#include <TANNModeler.h>

Inheritance diagram for URANIE::Modeler::TANNModeler:

[legend]

Collaboration diagram for URANIE::Modeler::TANNModeler:

[legend]

Public Types
enum	EProblem { kRegression , kClassification }

enum	ENorm { kCR , kMinusOneOne , kZeroOne }

Public Member Functions
Constructor and Destructor
	TANNModeler (URANIE::DataServer::TDataServer tds, TString architecture, Double_t dratio=0.80, Option_t option="")
	Constructor with a dataserver.

virtual	~TANNModeler ()
	Default destructor.

Setter and Getter function
Int_t	getNInput ()
	Gets the number of input.

Int_t	getNHidden ()
	Gets the number of hidden neurons.

Int_t	getNOutput ()
	Gets the number of output.

Int_t	getNAPP ()
	Gets the number of pure training patterns ("APPrentissage")

Int_t	getNTest ()
	Gets the number of pure training patterns ("APPrentissage")

void	setWeights (ColumnVector weights)
	Sets the weigts.

ColumnVector	getWeights ()

Double_t	getWeight (Int_t ind)
	Returns the weight givent by the index.

Double_t	getMinValue (Int_t ind)

Double_t	getMaxValue (Int_t ind)

Double_t *	getdval ()

void	setValue (Int_t ind, Int_t &ncrt, Double_t *xy, Bool_t normalize=true)

void	setDataSet (URANIE::DataServer::TPatternsEventList *tel)

void	setMix (Bool_t bmix=kTRUE)

void	setFcnTol (Double_t dprec)

Double_t	getFcnTol ()

void	setNormalization (ENorm n)
	Sets the normalisation.

ENorm	getNormalization ()
	Get the normalisation.

EProblem	getProblem ()
	Get the problem type.

void	setSeed (Int_t ind=0)
	Init the seed.

Int_t	getSeed ()
	Returns the seed value.

const char *	getInputName (Int_t i)
	Return the name of input attributes indexed by i.

const char *	getOutputName ()
	Get the attribute name of the output.

Training
void	init ()

void	randomizeWeights ()
	randomize the weights

void	train (Int_t niter=10, Int_t nInit=10, Option_t *option="text", Bool_t useGPU=false)

Double_t	getEQM (Option_t *option="")
	Gets the EQM.

Double_t	getWeightDecay ()
	Get the weight decay parameter.

void	saveIfBetter (const ColumnVector &weights, Double_t eqm_app, Double_t eqm_tst)
	Save a vector of weight if the RMS on the test database is better than what we currently have.

Exports function
The exports function to save an Artificial Neural Networks in an external format C/C++ Fortran PMML The user fonction is the exportFunction method. tann->exportFunction("c++", "uranie_ann_flowrate.C","ANNflowrate"); tann->exportFunction("fortran", "uranie_ann_flowrate.f","ANNflowrate"); tann->exportFunction("pmml", sPMMLFile, sANNName,"new"); Todo: Add different format in the exportFunction method tann->exportFunction("C++,fortran,pmml", "uranie_ann_flowrate","ANNflowrate");
void	exportModelCplusplus (const char file="", const char name="", Option_t *option="") const
	The only user interface method.

void	exportModelFortran (std::ofstream *sourcefile) const

void	exportModelPython (std::ofstream *sourcefile) const
	Export the model in Python langage in a file (not yet implemented)

Printing Log
virtual void	printLog (Option_t *option="")

Public Member Functions inherited from URANIE::Modeler::TModeler
	TModeler (URANIE::DataServer::TDataServer tds, TString architecture, Option_t option="")
	Constructor with a dataserver.

	TModeler (URANIE::DataServer::TDataServer tds, const char varexpinput, const char varexpoutput, Option_t option="")
	Default constructor with input and output attributes.

virtual	~TModeler ()
	Default destructor.

TVectorD	getParameters ()
	Return the parameter vector of the model.

Double_t	getR2 () const
	Return the R2 quality.

const char *	getInputName (Int_t i) const
	Return the name of input attributes.

const char *	getOutputName (Int_t i=0)
	Return the name of output attribute.

Int_t	getNInput ()
	Gets the number of input.

