sgpp::datadriven::SparseGridMinerSplitting Class Reference

SparseGridMiner models a datamining process that involves a dataset that is first split into validation and training data. More...

#include <SparseGridMinerSplitting.hpp>

Inheritance diagram for sgpp::datadriven::SparseGridMinerSplitting:

Public Member Functions
double	evaluateLambda (double lambda, bool verbose)
	Fit the model using a specific lambda value and return a score for this lambda.
double	learn (bool verbose) override
	Perform Learning cycle: Get samples from data source and based on the scoring procedure, generalize data by fitting and asses quality of the fit.
SparseGridMinerSplitting &	operator= (const SparseGridMinerSplitting &rhs)=delete
	Default copy assign operator deleted because not all members can be copied.
SparseGridMinerSplitting &	operator= (SparseGridMinerSplitting &&rhs)=default
	Default Move assign operator.
double	optimizeLambda (bool verbose)
	Optimizes the lambda value for regularization.
	SparseGridMinerSplitting (const SparseGridMinerSplitting &rhs)=delete
	Copy constructor deleted - not all members can be copied or cloned .
	SparseGridMinerSplitting (DataSourceSplitting *dataSource, ModelFittingBase *fitter, Scorer *scorer, Visualizer *visualizer)
	Constructor.
	SparseGridMinerSplitting (SparseGridMinerSplitting &&rhs)=default
	Default Move constructor .
	~SparseGridMinerSplitting () override=default
	Default destructor.
Public Member Functions inherited from sgpp::datadriven::SparseGridMiner
ModelFittingBase *	getModel ()
	Returns the trained model.
Visualizer *	getVisualizer ()
SparseGridMiner &	operator= (const SparseGridMiner &rhs)=delete
	Default copy assign operator deleted because not all members can be copied.
SparseGridMiner &	operator= (SparseGridMiner &&rhs)=default
	Default Move assign operator.
void	setModel (ModelFittingBase *model)
	SparseGridMiner (const SparseGridMiner &rhs)=delete
	Copy constructor deleted - not all members can be copied or cloned .
	SparseGridMiner (ModelFittingBase *fitter, Scorer *scorer, Visualizer *visualizer)
	Constructor.
	SparseGridMiner (SparseGridMiner &&rhs)=default
	Default Move constructor .
double	test (Dataset &testDataset)
	Evaluate the model on a certain test dataset.
virtual	~SparseGridMiner ()=default
	Default destructor.

Additional Inherited Members
Static Public Member Functions inherited from sgpp::datadriven::SparseGridMiner
static void	print (const char *message)
	Print output on one process.
static void	print (const std::string &message)
	Print output on one process.
static void	print (std::ostringstream &messageStream)
	Print output on one process.
Protected Attributes inherited from sgpp::datadriven::SparseGridMiner
std::unique_ptr< ModelFittingBase >	fitter
	Fitter that trains a model based on data samples.
std::unique_ptr< Scorer >	scorer
	Scorer that quantifies the quality of a fit.
std::unique_ptr< Visualizer >	visualizer

Detailed Description

SparseGridMiner models a datamining process that involves a dataset that is first split into validation and training data.

The model is then trained on the training data for several epochs.

Constructor & Destructor Documentation

SparseGridMinerSplitting() [1/3]

sgpp::datadriven::SparseGridMinerSplitting::SparseGridMinerSplitting	(	DataSourceSplitting *	dataSource,
		ModelFittingBase *	fitter,
		Scorer *	scorer,
		Visualizer *	visualizer
	)

Constructor.

Parameters

dataSource	configured instance of data source object, that will provide samples to learn from. The miner instance will take ownership of the passed object.
fitter	configured instance of fitter object that generalize the model. The miner instance will take ownership of the passed object.
scorer	configured instance of scorer object that will assess the quality of the generalization provided by the fitter on testing data. The miner instance will take ownership of the passed object.
visualizer	configured instance of viusalizer object that will produce the output to visualize the model and its results

SparseGridMinerSplitting() [2/3]

sgpp::datadriven::SparseGridMinerSplitting::SparseGridMinerSplitting ( const SparseGridMinerSplitting &  rhs )

delete

Copy constructor deleted - not all members can be copied or cloned .

Parameters

rhs	the object to copy from

SparseGridMinerSplitting() [3/3]

sgpp::datadriven::SparseGridMinerSplitting::SparseGridMinerSplitting ( SparseGridMinerSplitting &&  rhs )

default

Default Move constructor .

Parameters

rhs	the object to move from

~SparseGridMinerSplitting()

sgpp::datadriven::SparseGridMinerSplitting::~SparseGridMinerSplitting ( )

overridedefault

Default destructor.

Member Function Documentation

evaluateLambda()

double sgpp::datadriven::SparseGridMinerSplitting::evaluateLambda	(	double	lambda,
		bool	verbose
	)

Fit the model using a specific lambda value and return a score for this lambda.

References dataset, sgpp::datadriven::SparseGridMiner::fitter, sgpp::datadriven::Dataset::getNumberInstances(), lambda, and sgpp::datadriven::SparseGridMiner::scorer.

Referenced by optimizeLambda().

learn()

double sgpp::datadriven::SparseGridMinerSplitting::learn ( bool  verbose )

overridevirtual

Perform Learning cycle: Get samples from data source and based on the scoring procedure, generalize data by fitting and asses quality of the fit.

The learning process first divides the data into training and validation data and trains the model for several epochs on the training data.

Implements sgpp::datadriven::SparseGridMiner.

References sgpp::datadriven::RefinementMonitorFactory::createRefinementMonitor(), dataset, sgpp::datadriven::SparseGridMiner::fitter, sgpp::datadriven::Dataset::getNumberInstances(), lambda, optimizeLambda(), sgpp::datadriven::SparseGridMiner::print(), sgpp::datadriven::RefinementMonitor::pushToBuffer(), sgpp::datadriven::RefinementMonitor::refinementsNecessary(), sgpp::datadriven::SparseGridMiner::scorer, and sgpp::datadriven::SparseGridMiner::visualizer.

operator=() [1/2]

SparseGridMinerSplitting & sgpp::datadriven::SparseGridMinerSplitting::operator= ( const SparseGridMinerSplitting &  rhs )

delete

Default copy assign operator deleted because not all members can be copied.

Parameters

rhs	the object to copy from

operator=() [2/2]

SparseGridMinerSplitting & sgpp::datadriven::SparseGridMinerSplitting::operator= ( SparseGridMinerSplitting &&  rhs )

default

Default Move assign operator.

Parameters

rhs	the object to move from

optimizeLambda()

double sgpp::datadriven::SparseGridMinerSplitting::optimizeLambda

(

bool

verbose

)

Optimizes the lambda value for regularization.

In order to do this, multiple alpha vecotrs are created with different lambda values and Golden section search is used to quickly find the optimum.

References evaluateLambda(), sgpp::datadriven::SparseGridMiner::fitter, and sgpp::datadriven::SparseGridMiner::print().

Referenced by learn().

---

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

• datadriven/src/sgpp/datadriven/datamining/base/SparseGridMinerSplitting.hpp
• datadriven/src/sgpp/datadriven/datamining/base/SparseGridMinerSplitting.cpp