BaseMVAValueMapProducer< T > Class Template Reference

#include <PhysicsTools/PatAlgos/plugins/BaseMVAValueMapProducer.cc>

Inheritance diagram for BaseMVAValueMapProducer< T >:

Public Member Functions
	BaseMVAValueMapProducer (const edm::ParameterSet &iConfig)
void	setValue (const std::string var, float val)
	~BaseMVAValueMapProducer () override
Public Member Functions inherited from edm::stream::EDProducer<>
	EDProducer ()=default
bool	hasAbilityToProduceInLumis () const final
bool	hasAbilityToProduceInRuns () const final

Static Public Member Functions
static void	fillDescriptions (edm::ConfigurationDescriptions &descriptions)
static edm::ParameterSetDescription	getDescription ()

Private Member Functions
void	beginStream (edm::StreamID) override
void	endStream () override
virtual void	fillAdditionalVariables (const T &)
void	produce (edm::Event &, const edm::EventSetup &) override
virtual void	readAdditionalCollections (edm::Event &, const edm::EventSetup &)
	to be implemented in derived classes, filling values for additional variables More...

Private Attributes
std::string	backend_
bool	batch_eval_
std::vector< std::pair< std::string, StringObjectFunction< T, true > > >	funcs_
tensorflow::GraphDef *	graph_
std::string	inputTensorName_
bool	isClassifier_
std::string	name_
bool	onnx_
std::unique_ptr< cms::Ort::ONNXRuntime >	ort_
std::vector< StringObjectFunction< std::vector< float > > >	output_formulas_
std::vector< std::string >	output_names_
std::string	outputTensorName_
std::map< std::string, size_t >	positions_
TMVA::Reader *	reader_
tensorflow::Session *	session_
std::string	singleThreadPool_
edm::EDGetTokenT< edm::View< T > >	src_
bool	tf_
bool	tmva_
std::vector< float >	values_
std::vector< std::string >	variablesOrder_
std::string	weightfilename_

Additional Inherited Members
Public Types inherited from edm::stream::EDProducer<>
typedef CacheContexts< T... >	CacheTypes
typedef CacheTypes::GlobalCache	GlobalCache
typedef AbilityChecker< T... >	HasAbility
typedef CacheTypes::LuminosityBlockCache	LuminosityBlockCache
typedef LuminosityBlockContextT< LuminosityBlockCache, RunCache, GlobalCache >	LuminosityBlockContext
typedef CacheTypes::LuminosityBlockSummaryCache	LuminosityBlockSummaryCache
typedef CacheTypes::RunCache	RunCache
typedef RunContextT< RunCache, GlobalCache >	RunContext
typedef CacheTypes::RunSummaryCache	RunSummaryCache

Detailed Description

template<typename T>
class BaseMVAValueMapProducer< T >

Description: [one line class summary]

Implementation: [Notes on implementation]

Definition at line 57 of file BaseMVAValueMapProducer.h.

Constructor & Destructor Documentation

template<typename T>

BaseMVAValueMapProducer< T >::BaseMVAValueMapProducer ( const edm::ParameterSet &  iConfig )

inlineexplicit

Definition at line 59 of file BaseMVAValueMapProducer.h.

      : src_(consumes<edm::View<T>>(iConfig.getParameter<edm::InputTag>("src"))),
        variablesOrder_(iConfig.getParameter<std::vector<std::string>>("variablesOrder")),
        name_(iConfig.getParameter<std::string>("name")),
        backend_(iConfig.getParameter<std::string>("backend")),
        weightfilename_(iConfig.getParameter<edm::FileInPath>("weightFile").fullPath()),
        isClassifier_(iConfig.getParameter<bool>("isClassifier")),
        tmva_(backend_ == "TMVA"),
        tf_(backend_ == "TF"),
        onnx_(backend_ == "ONNX"),
        batch_eval_(iConfig.getParameter<bool>("batch_eval")) {
    if (tmva_)
      reader_ = new TMVA::Reader();
    edm::ParameterSet const& varsPSet = iConfig.getParameter<edm::ParameterSet>("variables");
    for (const std::string& vname : varsPSet.getParameterNamesForType<std::string>()) {
      funcs_.emplace_back(
          std::pair<std::string, StringObjectFunction<T, true>>(vname, varsPSet.getParameter<std::string>(vname)));
    }