Bool_t	getIntercept ()
	Gets the boolean attribut _bIntercept.

void	setDrawProgressBar (Bool_t bbool=kTRUE)
	Set the "draw progress bar" flag.

Bool_t	getDrawProgressBar ()
	Get the "draw progress bar" flag.

void	train (Option_t *option="text")

virtual void	exportFunction (const char lang, const char file="", const char name="", Option_t option="")
	Export the model in an external file with a specified langage.

virtual void	exportModelCplusplus (std::ofstream *sourcefile) const =0
	Export the model in C++ langage in a file.

virtual void	exportModelPMML (const char file="", const char name="", Option_t *option="") const =0
	Export the model in a PMML file.

void	setLog ()

void	unsetLog ()

void	changeLog ()

Bool_t	getLog ()

Protected Attributes
Int_t	_nSeed

Private Attributes
Int_t	_nInput
	! Seed for the random weight initialization

Int_t	_nHidden
	! The number of inputs

Int_t	_nHiddenLayer
	! The number of hidden neurons

Int_t	_nOutput
	! the number of hidden layer

Int_t	_nNeurons
	! The number of outputs

Int_t	_nApp
	! The number of neurons

Int_t	_nTest
	! Number of samples in the training set dedicated to back-propagation

Double_t	_dLearn
	! Number of samples in the training set dedicated to validation

Double_t	_dTest
	! The learning quality

Double_t	_dWeightDecay
	! The test quality

URANIE::DataServer::TPatternsEventList *	_elAll
	! The weight decay parameter - when used, favorize a smoother ANN

Bool_t	_bMix
	! The mix list of patterns

TString	_sArchi
	! True if data set must be mixed

TString	_sInput
	! The architecture, e.g. "x:y:z,3,yhat" specifying inputs, number of hidden neurons, and output

TString	_sHidden

TString	_sOutput

Int_t	_nWeight

ColumnVector	_vecWeights
	! The number of weights

TempSolution *	_tmpSolution

ColumnVector	_vecMinValues

ColumnVector	_vecMaxValues
	! Min values, used for normalization

TList *	_listOfAttributes
	! Max values, used for normalization

Parameters of algorithm
Double_t	_dFcnTol

ENorm	_nNormType
	! The function tolerance for the trust region algorithm

EProblem	_nProlemType
	! The normalized for input and output

Double_t *	_dval
	! The type of problem (Regression, Classification)

static const int	OPTIM_MAX_ITER = 10000

static const int	OPTIM_MAX_FEVAL = 50000

Bool_t	_blog
	Boolean for edit the log.

URANIE::DataServer::TDataServer *	_tds

UMLP *	_UMLP
	Pointeur vers un TDS.

Additional Inherited Members
Public Attributes inherited from URANIE::Modeler::TModeler
Bool_t	_blog
	Boolean for edit the log.

Bool_t	_bdrawProgressBar
	Boolean to know if the progress bar has to be drawn.

Bool_t	_bStoreYHat
	Boolean to specify if we add the \hat{} attribute in the TDS [default kTRUE].

URANIE::DataServer::TDataServer *	_tds

Int_t	_nS
	Pointer to a TDS.

TString	_sInputAttributes
	The string of input attributes (separated by the character ":".

TString	_sOutputAttributes
	The string of output attributes (separated by the character ":".

TString	_sIntermediateArchitecture
	The intermediate (between the first and last character ",") information.

TList *	_listOfInputAttributes
	The list of input Leaf.

TList *	_listOfOutputAttributes
	The list of output Leaf.

Int_t	_nX
	The number of input attributes.

Int_t	_nY
	The number of output attributes.

TString	_sFileName
	The name of the file when export the model without extension.

TString	_sFunctionName
	The name of the function when export the model.

TVectorD	_parameterValues
	The list of parameters.

Double_t	_dr2
	The R2 quality.

