Interface to N-dimensional functional representations (THn and TMVA) More...

Functions
template<typename T >
vector< T >	add_to_vector (vector< T > &z, T v)
	generic variadic function - to get it from boost in the future More...

template<typename T , typename... Args>
vector< T >	add_to_vector (vector< T > &z, T v, Args...args)

template<typename T , typename... Args>
vector< T >	make_vector (T v, Args...args)
	Variadic function to create an vector (boost implementation ) More...

Double_t	GetInterpolationLinear (THn his, Double_t xyz, Int_t verbose)
	Linear interpolation of the bin content. More...

Double_t	GetDeltaInterpolationLinear (THn his, Double_t xyz, Int_t dIndex, Int_t verbose)
	Linear interpolation of the numerical derivative dV/dx (V(i+1)-V(i-1))/(2(delta) More...

Double_t	GetInterpolationLinear (Int_t index, Double_t *xyz, Int_t verbose)

Double_t	GetDeltaInterpolationLinear (Int_t index, Double_t *xyz, Int_t dIndex, Int_t verbose)

Double_t	GetInterpolationLinear (const char name, Double_t xyz, Int_t verbose)

template<typename T , typename... Args>
T	EvalTHnLinear (int id, T v, Args...args)

void	registerDefaultMVAMethods ()
	map of registered TMVA regression methods More...

void	registerFactory (std::string factory, std::string content)
	example registering default methods () More...

void	registerMethod (std::string method, std::string content, TMVA::Types::EMVA id)

Int_t	FitMVAClassification (const char output, const char inputTrees, const char cuts, const char variableList, const char methodList, const char factoryString="")

Int_t	FitMVARegression (const char output, TTree tree, const char varFit, TCut cut, const char variables, const char methods, const char factoryString="")
	MVA Classification. More...

TMVA::MethodBase *	LoadMVAReader (Int_t id, const char inputFile, const char method, const char *dir)
	new MVA Regression More...

Int_t	LoadMVAReaderArray (Int_t id, const char inputFile, const char methodMask, const char *dirMask)

Int_t	AppendMethodToArray (Int_t index, TMVA::MethodBase *method)
	Register TMVA method to the array at index Not assumed to be used by users. More...

Double_t	EvalMVAStatArray (int id, int statType, vector< float > point)
	Append method into array of methods - used e.g for bootstrap statistics. More...

template<typename T , typename... Args>
T	EvalMVA (int id, T v, Args...args)
	Evaluate statistic. More...

template<typename T , typename... Args>
T	EvalMVAClasification (int id, T v, Args...args)
	variadic function evaluating MVA More...

template<typename T , typename... Args>
T	EvalMVAStat (int id, int statType, T v, Args...args)
	variadic function evaluating MVA More...

Variables
Int_t	fVerbose =0

std::map< int, THn * >	hnMapArrayInt

std::map< std::string, THn * >	hnMapArrayName

map< int, TMVA::MethodBase * >	readerMethodBase
	TMVA interface. More...

map< int, TObjArray * >	readerMethodBaseArray
	map of registered TMVA::MethodBase More...

map< std::string, std::string >	regressionMethodSetting
	map of registered array of TMVA::MethodBase - used to define TMVA statistics (Mean, Median, RMS, quantile) More...

map< std::string, std::string >	FactorySetting
	map of registered TMVA regression methods More...

map< std::string, TMVA::Types::EMVA >	regressionMethodID
	map of registered TMVA regression methods More...

Detailed Description

Interface to N-dimensional functional representations (THn and TMVA)

Authors: Marian Ivanov maria.nosp@m.n.iv.nosp@m.anov@.nosp@m.cern.nosp@m..ch see example usage in

Function Documentation

template<typename T >

vector< T > AliNDFunctionInterface::add_to_vector	(	vector< T > &	z,
		T	v
	)

generic variadic function - to get it from boost in the future

Helper function to create std vector in variadic function.

verbosity

Definition at line 58 of file AliNDFunctionInterface.h.

template<typename T , typename... Args>

vector< T > AliNDFunctionInterface::add_to_vector	(	vector< T > &	z,
		T	v,
		Args...	args
	)

Definition at line 59 of file AliNDFunctionInterface.h.

