiredBasicDiscriminators_
const bool save_inputs_
std::string output_layer_
bool getByToken(EDGetToken token, Handle< PROD > &result) const
Definition: Event.h:526
std::unique_ptr< TauDiscriminator > get_value(const edm::Handle< TauCollection > &taus, const PredType &pred, const WPList *working_points, bool is_online) const
const int debug_level
reco::SuperClusterRef superCluster() const override
override the reco::GsfElectron::superCluster method, to access the internal storage of the superclust...
void countMatches(const reco::Muon &muon, std::vector< int > &numMatchesDT, std::vector< int > &numMatchesCSC, std::vector< int > &numMatchesRPC)
static float calculateGottfriedJacksonAngleDifference(const TauCastType &tau, const size_t tau_index, TauFunc tau_funcs)
void find(edm::Handle< EcalRecHitCollection > &hits, DetId thisDet, std::vector< EcalRecHitCollection::const_iterator > &hit, bool debug=false)
Definition: FindCaloHit.cc:19
static bool calculateElectronClusterVarsV2(const pat::Electron &ele, float &cc_ele_energy, float &cc_gamma_energy, int &cc_n_gamma)
std::vector< Vertex > VertexCollection
Definition: Vertex.h:31
edm::RefProd< PFTauCollection > PFTauRefProd
references to PFTau collection
Definition: PFTauFwd.h:15
muons
the two sets of parameters below are mutually exclusive, depending if RECO or ALCARECO is used the us...
Definition: DiMuonV_cfg.py:214
const std::map< ValueQuantityType, double > min_value
ParameterSet const & parameterSet(StableProvenance const &provenance, ProcessHistory const &history)
Definition: Provenance.cc:11
static constexpr float default_value
void createEgammaBlockInputs(unsigned idx, const TauCastType &tau, const size_t tau_index, const edm::RefToBase< reco::BaseTau > tau_ref, const reco::Vertex &pv, double rho, const std::vector< pat::Electron > *electrons, const edm::View< reco::Candidate > &pfCands, const Cell &cell_map, TauFunc tau_funcs, bool is_inner, EgammaBlockType &egammaBlockInputs)
static float getValue(T value)
static const double deltaEta
Definition: CaloConstants.h:8
edm::Handle< ConsumeType > handle
static bool isAbove(double value, double min)
static bool calculateGottfriedJacksonAngleDifference(const TauCastType &tau, const size_t tau_index, double &gj_diff, TauFunc tau_funcs)
static void processIsolationPFComponents(const TauCastType &tau, const CandidateCollection &candidates, LorentzVectorXYZ &p4, float &pt, float &d_eta, float &d_phi, float &m, float &n)
Definition: HeavyIon.h:7
uint8_t andPrediscriminants_
edm::EDGetTokenT< CandidateCollection > pfcandToken_
const std::map< std::pair< FeatureT, bool >, ScalingParams > scalingParamsMap_v2p5
edm::EDGetTokenT< std::vector< pat::Muon > > muons_token_
static const int tauID
Definition: TopGenEvent.h:20
char const * label
static void fillDescriptionsHelper(edm::ParameterSetDescription &desc)
const bool disable_hcalFraction_workaround_
ALPAKA_FN_ACC static ALPAKA_FN_INLINE float dR2(Position4 pos1, Position4 pos2)
std::vector< Electron > ElectronCollection
Definition: Electron.h:36
static float getValueNorm(T value, float mean, float sigma, float n_sigmas_max=5)
Definition: Muon.py:1
float pt_weighted_deta_strip(const reco::PFTau &tau, int dm)
static double getInnerSignalConeRadius(double pt)
const std::map< BasicDiscriminator, std::string > stringFromDiscriminator_
edm::EDGetTokenT< reco::TauDiscriminatorContainer > basicTauDiscriminators_inputToken_
std::map< std::string, IDOutput > IDOutputCollection
edm::EDGetTokenT< TauCollection > tausToken_
std::map< BasicDiscriminator, size_t > basicDiscrdR03IndexMap_
void createHadronsBlockInputs(unsigned idx, const TauCastType &tau, const size_t tau_index, const edm::RefToBase< reco::BaseTau > tau_ref, const reco::Vertex &pv, double rho, const edm::View< reco::Candidate > &pfCands, const Cell &cell_map, TauFunc tau_funcs, bool is_inner, HadronBlockType &hadronBlockInputs)
bool discrIndicesMapped_
void createTauBlockInputs(const TauCastType &tau, const size_t &tau_index, const edm::RefToBase< reco::BaseTau > tau_ref, const reco::Vertex &pv, double rho, TauFunc tau_funcs, TauBlockType &tauBlockInputs)
std::vector< std::string > getParameterNamesForType(bool trackiness=true) const
Definition: ParameterSet.h:180
const std::vector< BasicDiscriminator > requiredBasicDiscriminatorsdR03_
Abs< T >::type abs(const T &t)
Definition: Abs.h:22
float pt_weighted_dr_iso(const reco::PFTau &tau, int dm)
double f[11][100]
static const pat::Electron * findMatchedElectron(const TauCastType &tau, const std::vector< pat::Electron > *electrons, double deltaR)
def template(fileName, svg, replaceme="REPLACEME")
Definition: svgfig.py:521
edm::EDGetTokenT< reco::TauDiscriminatorContainer > basicTauDiscriminatorsdR03_inputToken_
Definition: value.py:1
ParameterDescriptionBase * add(U const &iLabel, T const &value)
const_reference at(size_type pos) const
void createMuonBlockInputs(unsigned idx, const TauCastType &tau, const size_t tau_index, const edm::RefToBase< reco::BaseTau > tau_ref, const reco::Vertex &pv, double rho, const std::vector< pat::Muon > *muons, const edm::View< reco::Candidate > &pfCands, const Cell &cell_map, TauFunc tau_funcs, bool is_inner, MuonBlockType &muonBlockInputs)
ROOT::Math::LorentzVector< ROOT::Math::PxPyPzE4D< double > > LorentzVectorXYZ
std::vector< CutterPtr > WPList
Analysis-level tau class.
