Implement the Neural Gas algorithm (Martinetz and Schulten, 1991) More...

#include <TNeuralGas.h>

Inheritance diagram for URANIE::Sampler::TNeuralGas:

[legend]

Collaboration diagram for URANIE::Sampler::TNeuralGas:

[legend]

Public Member Functions
Constructor and Destructor
	TNeuralGas (URANIE::DataServer::TDataServer tds, Option_t option, Int_t ns)
	Constructor with a dataserver.

virtual	~TNeuralGas ()
	Default destructor.

Setting and Getting attributs
void	setLambda (Double_t dinitlambda, Double_t dfinallambda)
	Set the lambda parameters.

void	setAlpha (Double_t dinitalpha, Double_t dfinalalpha)
	Set the alpha parameters.

void	setTMax (Int_t dtmax)
	Set the $t_{max}$ parameter.

Int_t	getTMax ()
	Get the $t_{max}$ parameter.

State machine
For the stochastic methods, two states can be created init and terminate which are executed just before the generation of the sample for the first one, and after the generation for the second one. They can be used for pre and post treatment purpose.
void	init ()
	The preprocessing step.

void	learn (Int_t index)
	The learning step for the index pattern.

Generate the sample
the generate sample procedure
void	generateSample (Option_t *option="")
	Generates the sample.

Printing Log
virtual void	printLog (Option_t *option="")
	Prints the log.

Public Member Functions inherited from URANIE::Sampler::TVQ
	TVQ (URANIE::DataServer::TDataServer tds, Option_t option, Int_t ns)
	Constructor with a dataserver.

virtual	~TVQ ()
	Default destructor.

Bool_t	getSaveAll ()
	Returns the flag SaveAll.

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

Bool_t	getDrawProgressBar ()
	Get the clean flag.

void	createListOfAttributes ()
	Creates the list of attributes to select patterns.

void	shuffle ()
	shuffle the learning database (original TDataServer)

URANIE::DataServer::TDataServer *	getSubSample (Option_t *option="")
	Generate the Sample and return the TDataServer.

virtual void	terminate ()
	The post-processing step.

Public Member Functions inherited from URANIE::Sampler::TSamplerStochastic
	TSamplerStochastic (URANIE::DataServer::TDataServer tds, Option_t option, Int_t nCalcul)
	Constructor with a dataserver.

virtual	~TSamplerStochastic ()
	Default destructor.

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

Int_t	getSeed ()
	Returns the seed value.

Public Member Functions inherited from URANIE::Sampler::TSampler
	TSampler (URANIE::DataServer::TDataServer tds, Option_t option, Int_t nCalcul)
	Constructor with a TDataServer, the options and the size of the sample.

virtual	~TSampler ()
	Default destructor.

Int_t	GetID ()
	Returns the ID of the class.

void	setMethodName (TString str)
	Sets the method name in a global variable.

TString	getMethodName ()
	Gets the method name.

virtual URANIE::DataServer::TDataServer *	getTDS ()
	Return the TDS filling by the sampling algorithm.

void	parseOption (Option_t *option)
	Parse the option.

virtual void	createTuple ()
	Creates the TDSNtupleD of data with only the TStochasticAttributes.

URANIE::DataServer::TDSNtupleD *	getTuple ()
	Returns the TDSNtupleD of data.

virtual void	fillOtherAttributes ()
	Fills the TDSNtupleD of data with other TFormulaAttributes.

void	setLog ()

void	unsetLog ()

void	changeLog ()

Bool_t	getLog ()

Public Attributes
TMatrixD	_datamatrix
	The matrix of data.

TMatrixD	_dvw
	The matrix of difference.

TString	_svardraw

Int_t	_ntemperature

Double_t	_dtmax
	The $t_{max}$ parameter.

Double_t	_dinitlambda
	The initial value of $\lambda_i$ .

Double_t	_dlambda
	The current value of $\lambda_i$ .

Double_t	_dfinalambda
	The final value of $\lambda_f$ .

Double_t	_dinitalpha
	The initial value of $\alpha_i$ .

Double_t	_dalpha
	The current value of $\alpha_i$ .