Int_t AliNDFunctionInterface::AppendMethodToArray	(	Int_t	index,
		TMVA::MethodBase *	method
	)

Register TMVA method to the array at index Not assumed to be used by users.

Parameters

index	- registered array index
method	- pointer to the method

Returns

Definition at line 471 of file AliNDFunctionInterface.cxx.

Referenced by LoadMVAReaderArray(), and registerMethod().

template<typename T , typename... Args>

T AliNDFunctionInterface::EvalMVA	(	int	id,
		T	v,
		Args...	args
	)

Evaluate statistic.

Variadic function to evaluate MVA regression method registered using method ID.

Template Parameters

T
Args

Parameters

id
v
args

Returns

Example usage:

usage in TTreeFormula. e.g calculate and visualize second derivative of regression
MVAInput->Draw("AliNDFunctionInterface::EvalMVA(0,fraction,Z,0)-(AliNDFunctionInterface::EvalMVA(0,fraction,Z-1,0)+AliNDFunctionInterface::EvalMVA(0,fraction,Z+1,0))*0.5:Z","Z>2&&fraction>5","");

some optional verbosity needed

Definition at line 114 of file AliNDFunctionInterface.h.

Referenced by registerMethod().

template<typename T , typename... Args>

T AliNDFunctionInterface::EvalMVAClasification	(	int	id,
		T	v,
		Args...	args
	)

variadic function evaluating MVA

some optional verbosity needed

Definition at line 122 of file AliNDFunctionInterface.h.

Referenced by registerMethod().

template<typename T , typename... Args>

T AliNDFunctionInterface::EvalMVAStat	(	int	id,
		int	statType,
		T	v,
		Args...	args
	)

variadic function evaluating MVA

Template variadic function - Evaluate statistic on top of array (readerMethodBaseArray;) of MVA methods To use the method - array o TMVA methods should be registered before using LoadMVAReaderArray or AppendMethodToArray

Template Parameters

T	- template type - be default float
Args	-

Parameters

id	- id of the registered array to evaluate
statType	- type of statistic (0-mean, 1-median, 2-rms)
v	-
args

Returns

Example usage:

compare mean and median statistic of array 2
MVAInput->Draw("AliNDFunctionInterface::EvalMVAStat(2,0,interactionRate, bz0, qmaxQASum, qmaxQASumR):AliNDFunctionInterface::EvalMVAStat(2,1,interactionRate, bz0, qmaxQASum, qmaxQASumR)","run==QA.EVS.run","");

variadic function evaluating MVA array stat

Definition at line 147 of file AliNDFunctionInterface.h.

Referenced by registerMethod().

Double_t AliNDFunctionInterface::EvalMVAStatArray	(	int	id,
		int	statType,
		vector< float >	point
	)

Append method into array of methods - used e.g for bootstrap statistics.

Return statistic variable using array of MVA methods

Parameters

id	- id of the TMVA method array
statType	- type of the statistic (mean, median, rms, TODO cumulant)
point	- point to evaluate

Returns: - requested statistic
TODO:

TODO - for the moment in evaluation we assume only one variable

using the DNN or MLP we should start to support vectors TODO - extend statistic list

some optional verbosity needed

Definition at line 491 of file AliNDFunctionInterface.cxx.

