PIDNeuralNetwork Class Reference

Class to call PID neural network. More...

#include <PIDNeuralNetwork.h>

Collaboration diagram for PIDNeuralNetwork:

Public Member Functions
	PIDNeuralNetwork ()
	Constructor with default paramenter-set name.
	PIDNeuralNetwork (const std::string &parameterName)
	Constructor with given paramenter-set name.
	PIDNeuralNetwork (PIDNeuralNetwork &&other)
	Move constructor.
std::map< int, double >	predict (std::vector< float > inputs) const
	Predict neural-network output for all implemented hypotheses using the given inputs. More...
size_t	getInputSize () const
	Get number of inputs. More...
size_t	getInputIndex (const std::string &name) const
	Get input names. More...
const std::vector< std::string > &	getInputNames () const
	Get input names. More...
const std::vector< int > &	getOutputSpeciesPdg () const
	Get the list of pdg codes of species hypotheses, for which the network predicts the probability in the order defined by the network output layer. More...
const std::string &	getPIDNeuralNetworkParametersName () const
	Get the name of the used neural network. More...
bool	hasPdgCode (const int pdg, const bool throwException=false) const

Private Member Functions
void	loadParametersFromDB ()
	Load neural-network parameters with name m_pidNeuralNetworkParametersName from the conditions data base.

Private Attributes
std::string	m_pidNeuralNetworkParametersName = "PIDNeuralNetworkParameters"
	name of the parameter set
std::unique_ptr< DBObjPtr< PIDNeuralNetworkParameters > >	m_pidNeuralNetworkParametersDB
	db object for the parameter set
std::unique_ptr< const fdeep::model >	m_model
	frugally-deep neural network

Detailed Description

Class to call PID neural network.

Definition at line 31 of file PIDNeuralNetwork.h.

Member Function Documentation

getInputIndex()

size_t getInputIndex ( const std::string &  name ) const

inline

Get input names.

Returns

const std::vector<std::string>& input names

Definition at line 85 of file PIDNeuralNetwork.h.

{ return (*m_pidNeuralNetworkParametersDB)->getInputIndex(name); }

getInputNames()

const std::vector<std::string>& getInputNames ( ) const

inline

Get input names.

Returns

const std::vector<std::string>& input names

Definition at line 91 of file PIDNeuralNetwork.h.

getInputSize()

size_t getInputSize ( ) const

inline

Get number of inputs.

Returns

size_t number of inputs

Definition at line 79 of file PIDNeuralNetwork.h.

getOutputSpeciesPdg()

const std::vector<int>& getOutputSpeciesPdg ( ) const

inline

Get the list of pdg codes of species hypotheses, for which the network predicts the probability in the order defined by the network output layer.

Returns

const std::vector<int>& list of pdg codes of hypotheses predicted by the network

Definition at line 98 of file PIDNeuralNetwork.h.

getPIDNeuralNetworkParametersName()

const std::string& getPIDNeuralNetworkParametersName ( ) const

inline

Get the name of the used neural network.

Returns

const std::string& name of the used neural network

Definition at line 104 of file PIDNeuralNetwork.h.

hasPdgCode()

bool hasPdgCode ( const int  pdg, const bool  throwException = false  ) const

inline

Parameters

pdg	pdg code of hypothesis
throwException	throw exception if pdg code is not predicted

Returns

true if neural network predicts probability for the given hypothesis

false if neural network does not predict probability for the given hypothesis

Definition at line 112 of file PIDNeuralNetwork.h.

predict()

std::map< int, double > predict

(

std::vector< float >

inputs

)

const

Predict neural-network output for all implemented hypotheses using the given inputs.

Parameters

inputs

set of input variables

Returns

double probability of the asked particle-species hypothesis

std::map<int, double> map of pdg codes to the corresponding probabilities

Definition at line 32 of file PIDNeuralNetwork.cc.

{
 
  // apply cuts, ie. overwrite certain input values with index `inputSetIndex` with the value `setValue`
  // if the input with index `inputCutIndex` is in the range (`rangeStart`, `rangeEnd`)
  for (auto const& inputToCut : (*m_pidNeuralNetworkParametersDB)->getInputsToCut()) {
    const auto [inputSetIndex, inputCutIndex, rangeStart, rangeEnd, setValue] = inputToCut;
    if (!std::isnan(rangeStart) and !std::isnan(rangeEnd)) {
      if (input[inputCutIndex] >= rangeStart and input[inputCutIndex] <= rangeEnd)
        input[inputSetIndex] = setValue;
    } else if (!std::isnan(rangeStart)) {
      if (input[inputCutIndex] >= rangeStart)
        input[inputSetIndex] = setValue;
    } else if (!std::isnan(rangeEnd)) {
      if (input[inputCutIndex] <= rangeEnd)
        input[inputSetIndex] = setValue;
    }
  }
 
  // Normalize inputs
  // By definition, this has to be done after applying the cuts and before handling missing information
  for (size_t i = 0; i < input.size(); ++i) {
    input[i] = (input[i] - (*m_pidNeuralNetworkParametersDB)->getMeanValues()[i]) /
               (*m_pidNeuralNetworkParametersDB)->getStandardDeviations()[i];
  }
 
  // handle missing information
  for (auto const& index_value : (*m_pidNeuralNetworkParametersDB)->getHandleMissingInputs()) {
    const auto [index, value] = index_value;
    if (std::isnan(input[index])) input[index] = value;
  }
 
  // apply neural network
  const auto inputFdeep = fdeep::tensor(fdeep::tensor_shape(input.size()), input);
  const auto result = m_model->predict({inputFdeep});
 
  std::map<int, double> probabilities;
  for (const auto pdgCode : getOutputSpeciesPdg()) {
    const int outputIndex = (*m_pidNeuralNetworkParametersDB)->pdg2OutputIndex(pdgCode);
    probabilities[pdgCode] = result.front().get(fdeep::tensor_pos(outputIndex));
  }
  return probabilities;
}

---

The documentation for this class was generated from the following files:

• analysis/utility/include/PIDNeuralNetwork.h
• analysis/utility/src/PIDNeuralNetwork.cc