GBDT-Based Displaced Vertex Producer #1257

ryanm124 · 2024-05-14T08:58:32Z

Updated version of the displaced vertex producer presented at L1 DPG: https://indico.cern.ch/event/1408668/contributions/5920520/attachments/2850847/4984793/Displaced%20Vertexing%20Low%20Fake%20DP%20Note.pdf

Contains the latest GBDT trained on Dark Photon and Neutrino Gun samples, trackAndVert_NewTQMVA_model.onnx.

…e, and updating displaced vertex BDT

triggerDoctor · 2024-05-14T12:43:35Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

Attempts to compile this PR succeeded!

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b -j 8`

triggerDoctor · 2024-05-14T13:04:46Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

I found no issues with the code checks!

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b -k -j 8 code-checks && scram b -k -j 8 code-checks`

I found issues with the headers. These issues likely arise from improper build files.

The following is the stderr of the header checking:



In file included from src/L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h:12:
/cvmfs/cms.cern.ch/el9_amd64_gcc12/cms/cmssw/CMSSW_14_0_0_pre3/src/PhysicsTools/ONNXRuntime/interface/ONNXRuntime.h:19:10: fatal error: onnxruntime/core/session/onnxruntime_cxx_api.h: No such file or directory
   19 | #include "onnxruntime/core/session/onnxruntime_cxx_api.h"
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
gmake: *** [config/SCRAM/GMake/Makefile.rules:1833: tmp/el9_amd64_gcc12/src/L1Trigger/L1TTrackMatch/src/L1TriggerL1TTrackMatch/DisplacedVertexProducer.h.chk_header] Error 1
gmake: Target 'check-headers' not remade because of errors.

Info	Value
return code	`2`
command	`eval` scramv1 runtime -sh `&& scram b -k -j 8 check-headers`

triggerDoctor · 2024-05-14T13:05:22Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

I found 10 files that did not meet formatting requirements:

DataFormats/L1Trigger/interface/DisplacedVertex.h
L1Trigger/L1TMuon/src/MicroGMTConfiguration.cc
L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h
L1Trigger/L1TTrackMatch/plugins/DisplacedVertexProducer.cc
L1Trigger/L1TTrackMatch/test/Analyzer_DisplacedMuon.C
L1Trigger/L1TTrackMatch/test/L1TrackObjectNtupleMaker.cc
L1Trigger/Phase2L1GMT/interface/TPSLUTs.h
L1Trigger/Phase2L1GMT/plugins/Phase2L1TGMTFwdMuonTranslator.cc
L1Trigger/TrackFindingTracklet/plugins/L1FPGATrackProducer.cc
L1Trigger/TrackTrigger/src/L1TrackQuality.cc

Please run scram b code-format to auto-apply code formatting

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b -k -j 8 code-format-all`

triggerDoctor · 2024-05-14T13:26:55Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

This PR passes available unit tests!

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b runtests`

triggerDoctor · 2024-05-14T13:37:13Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation

I found a non-zero return code running the relval workflows for this PR!

Info	Value
return code	`1`
command	`eval` scramv1 runtime -sh `&& runTheMatrix.py --what upgrade -l 26834.78`

cgsavard · 2024-05-14T13:53:22Z

L1Trigger/TrackTrigger/data/clf_GBDT_newbin_disp_xgb.onnx

We normally put these files here (https://github.com/cms-data/L1Trigger-TrackTrigger) to avoid having non-readable files in this repository. You will have to put in a PR to that repository to put it in there.

Done, cms-data/L1Trigger-TrackTrigger#4

cgsavard · 2024-05-14T13:54:47Z

L1Trigger/TrackTrigger/src/L1TrackQuality.cc

@@ -90,7 +90,7 @@ std::vector<float> L1TrackQuality::featureTransform(TTTrack<Ref_Phase2TrackerDig
  feature_map["bendchi2_bin"] = tmp_trk_bendchi2_bin;
  feature_map["chi2rphi_bin"] = tmp_trk_chi2rphi_bin;
  feature_map["chi2rz_bin"] = tmp_trk_chi2rz_bin;
-  feature_map["tanl"] = tmp_trk_tanl;


Why is this being removed? Doesn't the prompt classifier need this?