Double_t	_dfinalalpha
	The final value of $\alpha_f$ .

Public Attributes inherited from URANIE::Sampler::TVQ
TVector	_vecindexorg
	The vector of index of the database.

URANIE::DataServer::TDataServer *	_subtds
	The subsampler returened by the getSubSample method.

Bool_t	_bsaveall
	The boolean to save all the data in the learning (default = kFALSE)

TString	_sattributes
	the list of attributes

Public Attributes inherited from URANIE::Sampler::TSampler
Int_t	_nS
	The size of the sample.

Int_t	_nX
	The size of attributes to sample.

URANIE::DataServer::TDSNtupleD *	_ntsample
	the tntuple of data

TString	_sMethod
	The title of the sampler method.

Bool_t	_blog
	Log Printing.

Bool_t	_bupdateFile
	Update the back up file when generating the attributeformula if there is some.

URANIE::DataServer::TDataServer *	_tds
	Pointer to a TDS.

TList *	_lstOfAttributesToSample
	The list of Stochastic Attributes to sample.

Additional Inherited Members
State machine
For the stochastic methods, two2 states can be created init and terminate which are executed just before the generation of the sample for the first one, and after the generation for the second one. They can be used for pre and post treatment purpose.
Protected Attributes inherited from URANIE::Sampler::TVQ
URANIE::DataServer::UMessageLogger *	_fLogger

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

Protected Attributes inherited from URANIE::Sampler::TSamplerStochastic
Int_t	_nSeed

Detailed Description

Implement the Neural Gas algorithm (Martinetz and Schulten, 1991)

The Neural Gas Algorithm (Martinetz and Schulten, 1991) sorts for each input signal the units of the network according to the distance of their reference vectors to $\xi$ . Based on this rank order a certain number of units is adapted. Both the number of adapted units and the adaptation strength are decreased according to a fixed schedule.

The reference vectors are adjusted according to

$\Delta \xi_j \: = \: \alpha(t) \exp^{-Rank(j)/\lambda(t)} \times (\xi_j - x)$

with the following time-dependencies:

$\alpha(t) = \alpha_i \; (\frac{\alpha_f}{\alpha_i})^{t/t_{max}}$

$\lambda(t) = \lambda_i \; (\frac{\lambda_f}{\lambda_i})^{t/t_{max}}$

Default values

$t_{max} = 40000$
$\alpha_i = 0.5$
$\alpha_f = 0.005$
$\lambda_f = 0.01$
$\lambda_i = \frac{Card(\xi)-1}{3}$

The initial cluster are selected randomly from the original dataset.

In the subsample dataserver, it exists an internal attribute "__neural__gas_iter__" which contains the number of loop need by the learning process.

Example: TDataServer *tds = new TDataServer("tdsgeyser", "Neural Gas for Geyser dataset");

tds->fileDataRead("geyser.dat");

TNeuralGas * tvq = new TNeuralGas(tds,"", 50);

TDataServer *tdsng = tvq->getSubSample("loop=100");

tds->draw("x2:x1");

tdsng->draw("x2:x1","","same");

URANIE::Sampler::TNeuralGas
Implement the Neural Gas algorithm (Martinetz and Schulten, 1991)
Definition TNeuralGas.h:92

URANIE::Sampler::TVQ::getSubSample
URANIE::DataServer::TDataServer * getSubSample(Option_t *option="")
Generate the Sample and return the TDataServer.
Definition TVQ.cxx:205

Constructor & Destructor Documentation

◆ TNeuralGas()

URANIE::Sampler::TNeuralGas::TNeuralGas	(	URANIE::DataServer::TDataServer *	tds,
		Option_t *	option,
		Int_t	ns
	)

Constructor with a dataserver.

Parameters

tds	(URANIE::DataServer::TDataServer *) the original data
option	(Option_t) contains the options [default ""] "saveall" Save all the data generate by the algorithm ( _bsaveall = kTRUE).
ns	(Int_t) the size of the subsample

References _dinitlambda, URANIE::Sampler::TSampler::_nS, and URANIE::Sampler::TSampler::_tds.