Definition: Tau.h:53
void fill(const edm::Event &evt)
std::vector< size_t > den_
const std::map< BasicDiscriminator, size_t > matchDiscriminatorIndices(edm::Event const &event, edm::EDGetTokenT< reco::TauDiscriminatorContainer > discriminatorContainerToken, std::vector< BasicDiscriminator > requiredDiscr)
const std::map< std::pair< FeatureT, bool >, ScalingParams > scalingParamsMap_v2p1
#define M_PI
void createOutputs(edm::Event &event, const PredType &pred, edm::Handle< TauCollection > taus)
constexpr auto deltaR(const T1 &t1, const T2 &t2) -> decltype(t1.eta())
Definition: deltaR.h:30
std::ofstream * json_file_
static void calculateElectronClusterVars(const pat::Electron *ele, float &elecEe, float &elecEgamma)
IDOutput(const std::vector< size_t > &num, const std::vector< size_t > &den)
CellObjectType
constexpr auto deltaR2(const T1 &t1, const T2 &t2) -> decltype(t1.eta())
Definition: deltaR.h:16
edm::ValueMap< SingleTauDiscriminatorContainer > TauDiscriminatorContainer
const bool disable_CellIndex_workaround_
float pt_weighted_dphi_strip(const reco::PFTau &tau, int dm)
float eratio(const reco::PFTau &tau)
return ratio of energy in ECAL over sum of energy in ECAL and HCAL
void loadPrediscriminants(edm::Event const &event, edm::Handle< TauCollection > const &taus)
static constexpr int RPC
Definition: MuonSubdetId.h:13
Analysis-level electron class.
Definition: Electron.h:51
const bool is_online_
edm::Ref< PFTauTransverseImpactParameterCollection > PFTauTransverseImpactParameterRef
presistent reference to a PFTauTransverseImpactParameter
IDOutputCollection idoutputs_
std::vector< Muon > MuonCollection
Definition: Muon.h:35
DeepTauIdBase(const edm::ParameterSet &cfg)
const bool disable_dxy_pca_
void fillGrids(const TauCastType &tau, const Collection &objects, CellGrid &inner_grid, CellGrid &outer_grid)
Particle reconstructed by the particle flow algorithm.
Definition: PFCandidate.h:41
fixed size matrix
HLT enums.
const std::map< ValueQuantityType, double > max_value
const unsigned sub_version_
pat::ElectronCollection ElectronCollection
const unsigned version_
edm::EDGetTokenT< reco::VertexCollection > vtxToken_
std::pair< typename Association::data_type::first_type, double > match(Reference key, Association association, bool bestMatchByMaxValue)
Generic matching function.
Definition: Utils.h:10
float read_value(const tensorflow::Tensor &pred, size_t tau_index, size_t elem) const
Definition: output.py:1
static constexpr float pi
float lead_track_chi2(const reco::PFTau &tau)
return chi2 of the leading track ==> deprecated? <==
std::vector< TauDiscInfo< reco::PFTauDiscriminator > > recoPrediscriminants_
const unsigned year_
static constexpr int DT
Definition: MuonSubdetId.h:11
Log< level::Warning, false > LogWarning
TupleMultiplicity< TrackerTraits > const *__restrict__ uint32_t nHits
T operator[](int i) const
long double T
static constexpr int CSC
Definition: MuonSubdetId.h:12
static const std::string subdets[7]
Definition: TrackUtils.cc:60
const std::map< std::pair< FeatureT, bool >, ScalingParams > scalingParamsMap_PhaseIIv2p5
Analysis-level muon class.
Definition: Muon.h:51
static float getValueLinear(T value, float min_value, float max_value, bool positive)
void countHits(const reco::Muon &muon, std::vector< int > &numHitsDT, std::vector< int > &numHitsCSC, std::vector< int > &numHitsRPC)
Power< A, B >::type pow(const A &a, const B &b)
Definition: Power.h:29
def move(src, dest)
Definition: eostools.py:511
Definition: event.py:1
std::map< std::string, WPList > workingPoints_
constexpr int NumberOfOutputs
Definition: DeepTauIdBase.h:50
edm::EDGetTokenT< edm::AssociationVector< reco::PFTauRefProd, std::vector< reco::PFTauTransverseImpactParameterRef > > > pfTauTransverseImpactParameters_token_
unsigned transform(const HcalDetId &id, unsigned transformCode)