Just added this back in commit dbf8525

cgsavard · 2024-05-14T14:04:20Z

L1Trigger/TrackTrigger/src/L1TrackQuality.cc

@@ -159,7 +159,9 @@ void L1TrackQuality::setL1TrackQuality(TTTrack<Ref_Phase2TrackerDigi_>& aTrack)
    }

    else if (this->qualityAlgorithm_ == QualityAlgorithm::GBDT) {
-      aTrack.settrkMVA1(ortoutputs[1][1]);
+      if (this->featureNames_.back() == "d0"){ // temp fix, need better way to fill disp MVA


This really should have a better solution. I would suggest passing the output value through the function and setting the MVA variable in L1FPGATrackProducer.cc with an extended tracking flag since we shouldn't run this BDT unless extended tracking is running because d0 isn't filled.

I've implemented your suggestion in commit dbf8525

cgsavard · 2024-05-14T14:05:56Z

L1Trigger/TrackFindingTracklet/plugins/L1FPGATrackProducer.cc

    if (trackQuality_) {
      trackQualityModel_->setL1TrackQuality(aTrack);
+      trackQualityDispModel_->setL1TrackQuality(aTrack);


There should be an extended tracking flag here since this should only run for extended tracking. There is a extended_ variable in this file that can be used.

Done here dbf8525

cgsavard · 2024-05-14T14:10:27Z

L1Trigger/L1TTrackMatch/test/L1TrackObjectNtupleMaker_cfg.py

@@ -42,27 +42,28 @@
 # input and output
 ############################################################

-process.maxEvents = cms.untracked.PSet(input = cms.untracked.int32(20))
+


All the configurations in this file should be set back to the default configurations used by everyone. So the maxEvents, readFiles, skipEvents, useJobReport, fileName, etc.

Reverted these changes as part of d95ffe9

Chriisbrown · 2024-05-15T19:15:59Z

L1Trigger/L1TTrackMatch/python/DisplacedVertexProducer_cfi.py

+  l1TracksInputTag = cms.InputTag("l1tTTTracksFromExtendedTrackletEmulation", "Level1TTTracks"),
+  l1TrackVertexCollectionName = cms.string("dispVertices"),
+  mcTruthTrackInputTag = cms.InputTag("TTTrackAssociatorFromPixelDigisExtended", "Level1TTTracks"),
+  ONNXmodel = cms.string("/afs/cern.ch/user/r/rmccarth/private/dispVert/CMSSW_14_0_0_pre3/src/L1Trigger/L1TTrackMatch/test/trackAndVert_NewTQMVA_model.onnx"),


Similar to Claire's comment below this also needs to be in a cms-data repository, we have one for L1TTrackMatch here: https://github.com/cms-data/L1Trigger-L1TTrackMatch

Just made a PR: cms-data/L1Trigger-L1TTrackMatch#2

Chriisbrown · 2024-05-15T19:18:22Z

L1Trigger/L1TTrackMatch/plugins/DisplacedVertexProducer.cc

+      if(fabs(vertex.b.d0)<fabs(minD0)) minD0 = vertex.b.d0;
+      //std::vector<float> Transformed_features = {vertex.delta_z, vertex.R_T, vertex.cos_T, vertex.d_T, vertex.chi2rzdofSum, float(vertex.numStubsSum), vertex.chi2rphidofSum, minD0, vertex.a.pt+vertex.b.pt};
+      std::vector<float> Transformed_features = {selectedTracks[i].pt, selectedTracks[j].pt, selectedTracks[i].eta, selectedTracks[j].eta, selectedTracks[i].phi, selectedTracks[j].phi, selectedTracks[i].d0, selectedTracks[j].d0, selectedTracks[i].z0, selectedTracks[j].z0, selectedTracks[i].chi2rz, selectedTracks[j].chi2rz, selectedTracks[i].bendchi2, selectedTracks[j].bendchi2, selectedTracks[i].MVA1, selectedTracks[j].MVA1, selectedTracks[i].MVA2, selectedTracks[j].MVA2, vertex.d_T, vertex.R_T, vertex.cos_T, vertex.delta_z};
+      //cms::Ort::ONNXRuntime Runtime(this->ONNXmodel_);  //Setup ONNX runtime