    values_.resize(variablesOrder_.size());
    size_t i = 0;
    for (const auto& v : variablesOrder_) {
      positions_[v] = i;
      if (tmva_)
        reader_->AddVariable(v, (&values_.front()) + i);
      i++;
    }
    //      reader_.BookMVA(name_,iConfig.getParameter<edm::FileInPath>("weightFile").fullPath() );
    if (tmva_) {
      reco::details::loadTMVAWeights(reader_, name_, weightfilename_);
    } else if (tf_) {
      tensorflow::setLogging("3");
      graph_ = tensorflow::loadGraphDef(weightfilename_);
      size_t nThreads = iConfig.getParameter<unsigned int>("nThreads");
      session_ = tensorflow::createSession(graph_, nThreads);
    } else if (onnx_) {
      ort_ = std::make_unique<cms::Ort::ONNXRuntime>(weightfilename_);
    } else {
      throw cms::Exception("ConfigError") << "Only 'TF', 'ONNX' and 'TMVA' backends are supported\n";
    }
    if (tf_ || onnx_) {
      inputTensorName_ = iConfig.getParameter<std::string>("inputTensorName");
      outputTensorName_ = iConfig.getParameter<std::string>("outputTensorName");
      output_names_ = iConfig.getParameter<std::vector<std::string>>("outputNames");
      for (const auto& s : iConfig.getParameter<std::vector<std::string>>("outputFormulas")) {
        output_formulas_.push_back(StringObjectFunction<std::vector<float>>(s));
      }
    }
    if (tmva_)
      produces<edm::ValueMap<float>>();
    else {
      for (const auto& n : output_names_) {
        produces<edm::ValueMap<float>>(n);
      }
    }
  }

template<typename T>

BaseMVAValueMapProducer< T >::~BaseMVAValueMapProducer ( )

inlineoverride

Definition at line 115 of file BaseMVAValueMapProducer.h.

{}

Member Function Documentation

template<typename T>

void BaseMVAValueMapProducer< T >::beginStream ( edm::StreamID  )

inlineoverrideprivate

Definition at line 126 of file BaseMVAValueMapProducer.h.

{};

template<typename T>

void BaseMVAValueMapProducer< T >::endStream ( )

inlineoverrideprivate

Definition at line 128 of file BaseMVAValueMapProducer.h.

{};

template<typename T>

virtual void BaseMVAValueMapProducer< T >::fillAdditionalVariables ( const T &  )

inlineprivatevirtual

Reimplemented in BJetEnergyRegressionMVA.

Definition at line 132 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< T >::produce().

{}

template<typename T >

void BaseMVAValueMapProducer< T >::fillDescriptions ( edm::ConfigurationDescriptions &  descriptions )

static

Definition at line 283 of file BaseMVAValueMapProducer.h.

References edm::ConfigurationDescriptions::add(), BaseMVAValueMapProducer< T >::getDescription(), and AlCaHLTBitMon_QueryRunRegistry::string.

Referenced by BaseMVAValueMapProducer< pat::Jet >::setValue().

                                                                                            {
  edm::ParameterSetDescription desc = getDescription();
  std::string modname;
  if (typeid(T) == typeid(pat::Jet))
    modname += "Jet";
  else if (typeid(T) == typeid(pat::Muon))
    modname += "Muon";
  else if (typeid(T) == typeid(pat::Electron))
    modname += "Ele";
  modname += "BaseMVAValueMapProducer";
  descriptions.add(modname, desc);
}

template<typename T >

edm::ParameterSetDescription BaseMVAValueMapProducer< T >::getDescription ( )

static

Definition at line 257 of file BaseMVAValueMapProducer.h.

