MVAUtils::BDT Class Reference

Simplified Boosted Regression Tree, support TMVA, lgbm, and xgboost. More...

#include <BDT.h>

Collaboration diagram for MVAUtils::BDT:

Public Member Functions
	BDT (TTree *tree)
	Constructor.
	BDT (std::unique_ptr< IForest > forest)
	BDT ()=delete
	BDT (const BDT &)=delete
BDT &	operator= (const BDT &)=delete
	BDT (BDT &&)=default
	default move ctor, move assignment and dtor
BDT &	operator= (BDT &&)=default
	~BDT ()=default
unsigned int	GetNTrees () const
	Number of trees in the whole forest.
int	GetNVars () const
	Number of variables expected in the inputs.
float	GetOffset () const
	Get the offset to the whole forest.
float	GetResponse (const std::vector< float > &values) const
	Get response of the forest, for regression.
float	GetResponse (const std::vector< float * > &pointers) const
float	GetResponse () const
float	GetClassification (const std::vector< float > &values) const
	Get response of the forest, for classification.
float	GetClassification (const std::vector< float * > &pointers) const
float	GetClassification () const
float	GetGradBoostMVA (const std::vector< float > &values) const
float	GetGradBoostMVA (const std::vector< float * > &pointers) const
std::vector< float >	GetMultiResponse (const std::vector< float > &values, unsigned int numClasses) const
	Get response of the forest, for multiclassification (e.g.
std::vector< float >	GetMultiResponse (const std::vector< float * > &pointers, unsigned int numClasses) const
std::vector< float >	GetMultiResponse (unsigned int numClasses) const
std::vector< float >	GetValues () const
	Return the values corresponding to m_pointers (or an empty vector).
const std::vector< float * > &	GetPointers () const
	Return stored pointers (which are used by methods with no args).
void	SetPointers (const std::vector< float * > &pointers)
	Set the stored pointers so that one can use methods with no args.
TTree *	WriteTree (TString name="BDT") const
	Return a TTree representing the BDT: each entry is a binary tree, each element of the vectors is a node.
void	PrintForest () const
	for debugging, print out tree or forest to stdout
void	PrintTree (unsigned int itree) const
float	GetTreeResponse (const std::vector< float > &values, MVAUtils::index_t index) const
	for debugging, return the response of a sigle tree given the index of its top node
float	GetTreeResponse (const std::vector< float * > &pointers, MVAUtils::index_t index) const

Private Attributes
std::unique_ptr< IForest >	m_forest
	the implementation of the forest, doing the hard work
std::vector< float * >	m_pointers
	where vars to cut on can be set (but can also be passed)

Detailed Description

Simplified Boosted Regression Tree, support TMVA, lgbm, and xgboost.

Holds a forest (vector of top nodes of each decision tree) and a constant offset or set of weights (not always used).

Each entry of the TTree represents a binary tree and each element of the vectors stored in the TTree represent a node.

The response can be evaluated from a vector of floats or a vector of pointers (to avoid creating vectors at each call) which can be stored internally (m_pointers)

Definition at line 34 of file Reconstruction/MVAUtils/MVAUtils/BDT.h.

Constructor & Destructor Documentation

BDT() [1/5]

BDT::BDT ( ::TTree * tree )

explicit

Constructor.

c-tor from TTree

The input tree must be created with convertLGBMToRootTree.py (for lgbm training) or with convertXGBoostToRootTree.py (for xgboost training) or with convertXmlToRootTree (for tmva training)

Definition at line 57 of file BDT.cxx.

{
  // at runtime decide which flavour of BDT we need to build
  // the information is coming from the title of the TTree
  
  if(!tree){
      throw std::runtime_error("nullptr to a TTree passed ");
  }
  tree->SetCacheSize(0); // Avoid unnecessary memory allocations
  std::map<std::string, std::string> options = parseOptions(tree->GetTitle());
  std::string creator = get_default_string_map(options, std::string("creator"));
  if (creator == "lgbm")
  {
    std::string node_type = get_default_string_map (options, std::string("node_type"));
    if (node_type == "lgbm") {
      m_forest = std::make_unique<ForestLGBM>(tree);
    } else if (node_type == "lgbm_simple") {
      m_forest = std::make_unique<ForestLGBMSimple>(
        tree); // this do not support nan as inputs
    } else {
      throw std::runtime_error(
        "the title of the input tree is misformatted: cannot understand which "
        "BDT implementation to use");
    }
  } else if (creator == "xgboost") {
    // this do support nan as inputs
    m_forest = std::make_unique<ForestXGBoost>(tree);
  } else {
    // default for compatibility: old TTree (based on TMVA) don't have a special title
    m_forest = std::make_unique<ForestTMVA>(tree);
  }
}

BDT() [2/5]

MVAUtils::BDT::BDT ( std::unique_ptr< IForest > forest )

inlineexplicit

Definition at line 44 of file Reconstruction/MVAUtils/MVAUtils/BDT.h.