This comment doesn't need to be here, in fact we should remove the similar comment from the TrackQuality.cc as the Runtime is now initialised (as you do) in the constructor

Removed in commit 63cdfcf

Chriisbrown · 2024-05-15T19:19:58Z

L1Trigger/L1TTrackMatch/python/DisplacedVertexProducer_cfi.py

+  mcTruthTrackInputTag = cms.InputTag("TTTrackAssociatorFromPixelDigisExtended", "Level1TTTracks"),
+  ONNXmodel = cms.string("/afs/cern.ch/user/r/rmccarth/private/dispVert/CMSSW_14_0_0_pre3/src/L1Trigger/L1TTrackMatch/test/trackAndVert_NewTQMVA_model.onnx"),
+  ONNXInputName = cms.string("feature_input"),
+  featureNames = cms.vstring(['trkExt_pt_firstTrk', 'trkExt_pt', 'trkExt_eta_firstTrk', 'trkExt_eta', 'trkExt_phi_firstTrk', 'trkExt_phi', 'trkExt_d0_firstTrk', 'trkExt_d0', 'trkExt_z0_firstTrk', 'trkExt_z0', 'trkExt_chi2rz_firstTrk', 'trkExt_chi2rz', 'trkExt_bendchi2_firstTrk', 'trkExt_bendchi2', 'trkExt_MVA_firstTrk', 'trkExt_MVA', 'trkExt_MVA2_firstTrk', 'trkExt_MVA2', 'dv_d_T', 'dv_R_T', 'dv_cos_T', 'dv_del_Z'])


Are these featureNames used by the code anywhere, in the track quality code they are used in a mapping but you just use all of these in a single vector

They are read in by the DisplacedVertexProducer (https://github.com/ryanm124/cmssw/blob/d95ffe97501c7e003b3bc9de7e83685ecc20a6a6/L1Trigger/L1TTrackMatch/plugins/DisplacedVertexProducer.cc#L112) but not actually used. I could build up a map of inputs using this list or just remove the list entirely, whichever you think is best.

We have a map of inputs in the track quality to allow some flexibility over the inputs but if your inputs are set and you don't foresee any change I would not have these features names read in and unused and instead have a comment here in the python config about what features the BDT uses

I've turned this into a comment as you suggested here 63cdfcf

Chriisbrown · 2024-05-15T19:47:28Z

L1Trigger/L1TTrackMatch/plugins/DisplacedVertexProducer.cc

+    float MVA2 = iterL1Track->trkMVA2();
+    std::vector<edm::Ref<edmNew::DetSetVector<TTStub<Ref_Phase2TrackerDigi_> >, TTStub<Ref_Phase2TrackerDigi_> > > stubRefs = iterL1Track->getStubRefs();
+    int nstub = (int)stubRefs.size();
+    if( chi2rz<3.0 && MVA2>0.2 && MVA1>0.2 && pt>3.0 && fabs(eta)<2.4){


I would recommend moving these track selection criteria to the python config to avoid the hard coding of these criteria here or using the track selection producer

I second using the track selection producer as this is much cleaner. You will just have to create another config in the python config and add whatever varibles are not already there (like MVA2), and then pass this track collection into the displaced vertex config here.

I'm looking at the track selector and I'm wondering if using this is in fact cleaner than simply moving the parameters to a python config as Christopher suggested. The selector producer seems to be made for simpler, single variable cuts while I have several selections that depend on two variables such as the MVA score of displaced (d0>1cm) tracks. The selector also is cutting on all defined variables regardless of which ones the user has overwritten so there are many wasted operations such as checking if reducedBendChi2MaxNstub5<999.9 which would always be true.

I would agree with @Chriisbrown and @cgsavard. You should add these to the python configuration of this module so the values of the cuts can be set there, but keep the logic here. I don't see where wasted checks and operations would happen there?

