Stores all required information for the ECLChargedPIDMVA for a phasespace category. More...

#include <ECLChargedPIDMVAWeights.h>

Inheritance diagram for ECLChargedPIDPhasespaceCategory:

Public Types
enum class	MVAResponseTransformMode : unsigned int { c_LogTransform = 0 , c_LogTransformSingle = 1 , c_GaussianTransform = 2 , c_DecorrelationTransform = 3 , c_DirectMVAResponse = 4 , c_LogMVAResponse = 5 }
	Enum of implemented transformations which can be applied to the MVA response. More...

Public Member Functions
	ECLChargedPIDPhasespaceCategory ()
	Default constructor, necessary for ROOT to stream the object.

	ECLChargedPIDPhasespaceCategory (const std::string weightfilePath, const MVAResponseTransformMode &mvaResponeTransformMode, const std::unordered_map< unsigned int, unsigned int > &mvaIndexForHypothesis)
	Useful constructor.

	~ECLChargedPIDPhasespaceCategory ()
	Destructor.

const std::string	getSerialisedWeight () const
	Getter for serialised weightfile.

void	setSerialisedWeight (std::string weight)
	Setter for serialised weightfile.

MVAResponseTransformMode	getTransformMode () const
	Getter for the MVA transform mode.

const TF1 *	getPDF (const unsigned int iMVAResponse, const unsigned int hypoPDG) const
	Getter for pdfs.

const TH1F *	getCDF (const unsigned int iMVAResponse, const int hypoPDG) const
	Gets the cdf for the hypothesis pdg for a given response value.

const std::vector< float > *	getDecorrelationMatrix (const int hypoPDG) const
	Gets the decorrelation matrix for a given particle hypothesis.

void	setCDFs (std::vector< std::unordered_map< unsigned int, TH1F > > cdfs)
	Set the cdfs.

void	setPDFs (std::vector< std::unordered_map< unsigned int, TF1 > > &pdfs)
	Set the pdfs.

void	setDecorrelationMatrixMap (std::unordered_map< unsigned int, std::vector< float > > decorrelationMatrices)
	Set the decorrelation matrices.

void	setlogTransformOffset (const float &offset)
	Set the offset used in the log transformation to be consistent with the offset used when generating the p.d.f.s.

float	getLogTransformOffset () const
	Getter for the log transform offset.

void	setTemperature (const float &temperature)
	Set the temperature parameter used to calibrate the MVA.

float	getTemperature () const
	Getter for the temperature.

void	setMaxPossibleResponseValue (const float &offset)
	Set the max possible response value, used in log transformation of the responses.

float	getMaxPossibleResponseValue () const
	Get the max possible response value, used in log transformation of the responses.

unsigned int	getMVAIndexForHypothesis (const unsigned int hypoPDG) const
	Maps a charged stable pdg code to an index of the MVA response.

Private Member Functions
	ClassDef (ECLChargedPIDPhasespaceCategory, 1)
	ClassDef.

Private Attributes
TParameter< float >	m_log_transform_offset
	Small offset to avoid mva response values of 1.0 being log transformed to NaN.

TParameter< float >	m_max_possible_response_value
	Max possible value of the mva response.

TParameter< float >	m_temperature
	calibration factor for MVA responses.

std::string	m_weight
	Serialsed MVA weightfile.

MVAResponseTransformMode	m_mvaResponseTransformMode
	Stores which transformation mode to apply to the mva responses.

std::vector< std::unordered_map< unsigned int, TF1 > >	m_pdfs
	A vector of unodered maps.

std::unordered_map< unsigned int, unsigned int >	m_mvaIndexForHypothesis
	Unordered map of abs(pdg_code) for the 6 charged stable hypotheses to index of the MVA response vector.

std::vector< std::unordered_map< unsigned int, TH1F > >	m_cdfs
	CDFs for each mva return value for each hypothesis.

std::unordered_map< unsigned int, std::vector< float > >	m_decorrelationMatrices
	Decorrelation matrices.

Detailed Description

Stores all required information for the ECLChargedPIDMVA for a phasespace category.

This includes:

MVA weightfiles for multiclass MVA.
MVAResponseTransform mode detailing which transformations will be applied to the MVA response.
unordered_map mapping a particle hypothesis to the output index of a MVA.
(Optional) TF1 p.d.fs for each charged particle hypothesis for each mva output variable.
(Optional) TH1F for each charged particle hypothesis for each mva output variable for gaussianisation.
(Optional) vector of floats (flattened square matrix) for potential linear decorrelation of the gaussian transformed mva response variables.