Referenced by EvalMVAStat(), and registerMethod().

template<typename T , typename... Args>

T AliNDFunctionInterface::EvalTHnLinear	(	int	id,
		T	v,
		Args...	args
	)

Variadic function to linearly interpolate THn

Template Parameters

T	- template type
Args	- list of arguments

Parameters

id	- id of the function (e.g using hash of the name)
v
args

Returns: interpolated value
Example usage:

interpolation of 3D histogram at point xyz
THn has to be registered in map hnMapArrayInt[id] before
value = AliNDFunctionInterface::EvalTHnLinear(0,x,y,z);

Definition at line 91 of file AliNDFunctionInterface.h.

Referenced by GetInterpolationLinear().

Int_t AliNDFunctionInterface::FitMVAClassification	(	const char *	output,
		const char *	inputTrees,
		const char *	cuts,
		const char *	variables,
		const char *	methods,
		const char *	factoryString = `""`
	)

FitMVAClassification - TODO add documentation and test

Parameters

output
inputTrees
cuts
variables
methods
factoryString

Returns

Write ascii weight files to root file

Definition at line 138 of file AliNDFunctionInterface.cxx.

Referenced by registerMethod().

Int_t AliNDFunctionInterface::FitMVARegression	(	const char *	output,
		TTree *	tree,
		const char *	varFit,
		TCut	cut,
		const char *	variables,
		const char *	methods,
		const char *	factoryString = `""`
	)

MVA Classification.

FitMVARegression wrapper = Do TMVA regression and save weights into output root file

Parameters

output	- output path <file>#directory/ output file is opened in update mode - in case key (directory) exist - fit should failed - data are not written (currently it failed)
tree	- input tree
varFit	- variable to fit (for the moment 1)
cut	- cut formula
variables	- : separated list of explanatory variables
methods	- : separated list of the regression methods (have to be registered before)
factoryString	- factory string (e.g specifying number of training and test samples). Factory string to be registered before. If empty default used - Fit parameters will be saved in the output destination - return value 0 in case of success Example usage in test macro ( QAtrendingFitExample.C function makeMVAFits() and makeMVABootstrapMI() ), e.g:

TString output="TMVA_RegressionOutput.root#";
 for (Int_t iBoot=0; iBoot<nRegression; iBoot++) {
   AliNDFunctionInterface::FitMVARegression(output+"resolutionMIP"+iBoot,treeCache, "resolutionMIP", "interactionRate>0", "interactionRate:bz0:qmaxQASum:qmaxQASumR", "BDTRF25_8:BDTRF12_16:KNN", "");
   AliNDFunctionInterface::FitMVARegression(output+"meanMIPeleR"+iBoot,treeCache, "meanMIPeleR", "interactionRate>0", "interactionRate:bz0:qmaxQASum:qmaxQASumR", "BDTRF25_8:BDTRF12_16:KNN","");
   AliNDFunctionInterface::FitMVARegression(output+"tpcItsMatchA"+iBoot,treeCache, "tpcItsMatchA", "interactionRate>0", "interactionRate:bz0:qmaxQASum:qmaxQASumR", "BDTRF25_8:BDTRF12_16:KNN","");
 }

Algorithm:

0. Declare Factory
1.) Declare DataLoader

TODO check existence - validity of variable - in case of error - exit with error message

3.) Setup DataSet
4.) Book regression methods from the methods list
5.) Train all methods
6.) Write ascii weight files to root file
7.) Evaluate all MVAs using the set of test events /// TODO - make an option
8.) Evaluate and compare performance of all configured MVAs /// TODO - make an option
Save the output

Definition at line 272 of file AliNDFunctionInterface.cxx.

Referenced by makeMVABootstrapMI(), and registerMethod().

Double_t AliNDFunctionInterface::GetDeltaInterpolationLinear	(	THn *	his,
		Double_t *	xyz,
		Int_t	dIndex,
		Int_t	verbose
	)

Linear interpolation of the numerical derivative dV/dx (V(i+1)-V(i-1))/(2(delta)

Parameters

his
xyz
dIndex
verbose

Returns

Definition at line 85 of file AliNDFunctionInterface.cxx.

Referenced by GetDeltaInterpolationLinear().