@aloeliger For example, reducedChi2RZMaxNstub4 is defined which I'm currently not cutting on. This parameter is used as part of the chi2nstub selection which is used for every track here. So this track selection producer seems to cut on every defined variable even though for many of the default values such as this one the cut is essentially a Boolean true

You shouldn't need to set a magnitude size, this is for when a variable needs to be converted from a track word variable. Since d0 is a variable directly included in the track word, no conversion is needed. You should follow what is being done for getting the z0 using the proper d0 functions from the track word class.

I've modified the displaced vertex producer to use the track selection producer here 63cdfcf

Does the MC track association have to be redone after using the track selection producer? The truth association seems to not be working now after this change

No, it shouldn't change anything except for where track selection is being done. The only thing I can think of is some of the files don't have the truth association included so maybe you are looking at one of those files?

I've been using the same file for testing these PR changes and I was able to get the tracking particles before. Now it's giving a null pointer for each track. I looked at the other producers that use the track selection and it seems like they don't use truth information. Has anyone done this before?

Chriisbrown · 2024-05-15T19:54:15Z

L1Trigger/L1TTrackMatch/test/L1TrackObjectNtupleMaker_cfg.py

@@ -193,6 +194,16 @@
    process.pTkMHT = cms.Path(process.l1tTrackerHTMiss*process.l1tTrackerHTMissExtended)
    process.pTkMHTEmulator = cms.Path(process.l1tTrackerEmuHTMiss*process.l1tTrackerEmuHTMissExtended)
    process.pL1TrackTripletEmulator = cms.Path(process.l1tTrackTripletEmulation)
+    process.DispVert = cms.Path(process.DisplacedVertexProducer)
+    process.printTree = cms.EDAnalyzer("ParticleTreeDrawer",


Can this be added as an optional part of the ObjectNtupleMaker rather than by default unless it is something that everyone would want running when creating the L1TTrack Objects

I added a boolean in the cfg to run the displaced vertexing which is false by default here d95ffe9

cgsavard · 2024-05-15T20:44:41Z

L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h

+
+using namespace std;
+
+class Track_Parameters


Is it necessary to create this new class? Can you just use the TTTrack class?

It was convenient to define a class which contains the parameters of the projection of the track on the transverse plane just so I only calculate it once per track. Is there a preference for using functions or class methods for this type of thing?

The preference is not necessarily for functions versus class members, but for condensing the number of in use classes to try and consolidate code and functionality.

There are at least a few things conceptually similar between TTrack and it's parent class TTTrack_TrackWord and this track parameters class. It would be good if the track trigger developers could try to centralize as many common elements to most types of tracks in one place if possible

So what would you recommend? I could try inheriting from TTrack in my class and only add the extra functionality I need or I could try to do without an additional class at all.

My recommendation would be to remove this class entirely, pass the L1 track into these functions and use the TTTrack/TTTrack_Word variables directly, and similarly for everywhere else this is used like here.

… when setting track quality score

cgsavard · 2024-05-17T13:29:07Z

L1Trigger/TrackFindingTracklet/plugins/L1FPGATrackProducer.cc

+	  aTrack.settrkMVA2(qualityScore);
+	}
+	else{
+	  aTrack.settrkMVA1(qualityScore);


This is not quite doing what we want. It seems like when extended==false, trkMVA1 is being overwritten with the qualityScore from the displaced model. We should remove the else statement, and only MVA2 is filled if extended_==true.

Also, why is there an additional constraint on the displaced model being a BDT? The displaced model should just be whatever is in the configuration and this seems like it will not let any changes to occur? I think it's better to make all changes to the configs than to the hardset code

Currently only the BDT is defined for displaced tracks, right? I wanted to allow for the use of extended tracking with the prompt classifiers such as the NN. I've also found it useful to use both the prompt and displaced MVA score while using the extended tracking which is why I've set it up this way.

That's right. I think we should always fill MVA1 with the prompt classifier no matter what (so that would be the removal of the else statement) and if someone wants to switch the prompt classifier to another one, say a NN, then that is done in the config. Then whenever extended tracking is run, we should fill MVA2 with the displaced one in addition to the prompt MVA1 one. And if people want to change the displaced one then that should happen in the config, no additional filtering should really be done here.