Bool_t	_bIntercept
	For certain models, add the intercept (Default is TRUE)

Member Enumeration Documentation

◆ ENorm

enum URANIE::Modeler::TANNModeler::ENorm

Enumerator
kCR
kMinusOneOne
kZeroOne

◆ EProblem

enum URANIE::Modeler::TANNModeler::EProblem

Enumerator
kRegression
kClassification

Constructor & Destructor Documentation

◆ TANNModeler()

URANIE::Modeler::TANNModeler::TANNModeler	(	URANIE::DataServer::TDataServer *	tds,
		TString	architecture,
		Double_t	dratio = `0.80`,
		Option_t *	option = `""`
	)

Constructor with a dataserver.

Referenced by ClassImp().

◆ ~TANNModeler()

virtual URANIE::Modeler::TANNModeler::~TANNModeler ( )

virtual

Default destructor.

Referenced by ClassImp().

Member Function Documentation

◆ exportModelCplusplus()

void URANIE::Modeler::TANNModeler::exportModelCplusplus	(	const char *	file = `""`,
		const char *	name = `""`,
		Option_t *	option = `""`
	)		const

The only user interface method.

Parameters

lang	(const char*) the keyword for the external format [c++\|fortran\|pmml]
file	(const char*) the filename. If empty, it is the name of the TDS with the good extension.[""]
name	(const char*) name of the function.
option	(const char) option parameter. NEW* : create a new PMML file UPDATE : Write the current ANN in the PMML file and if already exist an ANN in the PMML file, erase it. BACKUP : Write the current ANN in the PMML file and if already exist ANNs in the PMML file with modelname "name_I", rename the old with "name_I+1" Export the ANN model in a ROOT macro file

Referenced by ClassImp().

◆ exportModelFortran()

void URANIE::Modeler::TANNModeler::exportModelFortran ( std::ofstream * sourcefile ) const

virtual

Todo:: Write this method as a TModeler method ‍/ void exportModelCplusplus(std::ofstream * sourcefile) const; { cerr << " --- Method TANNModeler::exportModelCplusplus(ofstream * sourcefile) not yet implemented" << endl; cerr << " --- Then using the method TANNModeler::exportModelCplusplus(file, name, option)" << endl; } ;

! Export the TANNModeler object in a PMML File void exportModelPMML(const char* file = "", const char* name = "", Option_t option = "") const; ! Export the ANN model in a Fortran file void exportModelFortran(const char file, const char* name, Option_t option = "") ; /!

Todo:: Write this method as a TModeler method

Implements URANIE::Modeler::TModeler.

Referenced by ClassImp().

◆ exportModelPython()

void URANIE::Modeler::TANNModeler::exportModelPython ( std::ofstream * sourcefile ) const

inlinevirtual

Export the model in Python langage in a file (not yet implemented)

Warning: Don't use this method. Use the main method TModeler::exportFunction.

Parameters

sourcefile (ofstream *) the pointer to the file to export the modeler.

Implements URANIE::Modeler::TModeler.

◆ getdval()

Double_t * URANIE::Modeler::TANNModeler::getdval ( )

inline

References _dval.

Referenced by ClassImp().

◆ getEQM()

Double_t URANIE::Modeler::TANNModeler::getEQM ( Option_t * option = "" )

Gets the EQM.

Parameters

option ["learn|test"] give the Mean Square Error (MSE ) on the learn database ("learn") or on the test database ("test").

Referenced by ClassImp().

◆ getFcnTol()

Double_t URANIE::Modeler::TANNModeler::getFcnTol ( )

inline

References _dFcnTol.

Referenced by ClassImp().

◆ getInputName()

const char * URANIE::Modeler::TANNModeler::getInputName ( Int_t i )

inline

Return the name of input attributes indexed by i.

References _listOfAttributes.

◆ getMaxValue()

Double_t URANIE::Modeler::TANNModeler::getMaxValue ( Int_t ind )

inline

References _vecMaxValues.

Referenced by ClassImp().

◆ getMinValue()

Double_t URANIE::Modeler::TANNModeler::getMinValue ( Int_t ind )

inline

References _vecMinValues.

Referenced by ClassImp().

◆ getNAPP()

Int_t URANIE::Modeler::TANNModeler::getNAPP ( )

inline

Gets the number of pure training patterns ("APPrentissage")

References _nApp.

Referenced by ClassImp().

◆ getNHidden()

Int_t URANIE::Modeler::TANNModeler::getNHidden ( )

inline

Gets the number of hidden neurons.

