Skip to content

Add function to refine FastSim DeepJet discriminators #40550

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wolfmor wants to merge 3 commits into from
Closed

Add function to refine FastSim DeepJet discriminators #40550

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
6c27d59
Add function to refine FastSim btagDeepFlav jet variables
wolfmor Jan 11, 2023
d2ff394
Add option to disable ONNX runtime graph optimization and apply it fo…
wolfmor Jan 16, 2023
cc24284
Do not pass disableONNXGraphOpt option as reference
wolfmor Jan 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
58 changes: 58 additions & 0 deletions PhysicsTools/NanoAOD/python/jetsAK4_CHS_cff.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -350,6 +350,64 @@ def nanoAOD_addDeepInfoAK4CHS(process,addDeepBTag,addDeepFlavour):
## DeepInfoAK4CHS:End
#################################################

# ML-based FastSim refinement
from Configuration.Eras.Modifier_fastSim_cff import fastSim
def nanoAOD_refineFastSim_bTagDeepFlav(process):

fastSim.toModify( process.jetTable.variables,
btagDeepFlavBunrefined = process.jetTable.variables.btagDeepFlavB.clone(),
btagDeepFlavCvBunrefined = process.jetTable.variables.btagDeepFlavCvB.clone(),
btagDeepFlavCvLunrefined = process.jetTable.variables.btagDeepFlavCvL.clone(),
btagDeepFlavQGunrefined = process.jetTable.variables.btagDeepFlavQG.clone(),
)

fastSim.toModify( process.jetTable.variables,
btagDeepFlavB = None,
btagDeepFlavCvB = None,
btagDeepFlavCvL = None,
btagDeepFlavQG = None,
)

fastSim.toModify( process.jetTable.externalVariables,
btagDeepFlavB = ExtVar(cms.InputTag("btagDeepFlavRefineNN:btagDeepFlavBrefined"), float, doc="DeepJet b+bb+lepb tag discriminator", precision=10),
btagDeepFlavCvB = ExtVar(cms.InputTag("btagDeepFlavRefineNN:btagDeepFlavCvBrefined"), float, doc="DeepJet c vs b+bb+lepb discriminator", precision=10),
btagDeepFlavCvL = ExtVar(cms.InputTag("btagDeepFlavRefineNN:btagDeepFlavCvLrefined"), float, doc="DeepJet c vs uds+g discriminator", precision=10),
btagDeepFlavQG = ExtVar(cms.InputTag("btagDeepFlavRefineNN:btagDeepFlavQGrefined"), float, doc="DeepJet g vs uds discriminator", precision=10),
)

process.btagDeepFlavRefineNN= cms.EDProducer("JetBaseMVAValueMapProducer",
backend = cms.string("ONNX"),
batch_eval = cms.bool(True),
disableONNXGraphOpt = cms.bool(True),

src = cms.InputTag("linkedObjects","jets"),

weightFile=cms.FileInPath("PhysicsTools/NanoAOD/data/btagDeepFlavRefineNN_CHS.onnx"),
name = cms.string("btagDeepFlavRefineNN"),