Definition at line 129 of file ECLChargedPIDMVAWeights.h.

Member Enumeration Documentation

◆ MVAResponseTransformMode

enum class MVAResponseTransformMode : unsigned int

strong

Enum of implemented transformations which can be applied to the MVA response.

Enumerator
c_LogTransform	Log transform the mva responses. And take the likelihood as the product of likelihoods from all mva responses.
c_LogTransformSingle	Log transform the mva responses. Take the likelihood from only the mva response for the hypothesis.
c_GaussianTransform	Gaussian transform of the log transformed mva response.
c_DecorrelationTransform	Decorrelation transform of the gaussian transformed mva responses.
c_DirectMVAResponse	Directly take the MVA response as the logL, useful if we train neural nets to learn the logL.
c_LogMVAResponse	Take the log of the MVA response. Essentially undoes the softmax function.

Definition at line 133 of file ECLChargedPIDMVAWeights.h.

                                        : unsigned int {
      c_LogTransform = 0,
      c_LogTransformSingle = 1,
      c_GaussianTransform = 2,
      c_DecorrelationTransform = 3,
      c_DirectMVAResponse = 4,
      c_LogMVAResponse = 5
    };

Constructor & Destructor Documentation

◆ ECLChargedPIDPhasespaceCategory() [1/2]

ECLChargedPIDPhasespaceCategory ( )

inline

Default constructor, necessary for ROOT to stream the object.

Definition at line 151 of file ECLChargedPIDMVAWeights.h.

                                      :
      m_log_transform_offset("logTransformOffset", 1e-15),
      m_max_possible_response_value("maxPossibleResponseValue", 1.0),
      m_temperature("temperature", 1.0)
    {};

◆ ECLChargedPIDPhasespaceCategory() [2/2]

ECLChargedPIDPhasespaceCategory	(	const std::string	weightfilePath,
		const MVAResponseTransformMode &	mvaResponeTransformMode,
		const std::unordered_map< unsigned int, unsigned int > &	mvaIndexForHypothesis )

inline

Useful constructor.

Parameters

weightfilePath	path to the MVA weightfile for this phasespace category.
mvaResponeTransformMode	mva response transform mode booked for this phasespace.
mvaIndexForHypothesis	unordered_map mapping hypothesis to index of mva response. Useful if we exclude a class from the MVA training but want to use the likelihood of a different particle. For example using the proton pdf for a deuteron.

Definition at line 164 of file ECLChargedPIDMVAWeights.h.

                                                                                                               :
      m_log_transform_offset("logTransformOffset", 1e-15),
      m_max_possible_response_value("maxPossibleResponseValue", 1.0),
      m_temperature("temperature", 1.0)
    {
      // Load and serialize the MVA::Weightfile object into a string for storage in the database,
      // otherwise there are issues w/ dictionary generation for the payload class...
      Belle2::MVA::Weightfile weightfile;
      if (weightfilePath.ends_with(".root")) {
        weightfile = Belle2::MVA::Weightfile::loadFromROOTFile(weightfilePath);
      } else if (weightfilePath.ends_with(".xml")) {
        weightfile = Belle2::MVA::Weightfile::loadFromXMLFile(weightfilePath);
      } else {
        B2WARNING("Unknown file extension for file: " << weightfilePath << ", fallback to xml...");
        weightfile = Belle2::MVA::Weightfile::loadFromXMLFile(weightfilePath);
      }
      std::stringstream ss;
      Belle2::MVA::Weightfile::saveToStream(weightfile, ss);
 
      // store
      m_weight = ss.str();
      m_mvaResponseTransformMode = mvaResponeTransformMode;
      m_mvaIndexForHypothesis = mvaIndexForHypothesis;
    }

◆ ~ECLChargedPIDPhasespaceCategory()

~ECLChargedPIDPhasespaceCategory ( )

inline

Destructor.

Definition at line 194 of file ECLChargedPIDMVAWeights.h.

194{};

Member Function Documentation

◆ getCDF()

const TH1F * getCDF	(	const unsigned int	iMVAResponse,
		const int	hypoPDG ) const

inline

Gets the cdf for the hypothesis pdg for a given response value.

Parameters

iMVAResponse	index of MVA response.
hypoPDG	hypothesis pdg.

Definition at line 226 of file ECLChargedPIDMVAWeights.h.