Double_t AliNDFunctionInterface::GetDeltaInterpolationLinear	(	Int_t	index,
		Double_t *	xyz,
		Int_t	dIndex,
		Int_t	verbose
	)

Definition at line 31 of file AliNDFunctionInterface.h.

Double_t AliNDFunctionInterface::GetInterpolationLinear	(	THn *	his,
		Double_t *	xyz,
		Int_t	verbose
	)

Linear interpolation of the bin content.

Parameters

his	- input N-dimensional histogram to interpolate
xyz	- NDimensional point where to interpolate
verbose	- verbosity flag

Returns

Definition at line 52 of file AliNDFunctionInterface.cxx.

Referenced by EvalTHnLinear(), and GetInterpolationLinear().

Double_t AliNDFunctionInterface::GetInterpolationLinear	(	Int_t	index,
		Double_t *	xyz,
		Int_t	verbose
	)

Definition at line 30 of file AliNDFunctionInterface.h.

Double_t AliNDFunctionInterface::GetInterpolationLinear	(	const char *	name,
		Double_t *	xyz,
		Int_t	verbose
	)

Definition at line 34 of file AliNDFunctionInterface.h.

TMVA::MethodBase * AliNDFunctionInterface::LoadMVAReader	(	Int_t	id,
		const char *	inputFile,
		const char *	method,
		const char *	dir
	)

new MVA Regression

Load MVA regression object and register it in the map of available readers fr evaluation in TFormula Current ION of TMVA does not allow standard persistence in root file Way around: Writing - ASCII files are stored as TSting in the root file Reading - String written as ASCII file and read back by TMVA::Reader

Parameters

id	- regression method ID (using e.g. hash value)
inputFile	- input file (e.g TMVA_RegressionOutput.root)
method	- method name (e.g MLP)
dir	- directory usually coding regression variable with dir_ prefix (e.g. dir_meanMIPele)

Returns: - error code
TODOs:

TODO - ASK TMVA team to implement standard IO
TODO - ASK TMVA team to disable requirement to set variable addresses or find switch to make it

Example usage (see also QAtrendingFitExample.C loadMVAReaders()):

 void loadMVAReaders(){
  AliNDFunctionInterface::LoadMVAReader(0,"TMVA_RegressionOutput.root","BDTRF25_8","resolutionMIP0");
  AliNDFunctionInterface::LoadMVAReader(1,"TMVA_RegressionOutput.root","BDTRF12_16","resolutionMIP0");
  AliNDFunctionInterface::LoadMVAReader(2,"TMVA_RegressionOutput.root","KNN","resolutionMIP0");
}

TODO check existence - validity of variable - in case of error - exit with error message

write weight from the root file to txt files as it is expected by reader

Definition at line 371 of file AliNDFunctionInterface.cxx.

Referenced by LoadMVAReaderArray(), loadMVAReaders(), and registerMethod().

Int_t AliNDFunctionInterface::LoadMVAReaderArray	(	Int_t	id,
		const char *	inputFile,
		const char *	methodMask,
		const char *	dirMask
	)

Load array of the MVA reader and register it in the AliNDFunctionInterface method array maps ( readerMethodBaseArray) Method arrays could be later use for the TMVA array evaluation ()

Parameters

id
inputFile
methodMask
dirMask

Returns: error code - 0 mean OK

Example usage in (see also QAtrendingFitExample.C loadMVAReadersBootstrap())

void loadMVAReadersBootstrap() {
 AliNDFunctionInterface::LoadMVAReaderArray(0,"TMVA_RegressionOutput.root","BDTRF12_16",".*resolutionMIP");
 AliNDFunctionInterface::LoadMVAReaderArray(1,"TMVA_RegressionOutput.root","BDTRF25_8",".*resolutionMIP");
 AliNDFunctionInterface::LoadMVAReaderArray(2,"TMVA_RegressionOutput.root","KNN",".*resolutionMIP");
}