                                               : 
      m_forest(std::move(forest)){
      }

BDT() [3/5]

MVAUtils::BDT::BDT ( )

delete

BDT() [4/5]

MVAUtils::BDT::BDT ( const BDT & )

delete

BDT() [5/5]

MVAUtils::BDT::BDT ( BDT && )

default

default move ctor, move assignment and dtor

default move ctor, move assignment

~BDT()

MVAUtils::BDT::~BDT ( )

default

Member Function Documentation

GetClassification() [1/3]

float MVAUtils::BDT::GetClassification

(

)

const

GetClassification() [2/3]

float MVAUtils::BDT::GetClassification

(

const std::vector< float * > &

pointers

)

const

GetClassification() [3/3]

float MVAUtils::BDT::GetClassification

(

const std::vector< float > &

values

)

const

Get response of the forest, for classification.

GetGradBoostMVA() [1/2]

float MVAUtils::BDT::GetGradBoostMVA

(

const std::vector< float * > &

pointers

)

const

GetGradBoostMVA() [2/2]

float MVAUtils::BDT::GetGradBoostMVA

(

const std::vector< float > &

values

)

const

GetMultiResponse() [1/3]

std::vector< float > MVAUtils::BDT::GetMultiResponse	(	const std::vector< float * > &	pointers,
		unsigned int	numClasses ) const

GetMultiResponse() [2/3]

std::vector< float > MVAUtils::BDT::GetMultiResponse	(	const std::vector< float > &	values,
		unsigned int	numClasses ) const

Get response of the forest, for multiclassification (e.g.

b-tagging)

GetMultiResponse() [3/3]

std::vector< float > MVAUtils::BDT::GetMultiResponse

(

unsigned int

numClasses

)

const

GetNTrees()

unsigned int MVAUtils::BDT::GetNTrees

(

)

const

Number of trees in the whole forest.

GetNVars()

int MVAUtils::BDT::GetNVars

(

)

const

Number of variables expected in the inputs.

GetOffset()

float MVAUtils::BDT::GetOffset

(

)

const

Get the offset to the whole forest.

GetPointers()

const std::vector< float * > & MVAUtils::BDT::GetPointers

(

)

const

Return stored pointers (which are used by methods with no args).

GetResponse() [1/3]

float MVAUtils::BDT::GetResponse

(

)

const

GetResponse() [2/3]

float MVAUtils::BDT::GetResponse

(

const std::vector< float * > &

pointers

)

const

GetResponse() [3/3]

float MVAUtils::BDT::GetResponse

(

const std::vector< float > &

values

)

const

Get response of the forest, for regression.

GetTreeResponse() [1/2]

float MVAUtils::BDT::GetTreeResponse	(	const std::vector< float * > &	pointers,
		MVAUtils::index_t	index ) const

GetTreeResponse() [2/2]

float MVAUtils::BDT::GetTreeResponse	(	const std::vector< float > &	values,
		MVAUtils::index_t	index ) const

for debugging, return the response of a sigle tree given the index of its top node

GetValues()

std::vector< float > MVAUtils::BDT::GetValues

(

)

const

Return the values corresponding to m_pointers (or an empty vector).

operator=() [1/2]

BDT & MVAUtils::BDT::operator= ( BDT && )

default

operator=() [2/2]

BDT & MVAUtils::BDT::operator= ( const BDT & )

delete

PrintForest()

void BDT::PrintForest

(

)

const

for debugging, print out tree or forest to stdout

Definition at line 92 of file BDT.cxx.

{ m_forest->PrintForest(); }

PrintTree()

void BDT::PrintTree

(

unsigned int

itree

)

const

Definition at line 93 of file BDT.cxx.

{ m_forest->PrintTree(itree); }

SetPointers()

void MVAUtils::BDT::SetPointers

(

const std::vector< float * > &

pointers

)

Set the stored pointers so that one can use methods with no args.

WriteTree()

TTree * BDT::WriteTree

(

TString

name = "BDT"

)

const

Return a TTree representing the BDT: each entry is a binary tree, each element of the vectors is a node.

Definition at line 91 of file BDT.cxx.

{ return m_forest->WriteTree(std::move(name)); }

Member Data Documentation

m_forest

std::unique_ptr<IForest> MVAUtils::BDT::m_forest

private

the implementation of the forest, doing the hard work

Definition at line 107 of file Reconstruction/MVAUtils/MVAUtils/BDT.h.

m_pointers

std::vector<float*> MVAUtils::BDT::m_pointers

private

where vars to cut on can be set (but can also be passed)

Definition at line 108 of file Reconstruction/MVAUtils/MVAUtils/BDT.h.

---

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

• Reconstruction/MVAUtils/MVAUtils/BDT.h
• BDT.cxx