    {
      return &m_cdfs.at(iMVAResponse).at(hypoPDG);
    }

◆ getDecorrelationMatrix()

const std::vector< float > * getDecorrelationMatrix ( const int hypoPDG ) const

inline

Gets the decorrelation matrix for a given particle hypothesis.

Parameters

hypoPDG hypothesis pdg.

Definition at line 235 of file ECLChargedPIDMVAWeights.h.

    {
      return &m_decorrelationMatrices.at(hypoPDG);
    }

◆ getLogTransformOffset()

float getLogTransformOffset ( ) const

inline

Getter for the log transform offset.

Definition at line 274 of file ECLChargedPIDMVAWeights.h.

    {
      return m_log_transform_offset.GetVal();
    }

◆ getMaxPossibleResponseValue()

float getMaxPossibleResponseValue ( ) const

inline

Get the max possible response value, used in log transformation of the responses.

Definition at line 306 of file ECLChargedPIDMVAWeights.h.

    {
      return m_max_possible_response_value.GetVal();
    }

◆ getMVAIndexForHypothesis()

unsigned int getMVAIndexForHypothesis ( const unsigned int hypoPDG ) const

inline

Maps a charged stable pdg code to an index of the MVA response.

In general this is a one-to-one mapping however in cases where we do not include all six stable charged particles in the MVA training we may have a many-to-one mapping. For example if we take the proton response value also for deuterons.

Definition at line 317 of file ECLChargedPIDMVAWeights.h.

    {
      return m_mvaIndexForHypothesis.at(hypoPDG);
    }

◆ getPDF()

const TF1 * getPDF	(	const unsigned int	iMVAResponse,
		const unsigned int	hypoPDG ) const

inline

Getter for pdfs.

Parameters

iMVAResponse	index of MVA response.
hypoPDG	hypothesis pdg.

Definition at line 216 of file ECLChargedPIDMVAWeights.h.

    {
      return &m_pdfs.at(iMVAResponse).at(hypoPDG);
    }

◆ getSerialisedWeight()

const std::string getSerialisedWeight ( ) const

inline

Getter for serialised weightfile.

Definition at line 199 of file ECLChargedPIDMVAWeights.h.

199{return m_weight;}

◆ getTemperature()

float getTemperature ( ) const

inline

Getter for the temperature.

Definition at line 290 of file ECLChargedPIDMVAWeights.h.

    {
      return m_temperature.GetVal();
    }

◆ getTransformMode()

MVAResponseTransformMode getTransformMode ( ) const

inline

Getter for the MVA transform mode.

Definition at line 209 of file ECLChargedPIDMVAWeights.h.

209{return m_mvaResponseTransformMode;}

◆ setCDFs()

void setCDFs ( std::vector< std::unordered_map< unsigned int, TH1F > > cdfs )

inline

Set the cdfs.

Parameters

cdfs	vector of map of cdfs to be stored in the payload.

Definition at line 244 of file ECLChargedPIDMVAWeights.h.

244{m_cdfs = cdfs;}

◆ setDecorrelationMatrixMap()

void setDecorrelationMatrixMap ( std::unordered_map< unsigned int, std::vector< float > > decorrelationMatrices )

inline

Set the decorrelation matrices.

Parameters

decorrelationMatrices map of decorrelation matrices to be stored in the payload.

Definition at line 258 of file ECLChargedPIDMVAWeights.h.

    {
      m_decorrelationMatrices = decorrelationMatrices;
    }

◆ setlogTransformOffset()

void setlogTransformOffset ( const float & offset )

inline

Set the offset used in the log transformation to be consistent with the offset used when generating the p.d.f.s.

Definition at line 266 of file ECLChargedPIDMVAWeights.h.

    {
      m_log_transform_offset.SetVal(offset);
    }

◆ setMaxPossibleResponseValue()

void setMaxPossibleResponseValue ( const float & offset )

inline

Set the max possible response value, used in log transformation of the responses.

Definition at line 298 of file ECLChargedPIDMVAWeights.h.

    {
      m_max_possible_response_value.SetVal(offset);
    }

◆ setPDFs()

void setPDFs ( std::vector< std::unordered_map< unsigned int, TF1 > > & pdfs )

inline

Set the pdfs.

Parameters

pdfs	vector of map of pdfs to be stored in the payload.

Definition at line 251 of file ECLChargedPIDMVAWeights.h.

