|| || || || || ||

---

Impala::Core::Training::TrainDataSrc Class Reference

Abstraction of training data. More...

#include <TrainDataSrc.h>

Inheritance diagram for Impala::Core::Training::TrainDataSrc:

[legend]Collaboration diagram for Impala::Core::Training::TrainDataSrc:

[legend]List of all members.

Public Member Functions
	TrainDataSrc (Table::AnnotationTable *annotation)
virtual	~TrainDataSrc ()
virtual svm_problem *	MakeSvmProblem ()=0
	Create an svm-problem of all available data.
virtual svm_problem *	MakeSvmProblem (int i)=0
	create an svm_problem of *one* example
virtual int	GetVectorLength ()=0
virtual Quid	GetQuid (int i)
virtual int	Size ()
virtual void	FilterTestFold (int f, int foldCount, int repetition, bool episodeConstrained, int restrictSet)
virtual void	FilterTrainFold (int f, int foldCount, int repetition, bool episodeConstrained)
virtual void	FreeProblem (svm_problem *p)
int	GetTotalPositiveCount ()
int	GetTotalNegativeCount ()
void	PrintSelection ()
Protected Member Functions
std::vector< Table::QuidTable * >	MakeFolds (int f, int foldCount, int repetition, bool episodeConstrained)
void	Clear ()
void	SetAnnotation (Table::AnnotationTable *anno)
svm_problem *	MakeEmptyProblem ()
void	ClearSelection ()
Protected Attributes
Table::AnnotationTable *	mAnnotation
Table::QuidTable *	mQuids
Table::QuidTable *	mSelection
	ILOG_VAR_DECL

---

Detailed Description

Abstraction of training data.

TrainData kan have multiple flavours: feature data in one table (FeatureIndex); feature data in multiple tables (FeatureData); precomputed kernels in a matrix; precomputed kernels stored on ohter machines (DistributedAccess). This class provides a single interface to all different kinds of data.

A TrainDataSrc kan be 'filtered'; this means that some data is made unavailable. Currently only filtering on folds is used.

Note:

The current implementation is dependant on SVMs interface, because SVM was the only use case during design/implementation.

Todo:

Ideally this class should be independant of the Classifier used. Perhaps the MakeSvmProblem will be replaced with an interface to the vectors, and labels of individual samples.

Todo:

The function free problem does not belong in this class. In stead of a funcitno like this, svm_problem should have a wrapper so the destructor can take care of this.

Definition at line 42 of file TrainDataSrc.h.

---

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

• Core/Training/TrainDataSrc.h

---