References _nHidden.

Referenced by ClassImp().

◆ getNInput()

Int_t URANIE::Modeler::TANNModeler::getNInput ( )

inline

Gets the number of input.

References _nInput.

Referenced by ClassImp().

◆ getNormalization()

ENorm URANIE::Modeler::TANNModeler::getNormalization ( )

inline

Get the normalisation.

References _nNormType.

Referenced by ClassImp().

◆ getNOutput()

Int_t URANIE::Modeler::TANNModeler::getNOutput ( )

inline

Gets the number of output.

References _nOutput.

Referenced by ClassImp().

◆ getNTest()

Int_t URANIE::Modeler::TANNModeler::getNTest ( )

inline

Gets the number of pure training patterns ("APPrentissage")

References _nTest.

Referenced by ClassImp().

◆ getOutputName()

const char * URANIE::Modeler::TANNModeler::getOutputName ( )

inline

Get the attribute name of the output.

References _sOutput.

◆ getProblem()

EProblem URANIE::Modeler::TANNModeler::getProblem ( )

inline

Get the problem type.

References _nProlemType.

Referenced by ClassImp().

◆ getSeed()

Int_t URANIE::Modeler::TANNModeler::getSeed ( )

inline

Returns the seed value.

References _nSeed.

◆ getWeight()

Double_t URANIE::Modeler::TANNModeler::getWeight ( Int_t ind )

inline

Returns the weight givent by the index.

the index begins to 0 and ends to nW-1

References _vecWeights.

◆ getWeightDecay()

Double_t URANIE::Modeler::TANNModeler::getWeightDecay ( )

inline

Get the weight decay parameter.

References _dWeightDecay.

Referenced by ClassImp().

◆ getWeights()

ColumnVector URANIE::Modeler::TANNModeler::getWeights ( )

inline

References _vecWeights.

Referenced by ClassImp().

◆ init()

void URANIE::Modeler::TANNModeler::init ( )

Referenced by ClassImp().

◆ printLog()

virtual void URANIE::Modeler::TANNModeler::printLog ( Option_t * option = "" )

virtual

Reimplemented from URANIE::Modeler::TModeler.

Referenced by ClassImp().

◆ randomizeWeights()

void URANIE::Modeler::TANNModeler::randomizeWeights ( )

randomize the weights

Referenced by ClassImp().

◆ saveIfBetter()

void URANIE::Modeler::TANNModeler::saveIfBetter	(	const ColumnVector &	weights,
		Double_t	eqm_app,
		Double_t	eqm_tst
	)

Save a vector of weight if the RMS on the test database is better than what we currently have.

Referenced by ClassImp().

◆ setDataSet()

void URANIE::Modeler::TANNModeler::setDataSet ( URANIE::DataServer::TPatternsEventList * tel )

inline

References _elAll.

◆ setFcnTol()

void URANIE::Modeler::TANNModeler::setFcnTol ( Double_t dprec )

inline

References _dFcnTol.

◆ setMix()

void URANIE::Modeler::TANNModeler::setMix ( Bool_t bmix = kTRUE )

inline

References _bMix.

◆ setNormalization()

void URANIE::Modeler::TANNModeler::setNormalization ( ENorm n )

Sets the normalisation.

Referenced by ClassImp().

◆ setSeed()

void URANIE::Modeler::TANNModeler::setSeed ( Int_t ind = 0 )

Init the seed.

If the value is zero, the seed is initialized with the hour in milliseconds

Parameters

ind	(Int_t) the seed value

Referenced by ClassImp().

◆ setValue()

void URANIE::Modeler::TANNModeler::setValue	(	Int_t	ind,
		Int_t &	ncrt,
		Double_t *	xy,
		Bool_t	normalize = `true`
	)

Referenced by ClassImp().

◆ setWeights()

void URANIE::Modeler::TANNModeler::setWeights ( ColumnVector weights )

inline

Sets the weigts.

References _vecWeights.

Referenced by ClassImp().

◆ train()

void URANIE::Modeler::TANNModeler::train	(	Int_t	niter = `10`,
		Int_t	nInit = `10`,
		Option_t *	option = `"text"`,
		Bool_t	useGPU = `false`
	)

Launch the training of the ANN. This potentially uses the GPU if available.