So here is what I'm imagining:

if (trackQuality_) { aTrack.settrkMVA1(trackQualityModel_->getL1TrackQuality(aTrack)); if(extended_) aTrack.settrkMVA2(trackQualityDispModel_->getL1TrackQuality(aTrack));

Okay, I've made the fix d95ffe9

…changes to ntuple maker params, removing BDT requirement on displaced track quality

triggerDoctor · 2024-05-20T08:20:22Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

Attempts to compile this PR succeeded!

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b -j 8`

triggerDoctor · 2024-05-20T08:57:26Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

I found no issues with the code checks!

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b -k -j 8 code-checks && scram b -k -j 8 code-checks`

I found issues with the headers. These issues likely arise from improper build files.

The following is the stderr of the header checking:

In file included from src/L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h:12:
/cvmfs/cms.cern.ch/el9_amd64_gcc12/cms/cmssw/CMSSW_14_0_0_pre3/src/PhysicsTools/ONNXRuntime/interface/ONNXRuntime.h:19:10: fatal error: onnxruntime/core/session/onnxruntime_cxx_api.h: No such file or directory
   19 | #include "onnxruntime/core/session/onnxruntime_cxx_api.h"
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
gmake: *** [config/SCRAM/GMake/Makefile.rules:1833: tmp/el9_amd64_gcc12/src/L1Trigger/L1TTrackMatch/src/L1TriggerL1TTrackMatch/DisplacedVertexProducer.h.chk_header] Error 1
gmake: Target 'check-headers' not remade because of errors.

Info	Value
return code	`2`
command	`eval` scramv1 runtime -sh `&& scram b -k -j 8 check-headers`

triggerDoctor · 2024-05-20T08:58:03Z

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

I found 11 files that did not meet formatting requirements:

DataFormats/L1Trigger/interface/DisplacedVertex.h
L1Trigger/L1TMuon/src/MicroGMTConfiguration.cc
L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h
L1Trigger/L1TTrackMatch/plugins/DisplacedVertexProducer.cc
L1Trigger/L1TTrackMatch/test/Analyzer_DisplacedMuon.C
L1Trigger/L1TTrackMatch/test/L1TrackObjectNtupleMaker.cc
L1Trigger/Phase2L1GMT/interface/TPSLUTs.h
L1Trigger/Phase2L1GMT/plugins/Phase2L1TGMTFwdMuonTranslator.cc
L1Trigger/TrackFindingTracklet/plugins/L1FPGATrackProducer.cc
L1Trigger/TrackTrigger/interface/L1TrackQuality.h
L1Trigger/TrackTrigger/src/L1TrackQuality.cc

Please run scram b code-format to auto-apply code formatting

Info	Value
return code	`0`
command	`eval` scramv1 runtime -sh `&& scram b -k -j 8 code-format-all`

aloeliger · 2024-05-20T14:36:36Z

L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h

+  if(a.charge>0){
+    p_vec *= -1;
+  }
+  p_vec /= TMath::Sqrt(pow(p_vec[0],2)+pow(p_vec[1],2));


Is it possible that the arguments to this vector will end up as 0? This will causes errors if so.

Added a check for this in this commit 63cdfcf

aloeliger · 2024-05-20T14:37:24Z

L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h

+    std::valarray<float> p_trk_2 = calcPVec(b_in,x_dv_in,y_dv_in);
+    std::valarray<float> p_tot = p_trk_1+p_trk_2;
+    p_mag = TMath::Sqrt(pow(p_tot[0],2)+pow(p_tot[1],2));
+    openingAngle = (p_trk_1[0]*p_trk_2[0]+p_trk_1[1]*p_trk_2[1]) / (TMath::Sqrt(pow(p_trk_1[0],2)+pow(p_trk_1[1],2))*TMath::Sqrt(pow(p_trk_2[0],2)+pow(p_trk_2[1],2)));


Same question about possible division by zero errors here.