isClassifier = cms.bool(False),
variablesOrder = cms.vstring(["GenJet_pt","GenJet_eta","Jet_hadronFlavour",
"Jet_btagDeepFlavB","Jet_btagDeepFlavCvB","Jet_btagDeepFlavCvL","Jet_btagDeepFlavQG"]),
variables = cms.PSet(
GenJet_pt = cms.string("?genJetFwdRef().backRef().isNonnull()?genJetFwdRef().backRef().pt():pt"),
GenJet_eta = cms.string("?genJetFwdRef().backRef().isNonnull()?genJetFwdRef().backRef().eta():eta"),
Jet_hadronFlavour = cms.string("hadronFlavour()"),
Jet_btagDeepFlavB = cms.string("bDiscriminator('pfDeepFlavourJetTags:probb')+bDiscriminator('pfDeepFlavourJetTags:probbb')+bDiscriminator('pfDeepFlavourJetTags:problepb')"),
Jet_btagDeepFlavCvB = cms.string("?(bDiscriminator('pfDeepFlavourJetTags:probc')+bDiscriminator('pfDeepFlavourJetTags:probb')+bDiscriminator('pfDeepFlavourJetTags:probbb')+bDiscriminator('pfDeepFlavourJetTags:problepb'))>0?bDiscriminator('pfDeepFlavourJetTags:probc')/(bDiscriminator('pfDeepFlavourJetTags:probc')+bDiscriminator('pfDeepFlavourJetTags:probb')+bDiscriminator('pfDeepFlavourJetTags:probbb')+bDiscriminator('pfDeepFlavourJetTags:problepb')):-1"),
Jet_btagDeepFlavCvL = cms.string("?(bDiscriminator('pfDeepFlavourJetTags:probc')+bDiscriminator('pfDeepFlavourJetTags:probuds')+bDiscriminator('pfDeepFlavourJetTags:probg'))>0?bDiscriminator('pfDeepFlavourJetTags:probc')/(bDiscriminator('pfDeepFlavourJetTags:probc')+bDiscriminator('pfDeepFlavourJetTags:probuds')+bDiscriminator('pfDeepFlavourJetTags:probg')):-1"),
Jet_btagDeepFlavQG = cms.string("?(bDiscriminator('pfDeepFlavourJetTags:probg')+bDiscriminator('pfDeepFlavourJetTags:probuds'))>0?bDiscriminator('pfDeepFlavourJetTags:probg')/(bDiscriminator('pfDeepFlavourJetTags:probg')+bDiscriminator('pfDeepFlavourJetTags:probuds')):-1"),
),
inputTensorName = cms.string("input"),
outputTensorName = cms.string("output"),
outputNames = cms.vstring(["btagDeepFlavBrefined","btagDeepFlavCvBrefined","btagDeepFlavCvLrefined","btagDeepFlavQGrefined"]),
outputFormulas = cms.vstring(["at(0)","at(1)","at(2)","at(3)"]),
)

fastSim.toModify(process.jetTablesTask, process.jetTablesTask.add(process.btagDeepFlavRefineNN))

return process


################################################################################
# JETS FOR MET type1
################################################################################
Expand Down
15 changes: 12 additions & 3 deletions PhysicsTools/PatAlgos/interface/BaseMVAValueMapProducer.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,12 +55,19 @@

class BaseMVACache {
public:
BaseMVACache(const std::string& model_path, const std::string& backend) {
BaseMVACache(const std::string& model_path, const std::string& backend, const bool disableONNXGraphOpt) {
if (backend == "TF") {
graph_.reset(tensorflow::loadGraphDef(model_path));
tf_session_ = tensorflow::createSession(graph_.get());
} else if (backend == "ONNX") {
ort_ = std::make_unique<cms::Ort::ONNXRuntime>(model_path);
if (disableONNXGraphOpt) {
Ort::SessionOptions sess_opts;
sess_opts = cms::Ort::ONNXRuntime::defaultSessionOptions();
sess_opts.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_DISABLE_ALL);
ort_ = std::make_unique<cms::Ort::ONNXRuntime>(model_path, &sess_opts);
} else {
ort_ = std::make_unique<cms::Ort::ONNXRuntime>(model_path);
}
}
}
~BaseMVACache() { tensorflow::closeSession(tf_session_); }
Expand Down Expand Up @@ -270,7 +277,8 @@ void BaseMVAValueMapProducer<T>::produce(edm::Event& iEvent, const edm::EventSet
template <typename T>
std::unique_ptr<BaseMVACache> BaseMVAValueMapProducer<T>::initializeGlobalCache(const edm::ParameterSet& cfg) {
return std::make_unique<BaseMVACache>(cfg.getParameter<edm::FileInPath>("weightFile").fullPath(),
cfg.getParameter<std::string>("backend"));
cfg.getParameter<std::string>("backend"),
cfg.getParameter<bool>("disableONNXGraphOpt"));
}

template <typename T>
Expand All @@ -295,6 +303,7 @@ edm::ParameterSetDescription BaseMVAValueMapProducer<T>::getDescription() {
desc.add<std::vector<std::string>>("outputFormulas", std::vector<std::string>())
->setComment("Formulas to be used to post process the output");
desc.add<bool>("batch_eval", false)->setComment("Run inference in batch instead of per-object");
desc.add<bool>("disableONNXGraphOpt", false)->setComment("Disable ONNX runtime graph optimization");

return desc;
}
Expand Down