Parameters

niter	number of times (a random permutation of) the test base will be presented for training
nInit	number of times the ANN is trained (from random start weights) with a given permutation of the test database

Referenced by ClassImp().

Member Data Documentation

◆ _blog

Bool_t URANIE::Modeler::TANNModeler::_blog

Boolean for edit the log.

Referenced by ClassImp().

◆ _bMix

Bool_t URANIE::Modeler::TANNModeler::_bMix

private

! The mix list of patterns

Referenced by ClassImp(), and setMix().

◆ _dFcnTol

Double_t URANIE::Modeler::TANNModeler::_dFcnTol

private

Referenced by ClassImp(), getFcnTol(), and setFcnTol().

◆ _dLearn

Double_t URANIE::Modeler::TANNModeler::_dLearn

private

! Number of samples in the training set dedicated to validation

Referenced by ClassImp().

◆ _dTest

Double_t URANIE::Modeler::TANNModeler::_dTest

private

! The learning quality

Referenced by ClassImp().

◆ _dval

Double_t* URANIE::Modeler::TANNModeler::_dval

private

! The type of problem (Regression, Classification)

Temporary array used during training

Referenced by ClassImp(), and getdval().

◆ _dWeightDecay

Double_t URANIE::Modeler::TANNModeler::_dWeightDecay

private

! The test quality

Referenced by ClassImp(), and getWeightDecay().

◆ _elAll

URANIE::DataServer::TPatternsEventList* URANIE::Modeler::TANNModeler::_elAll

private

! The weight decay parameter - when used, favorize a smoother ANN

Referenced by ClassImp(), and setDataSet().

◆ _listOfAttributes

TList* URANIE::Modeler::TANNModeler::_listOfAttributes

private

! Max values, used for normalization

Referenced by ClassImp(), and getInputName().

◆ _nApp

Int_t URANIE::Modeler::TANNModeler::_nApp

private

! The number of neurons

Referenced by ClassImp(), and getNAPP().

◆ _nHidden

Int_t URANIE::Modeler::TANNModeler::_nHidden

private

! The number of inputs

Referenced by ClassImp(), and getNHidden().

◆ _nHiddenLayer

Int_t URANIE::Modeler::TANNModeler::_nHiddenLayer

private

! The number of hidden neurons

Referenced by ClassImp().

◆ _nInput

Int_t URANIE::Modeler::TANNModeler::_nInput

private

! Seed for the random weight initialization

Referenced by ClassImp(), and getNInput().

◆ _nNeurons

Int_t URANIE::Modeler::TANNModeler::_nNeurons

private

! The number of outputs

Referenced by ClassImp().

◆ _nNormType

ENorm URANIE::Modeler::TANNModeler::_nNormType

private

! The function tolerance for the trust region algorithm

Referenced by ClassImp(), and getNormalization().

◆ _nOutput

Int_t URANIE::Modeler::TANNModeler::_nOutput

private

! the number of hidden layer

Referenced by ClassImp(), and getNOutput().

◆ _nProlemType

EProblem URANIE::Modeler::TANNModeler::_nProlemType

private

! The normalized for input and output

Referenced by ClassImp(), and getProblem().

◆ _nSeed

Int_t URANIE::Modeler::TANNModeler::_nSeed

protected

Referenced by ClassImp(), and getSeed().

◆ _nTest

Int_t URANIE::Modeler::TANNModeler::_nTest

private

! Number of samples in the training set dedicated to back-propagation

Referenced by ClassImp(), and getNTest().

◆ _nWeight

Int_t URANIE::Modeler::TANNModeler::_nWeight

private

Referenced by ClassImp().

◆ _sArchi

TString URANIE::Modeler::TANNModeler::_sArchi

private

! True if data set must be mixed

Referenced by ClassImp().

◆ _sHidden

TString URANIE::Modeler::TANNModeler::_sHidden

private

Referenced by ClassImp().

◆ _sInput

TString URANIE::Modeler::TANNModeler::_sInput

private

! The architecture, e.g. "x:y:z,3,yhat" specifying inputs, number of hidden neurons, and output

Referenced by ClassImp().