References edm::ParameterSetDescription::add(), edm::ParameterSetDescription::setAllowAnything(), edm::ParameterDescriptionNode::setComment(), AlCaHLTBitMon_QueryRunRegistry::string, and objects.autophobj::variables.

Referenced by BJetEnergyRegressionMVA::fillDescriptions(), BaseMVAValueMapProducer< T >::fillDescriptions(), and BaseMVAValueMapProducer< pat::Jet >::setValue().

                                                                      {
  edm::ParameterSetDescription desc;
  desc.add<edm::InputTag>("src")->setComment("input physics object collection");
  desc.add<std::vector<std::string>>("variablesOrder")->setComment("ordered list of MVA input variable names");
  desc.add<std::string>("name")->setComment("output score variable name");
  desc.add<bool>("isClassifier")->setComment("is a classifier discriminator");
  edm::ParameterSetDescription variables;
  variables.setAllowAnything();
  desc.add<edm::ParameterSetDescription>("variables", variables)->setComment("list of input variable definitions");
  desc.add<edm::FileInPath>("weightFile")->setComment("xml weight file");
  desc.add<std::string>("backend", "TMVA")->setComment("TMVA, TF or ONNX");
  desc.add<std::string>("inputTensorName", "")->setComment("Name of tensorflow input tensor in the model");
  desc.add<std::string>("outputTensorName", "")->setComment("Name of tensorflow output tensor in the model");
  desc.add<std::vector<std::string>>("outputNames", std::vector<std::string>())
      ->setComment("Names of the output values to be used in the output valuemap");
  desc.add<std::vector<std::string>>("outputFormulas", std::vector<std::string>())
      ->setComment("Formulas to be used to post process the output");
  desc.add<unsigned int>("nThreads", 1)->setComment("number of threads");
  desc.add<std::string>("singleThreadPool", "no_threads");
  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;
}

template<typename T >

void BaseMVAValueMapProducer< T >::produce ( edm::Event &  iEvent, const edm::EventSetup &  iSetup  )

overrideprivate

Definition at line 160 of file BaseMVAValueMapProducer.h.

References BaseMVAValueMapProducer< T >::batch_eval_, data, pat::helper::ParametrizationHelper::dimension(), edm::helper::Filler< Map >::fill(), BaseMVAValueMapProducer< T >::fillAdditionalVariables(), objects.autophobj::filler, BaseMVAValueMapProducer< T >::funcs_, edm::Event::getByToken(), mps_fire::i, haddnano::inputs, BaseMVAValueMapProducer< T >::inputTensorName_, edm::helper::Filler< Map >::insert(), createfilelist::int, BaseMVAValueMapProducer< T >::isClassifier_, gen::k, funct::m, eostools::move(), BaseMVAValueMapProducer< T >::name_, connectstrParser::o, BaseMVAValueMapProducer< T >::onnx_, BaseMVAValueMapProducer< T >::ort_, BaseMVAValueMapProducer< T >::output_formulas_, BaseMVAValueMapProducer< T >::output_names_, PatBasicFWLiteJetAnalyzer_Selector_cfg::outputs, BaseMVAValueMapProducer< T >::outputTensorName_, AlCaHLTBitMon_ParallelJobs::p, BaseMVAValueMapProducer< T >::positions_, edm::Event::put(), BaseMVAValueMapProducer< T >::readAdditionalCollections(), BaseMVAValueMapProducer< T >::reader_, tensorflow::run(), BaseMVAValueMapProducer< T >::session_, BaseMVAValueMapProducer< T >::setValue(), TrackRefitter_38T_cff::src, BaseMVAValueMapProducer< T >::src_, BaseMVAValueMapProducer< T >::tf_, BaseMVAValueMapProducer< T >::tmva_, findQualityFiles::v, and BaseMVAValueMapProducer< T >::values_.

Referenced by BaseMVAValueMapProducer< pat::Jet >::beginStream().