Added a check for this in this commit 63cdfcf

aloeliger · 2024-05-20T14:37:39Z

L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h

+    p_mag = TMath::Sqrt(pow(p_tot[0],2)+pow(p_tot[1],2));
+    openingAngle = (p_trk_1[0]*p_trk_2[0]+p_trk_1[1]*p_trk_2[1]) / (TMath::Sqrt(pow(p_trk_1[0],2)+pow(p_trk_1[1],2))*TMath::Sqrt(pow(p_trk_2[0],2)+pow(p_trk_2[1],2)));
+    R_T = TMath::Sqrt(pow(x_dv_in,2)+pow(y_dv_in,2));
+    cos_T = (p_tot[0]*x_dv_in+p_tot[1]*y_dv_in)/(R_T*TMath::Sqrt(pow(p_tot[0],2)+pow(p_tot[1],2)));


Added a check for this in this commit 63cdfcf

aloeliger · 2024-05-20T14:38:25Z

L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h

+
+class Vertex_Parameters
+{
+public:


Are there any extant Track Trigger classes for handling vertexes that would be good to centralize this class with?

My personal preference would be to get rid of this class as well and use these variables directly in the DisplacedVertexProducer class to avoid another vertex-like class, but I also see why you may want to set it up like this. It doesn't quite follow how the other GTT object classes are set up so I would suggest maybe changing this class to a struct instead as it's really only holding variables to be used by DisplacedVertexProducer.

There is no vertex class that has all of these variables.

can this (class => functions) be done as a "code improvement" in a second stage? L1 is requesting that this PR is merged in order for @ryanm124 to show the plots at an upcoming conference. that request is to have code to be able to reproduce the results, and it seems code improvements wouldn't be necessary for that?

This is ok with me for this comment. There does seem to be some potential logic issues brought up in other comments that might need more addressing. And at minimum, the code checks and formatting should be done and It doesn't seem like the formatting has been applied. You should run these two commands:

scram build code-checks scram build code-format

Code formatting has been applied 93ea36d

aloeliger · 2024-05-20T14:40:56Z

L1Trigger/L1TTrackMatch/interface/DisplacedVertexProducer.h

+class DisplacedVertexProducer : public edm::global::EDProducer<> {
+ public:
+  explicit DisplacedVertexProducer(const edm::ParameterSet&);
+  ~DisplacedVertexProducer() override {}


Suggested change

~DisplacedVertexProducer() override {}

~DisplacedVertexProducer() override = default;

I've implemented this

aloeliger · 2024-05-20T15:06:33Z

L1Trigger/L1TTrackMatch/plugins/DisplacedVertexProducer.cc

+    float MVA2 = iterL1Track->trkMVA2();
+    std::vector<edm::Ref<edmNew::DetSetVector<TTStub<Ref_Phase2TrackerDigi_> >, TTStub<Ref_Phase2TrackerDigi_> > > stubRefs = iterL1Track->getStubRefs();
+    int nstub = (int)stubRefs.size();
+    if( chi2rz<3.0 && MVA2>0.2 && MVA1>0.2 && pt>3.0 && fabs(eta)<2.4){


I would agree with @Chriisbrown and @cgsavard. You should add these to the python configuration of this module so the values of the cuts can be set there, but keep the logic here. I don't see where wasted checks and operations would happen there?

aloeliger · 2024-05-20T15:09:41Z

L1Trigger/L1TTrackMatch/test/Analyzer_DisplacedMuon.C

+  //preselection cuts and plots definitions
+  // Cut assumptions: first cut is maxEta
+  std::vector<std::unique_ptr<Cut>> preselCuts;
+  std::unique_ptr<TypedCut<float>> cut0(new TypedCut<float>("maxEta","max #eta",&trk_eta,2.4,true));


I know this is test code and isn't really intended for main CMSSW use, but std::make_unique is probably preferred.

aloeliger · 2024-05-20T15:19:14Z

L1Trigger/L1TTrackMatch/test/Analyzer_DisplacedMuon.C

Does this need to go into CMSSW? It seems like a large single purpose macro, designed for one person or group only.