◆ _sOutput

TString URANIE::Modeler::TANNModeler::_sOutput

private

Referenced by ClassImp(), and getOutputName().

◆ _tds

URANIE::DataServer::TDataServer* URANIE::Modeler::TANNModeler::_tds

Referenced by ClassImp().

◆ _tmpSolution

TempSolution* URANIE::Modeler::TANNModeler::_tmpSolution

private

Referenced by ClassImp().

◆ _UMLP

UMLP* URANIE::Modeler::TANNModeler::_UMLP

Pointeur vers un TDS.

Referenced by ClassImp().

◆ _vecMaxValues

ColumnVector URANIE::Modeler::TANNModeler::_vecMaxValues

private

! Min values, used for normalization

Referenced by ClassImp(), and getMaxValue().

◆ _vecMinValues

ColumnVector URANIE::Modeler::TANNModeler::_vecMinValues

private

Referenced by ClassImp(), and getMinValue().

◆ _vecWeights

ColumnVector URANIE::Modeler::TANNModeler::_vecWeights

private

! The number of weights

! The vector of weights, stored like this: [a][b(1) b(2) b(3).... b(_nInput+2)][c(1) c(2) c(3).... c(_nInput+2)][ ... ] a: scalar value giving the weight of the output biais (link betw. single output and biais neuron in hidden layer) b, c, d, ...: _nHidden blocks of (_nInput+2) scalars representing: b(1): weight link between second hidden neuron and output (c(1) for the 3rd hidden neuron, d(1) for the 4th, etc ...) b(2): weight between biais input neuron and second hidden neuron b(2+i): weight between i-th (real) input neuron and second hidden neuron See impl. of UMPL::evaluateInput().

Referenced by ClassImp(), getWeight(), getWeights(), and setWeights().

◆ OPTIM_MAX_FEVAL

const int URANIE::Modeler::TANNModeler::OPTIM_MAX_FEVAL = 50000

staticprivate

Referenced by ClassImp().

◆ OPTIM_MAX_ITER

const int URANIE::Modeler::TANNModeler::OPTIM_MAX_ITER = 10000

staticprivate

Referenced by ClassImp().

Documentation / Developer's manual

Available modules

Public Types

Public Member Functions

Protected Attributes

Private Attributes

Parameters of algorithm

Additional Inherited Members

Member Enumeration Documentation

◆ ENorm

◆ EProblem

Constructor & Destructor Documentation

◆ TANNModeler()

◆ ~TANNModeler()

Member Function Documentation

◆ exportModelCplusplus()

◆ exportModelFortran()

◆ exportModelPython()

◆ getdval()

◆ getEQM()

◆ getFcnTol()

◆ getInputName()

◆ getMaxValue()

◆ getMinValue()

◆ getNAPP()

◆ getNHidden()

◆ getNInput()

◆ getNormalization()

◆ getNOutput()

◆ getNTest()

◆ getOutputName()

◆ getProblem()

◆ getSeed()

◆ getWeight()

◆ getWeightDecay()

◆ getWeights()

◆ init()

◆ printLog()

◆ randomizeWeights()

◆ saveIfBetter()

◆ setDataSet()

◆ setFcnTol()

◆ setMix()

◆ setNormalization()

◆ setSeed()

◆ setValue()

◆ setWeights()

◆ train()

Member Data Documentation

◆ _blog

◆ _bMix

◆ _dFcnTol

◆ _dLearn

◆ _dTest

◆ _dval

◆ _dWeightDecay

◆ _elAll

◆ _listOfAttributes

◆ _nApp

◆ _nHidden

◆ _nHiddenLayer

◆ _nInput

◆ _nNeurons

◆ _nNormType

◆ _nOutput

◆ _nProlemType

◆ _nSeed

◆ _nTest

◆ _nWeight

◆ _sArchi

◆ _sHidden

◆ _sInput

◆ _sOutput

◆ _tds

◆ _tmpSolution

◆ _UMLP

◆ _vecMaxValues

◆ _vecMinValues

◆ _vecWeights

◆ OPTIM_MAX_FEVAL