                                                                                      {
  edm::Handle<edm::View<T>> src;
  iEvent.getByToken(src_, src);
  readAdditionalCollections(iEvent, iSetup);
  std::vector<std::vector<float>> mvaOut((tmva_) ? 1 : output_names_.size());
  for (auto& v : mvaOut)
    v.reserve(src->size());

  if (batch_eval_) {
    if (!src->empty()) {
      std::vector<float> data;
      data.reserve(src->size() * positions_.size());
      for (auto const& o : *src) {
        for (auto const& p : funcs_) {
          setValue(p.first, p.second(o));
        }
        fillAdditionalVariables(o);
        data.insert(data.end(), values_.begin(), values_.end());
      }

      std::vector<float> outputs;
      if (tf_) {
        //currently support only one input sensor to reuse the TMVA like config
        tensorflow::TensorShape input_size{(long long int)src->size(), (long long int)positions_.size()};
        tensorflow::NamedTensorList input_tensors;
        input_tensors.resize(1);
        input_tensors[0] =
            tensorflow::NamedTensor(inputTensorName_, tensorflow::Tensor(tensorflow::DT_FLOAT, input_size));

        for (unsigned i = 0; i < data.size(); ++i) {
          input_tensors[0].second.flat<float>()(i) = data[i];
        }
        std::vector<tensorflow::Tensor> output_tensors;
        tensorflow::run(session_, input_tensors, {outputTensorName_}, &output_tensors);
        for (unsigned i = 0; i < output_tensors.at(0).NumElements(); ++i) {
          outputs.push_back(output_tensors.at(0).flat<float>()(i));
        }
      } else if (onnx_) {
        cms::Ort::FloatArrays inputs{data};
        outputs = ort_->run({inputTensorName_}, inputs, {}, {outputTensorName_}, src->size())[0];
      }

      const unsigned outdim = outputs.size() / src->size();
      for (unsigned i = 0; i < src->size(); ++i) {
        std::vector<float> tmpOut(outputs.begin() + i * outdim, outputs.begin() + (i + 1) * outdim);
        for (size_t k = 0; k < output_names_.size(); k++) {
          mvaOut[k].push_back(output_formulas_[k](tmpOut));
        }
      }
    }
  } else {
    for (auto const& o : *src) {
      for (auto const& p : funcs_) {
        setValue(p.first, p.second(o));
      }
      fillAdditionalVariables(o);
      if (tmva_) {
        mvaOut[0].push_back(isClassifier_ ? reader_->EvaluateMVA(name_) : reader_->EvaluateRegression(name_)[0]);
      } else {
        std::vector<float> tmpOut;
        if (tf_) {
          //currently support only one input sensor to reuse the TMVA like config
          tensorflow::TensorShape input_size{1, (long long int)positions_.size()};
          tensorflow::NamedTensorList input_tensors;
          input_tensors.resize(1);
          input_tensors[0] =
              tensorflow::NamedTensor(inputTensorName_, tensorflow::Tensor(tensorflow::DT_FLOAT, input_size));
          for (size_t j = 0; j < values_.size(); j++) {
            input_tensors[0].second.matrix<float>()(0, j) = values_[j];
          }
          std::vector<tensorflow::Tensor> outputs;
          tensorflow::run(session_, input_tensors, {outputTensorName_}, &outputs);
          for (int k = 0; k < outputs.at(0).matrix<float>().dimension(1); k++)
            tmpOut.push_back(outputs.at(0).matrix<float>()(0, k));
        } else if (onnx_) {
          cms::Ort::FloatArrays inputs{values_};
          tmpOut = ort_->run({inputTensorName_}, inputs, {}, {outputTensorName_})[0];
        }

        for (size_t k = 0; k < output_names_.size(); k++)
          mvaOut[k].push_back(output_formulas_[k](tmpOut));
      }
    }
  }

  size_t k = 0;
  for (auto& m : mvaOut) {
    std::unique_ptr<edm::ValueMap<float>> mvaV(new edm::ValueMap<float>());
    edm::ValueMap<float>::Filler filler(*mvaV);
    filler.insert(src, m.begin(), m.end());
    filler.fill();
    iEvent.put(std::move(mvaV), (tmva_) ? "" : output_names_[k]);
    k++;
  }
}

template<typename T>

virtual void BaseMVAValueMapProducer< T >::readAdditionalCollections ( edm::Event &  , const edm::EventSetup &    )

inlineprivatevirtual

to be implemented in derived classes, filling values for additional variables

Reimplemented in BJetEnergyRegressionMVA.