No it doesn't, I can remove it from the PR

aloeliger · 2024-05-20T15:21:56Z

L1Trigger/L1TTrackMatch/test/trackAndVert_NewTQMVA_model.onnx

This cannot be committed to CMSSW. It's fine here I guess, but in a PR to CMSSW all non-code files need to be committed to the data repositories.

aloeliger · 2024-05-20T15:23:02Z

L1Trigger/TrackTrigger/data/clf_GBDT_newbin_disp_xgb.onnx

This cannot be committed to CMSSW. It's fine here I guess, but in a PR to CMSSW all non-code files need to be committed to the data repositories.

… selection producer, and added divide by 0 checks

… code formatting

…ng input track collection

…ting

ryanm124 · 2024-07-17T09:11:16Z

The track selection module has broken the truth association. I suspect it's changing the addresses of the tracks which breaks the mapping between the tracks and tracking particles. Can I move the cuts back to my displaced vertex producer but have the cut values be input from the python config? Otherwise, I need someone to help get this working

…ed bug in track truth association which affected BDT training boolean

ryanm124 · 2024-08-28T08:15:24Z

I've removed the track selection module and moved the cut values to the python configuration in my last commit. The code is working and should be ready to merge

ryanm124 added 4 commits May 8, 2024 14:14

Adding Displaced Vertexing Producer

0255f1a

Modifying track quality MVA inputs for displaced tracks

24413dc

Reverting changes to prompt track quality BDT, adding fix to MVA scor…

acde98b

…e, and updating displaced vertex BDT

Adding newest BDT model file and displaced vertexing analyzer

0dde2bd

cgsavard reviewed May 14, 2024

View reviewed changes

Chriisbrown reviewed May 15, 2024

View reviewed changes

cgsavard reviewed May 15, 2024

View reviewed changes

Adding back tanL to prompt MVA and adding check for extended tracking…

dbf8525

… when setting track quality score

cgsavard reviewed May 17, 2024

View reviewed changes

Adding option to run displaced vertexing in ntupler maker, reverting …

d95ffe9

…changes to ntuple maker params, removing BDT requirement on displaced track quality

aloeliger reviewed May 20, 2024

View reviewed changes

aloeliger added Phase-2 Pertains to phase-2 development Algorithm Development labels May 20, 2024

ryanm124 mentioned this pull request May 22, 2024

Adding displaced track quality model file cms-data/L1Trigger-TrackTrigger#4

Open

ryanm124 mentioned this pull request May 22, 2024

Adding displaced vertexing GBDT cms-data/L1Trigger-L1TTrackMatch#2

Open

ryanm124 added 6 commits May 28, 2024 13:21

Removing Displaced Vertexing Analyzer macro

ca83bc8

Changing BDT model to slimmed version, moved track selection to track…

63cdfcf

… selection producer, and added divide by 0 checks

Adding check for number of mother partucles in ntuple maker and fixed…

93ea36d

… code formatting

Removing unused class methods

0786d1f

Changing track selection producer collection output name and correcti…

8f39115

…ng input track collection

Adding displaced vertexing track selector to schedule and code format…

34584c5

…ting

Removing track selection module and moving cuts to python config. Fix…

bff29e9

…ed bug in track truth association which affected BDT training boolean

	~DisplacedVertexProducer() override {}
	~DisplacedVertexProducer() override = default;

GBDT-Based Displaced Vertex Producer #1257

Are you sure you want to change the base?

GBDT-Based Displaced Vertex Producer #1257

Conversation

ryanm124 commented May 14, 2024 • edited Loading

triggerDoctor commented May 14, 2024

triggerDoctor commented May 14, 2024

triggerDoctor commented May 14, 2024

triggerDoctor commented May 14, 2024

triggerDoctor commented May 14, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Chriisbrown May 15, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

triggerDoctor commented May 20, 2024

triggerDoctor commented May 20, 2024

triggerDoctor commented May 20, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ryanm124 commented Jul 17, 2024

ryanm124 commented Aug 28, 2024

ryanm124 commented May 14, 2024 •

edited

Loading

Chriisbrown May 15, 2024 •

edited

Loading