251{m_pdfs = pdfs;}

◆ setSerialisedWeight()

void setSerialisedWeight ( std::string weight )

inline

Setter for serialised weightfile.

Definition at line 204 of file ECLChargedPIDMVAWeights.h.

204{ m_weight = weight; }

◆ setTemperature()

void setTemperature ( const float & temperature )

inline

Set the temperature parameter used to calibrate the MVA.

Definition at line 282 of file ECLChargedPIDMVAWeights.h.

    {
      m_temperature.SetVal(temperature);
    }

Member Data Documentation

◆ m_cdfs

std::vector<std::unordered_map<unsigned int, TH1F> > m_cdfs

private

CDFs for each mva return value for each hypothesis.

The N vector elements correspond to the N MVA return values. The unordered map maps the hypothesis pdg values to their matching TH1F cdfs which can be used for a gaussianisation.

Definition at line 355 of file ECLChargedPIDMVAWeights.h.

◆ m_decorrelationMatrices

std::unordered_map<unsigned int, std::vector<float> > m_decorrelationMatrices

private

Decorrelation matrices.

To be used (optionally) after gaussianisation. The unordered map maps the hypothesis pdg values to their matching linearised decorrelation matrix.

Definition at line 361 of file ECLChargedPIDMVAWeights.h.

◆ m_log_transform_offset

TParameter<float> m_log_transform_offset

private

Small offset to avoid mva response values of 1.0 being log transformed to NaN.

Definition at line 324 of file ECLChargedPIDMVAWeights.h.

◆ m_max_possible_response_value

TParameter<float> m_max_possible_response_value

private

Max possible value of the mva response.

Used in the log transformation.

Definition at line 325 of file ECLChargedPIDMVAWeights.h.

◆ m_mvaIndexForHypothesis

std::unordered_map<unsigned int, unsigned int> m_mvaIndexForHypothesis

private

Unordered map of abs(pdg_code) for the 6 charged stable hypotheses to index of the MVA response vector.

Needed if we do not train with all 6 species to map several to the same MVA response value.

Definition at line 348 of file ECLChargedPIDMVAWeights.h.

◆ m_mvaResponseTransformMode

MVAResponseTransformMode m_mvaResponseTransformMode

private

Stores which transformation mode to apply to the mva responses.

Definition at line 336 of file ECLChargedPIDMVAWeights.h.

◆ m_pdfs

std::vector<std::unordered_map<unsigned int, TF1> > m_pdfs

private

A vector of unodered maps.

The vector corresponds to the return values of the MVA, one for each class correspondonding to charged stable particles considered by the MVA. In general this is the full six charged stable particles {e, mu, pi, K, p, d}. The unordered map maps the hypothesis pdg values to their matching TF1 pdfs from which the likelihood will be taken.

Definition at line 342 of file ECLChargedPIDMVAWeights.h.

◆ m_temperature

TParameter<float> m_temperature

private

calibration factor for MVA responses.

Follows arXiv:1706.04599

Definition at line 326 of file ECLChargedPIDMVAWeights.h.

◆ m_weight

std::string m_weight

private

Serialsed MVA weightfile.

Definition at line 331 of file ECLChargedPIDMVAWeights.h.

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

ecl/dbobjects/include/ECLChargedPIDMVAWeights.h

Public Types

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Member Enumeration Documentation

◆ MVAResponseTransformMode

Constructor & Destructor Documentation

◆ ECLChargedPIDPhasespaceCategory() [1/2]

◆ ECLChargedPIDPhasespaceCategory() [2/2]

◆ ~ECLChargedPIDPhasespaceCategory()

Member Function Documentation

◆ getCDF()

◆ getDecorrelationMatrix()

◆ getLogTransformOffset()

◆ getMaxPossibleResponseValue()

◆ getMVAIndexForHypothesis()

◆ getPDF()

◆ getSerialisedWeight()

◆ getTemperature()

◆ getTransformMode()

◆ setCDFs()

◆ setDecorrelationMatrixMap()

◆ setlogTransformOffset()

◆ setMaxPossibleResponseValue()

◆ setPDFs()

◆ setSerialisedWeight()

◆ setTemperature()

Member Data Documentation

◆ m_cdfs

◆ m_decorrelationMatrices

◆ m_log_transform_offset

◆ m_max_possible_response_value

◆ m_mvaIndexForHypothesis

◆ m_mvaResponseTransformMode

◆ m_pdfs

◆ m_temperature

◆ m_weight