Definition at line 131 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< T >::produce().

{}

template<typename T>

void BaseMVAValueMapProducer< T >::setValue ( const std::string  var, float  val  )

inline

Definition at line 117 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< T >::produce().

                                                {
    if (positions_.find(var) != positions_.end())
      values_[positions_[var]] = val;
  }

Member Data Documentation

template<typename T>

std::string BaseMVAValueMapProducer< T >::backend_

private

Definition at line 146 of file BaseMVAValueMapProducer.h.

template<typename T>

bool BaseMVAValueMapProducer< T >::batch_eval_

private

Definition at line 152 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::vector<std::pair<std::string, StringObjectFunction<T, true> > > BaseMVAValueMapProducer< T >::funcs_

private

Definition at line 136 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

tensorflow::GraphDef* BaseMVAValueMapProducer< T >::graph_

private

Definition at line 140 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer().

template<typename T>

std::string BaseMVAValueMapProducer< T >::inputTensorName_

private

Definition at line 153 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

bool BaseMVAValueMapProducer< T >::isClassifier_

private

Definition at line 148 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::string BaseMVAValueMapProducer< T >::name_

private

Definition at line 145 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

bool BaseMVAValueMapProducer< T >::onnx_

private

Definition at line 151 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::unique_ptr<cms::Ort::ONNXRuntime> BaseMVAValueMapProducer< T >::ort_

private

Definition at line 143 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::vector<StringObjectFunction<std::vector<float> > > BaseMVAValueMapProducer< T >::output_formulas_

private

Definition at line 156 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::vector<std::string> BaseMVAValueMapProducer< T >::output_names_

private

Definition at line 155 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::string BaseMVAValueMapProducer< T >::outputTensorName_

private

Definition at line 154 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::map<std::string, size_t> BaseMVAValueMapProducer< T >::positions_

private

Definition at line 135 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), BaseMVAValueMapProducer< T >::produce(), and BaseMVAValueMapProducer< pat::Jet >::setValue().

template<typename T>

TMVA::Reader* BaseMVAValueMapProducer< T >::reader_

private

Definition at line 139 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

tensorflow::Session* BaseMVAValueMapProducer< T >::session_

private

Definition at line 141 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::string BaseMVAValueMapProducer< T >::singleThreadPool_

private

Definition at line 142 of file BaseMVAValueMapProducer.h.

template<typename T>

edm::EDGetTokenT<edm::View<T> > BaseMVAValueMapProducer< T >::src_

private

Definition at line 134 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< T >::produce().

template<typename T>

bool BaseMVAValueMapProducer< T >::tf_

private

Definition at line 150 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

bool BaseMVAValueMapProducer< T >::tmva_

private

Definition at line 149 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), and BaseMVAValueMapProducer< T >::produce().

template<typename T>

std::vector<float> BaseMVAValueMapProducer< T >::values_

private

Definition at line 138 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer(), BaseMVAValueMapProducer< T >::produce(), and BaseMVAValueMapProducer< pat::Jet >::setValue().

template<typename T>

std::vector<std::string> BaseMVAValueMapProducer< T >::variablesOrder_

private

Definition at line 137 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer().

template<typename T>

std::string BaseMVAValueMapProducer< T >::weightfilename_

private

Definition at line 147 of file BaseMVAValueMapProducer.h.

Referenced by BaseMVAValueMapProducer< pat::Jet >::BaseMVAValueMapProducer().