Dataset Class Reference

The Dataset class encapsulates a realization of data from a DataSource. More...

#include <Dataset.h>

Public Member Functions
	Dataset (const Dataset &dataset)
	Construction of a Dataset from another Dataset.
	Dataset (DataSource &source, int train, int test)
	Construction of a Dataset from a generic DataSource.
	Dataset (const char *filename, int train, int test=0)
	Construction of a Dataset from a single file.
	Dataset (const char *filename, double train)
	Construction of a Dataset from a single file.
	Dataset (const char trainfile, const char testfile)
	Construction of a Dataset from a pair of files.
	Dataset (const char trainfile, const char testfile, int train)
	Construction of a Dataset from a pair of files, but using a different data separation into training and test set.
	Dataset (const char datafile, const char splitfile, double disambiguation)
	Construction of a Dataset from a data file and a split file.
	Dataset (const Array< double > &trainingData, const Array< double > &trainingTarget, const Array< double > &testData, const Array< double > &testTarget)
	Construction of a Dataset object from given arrays.
void	ShuffleTraining ()
	shuffles the training examples
void	ShuffleTest ()
	shuffles the test examples
void	ShuffleAll ()
	shuffles the union of training and test examples, such that the number of training and test examples remains unchanged
const Array< double > &	getTrainingData () const
	access to the training data as a constant array
const Array< double > &	getTrainingTarget () const
	access to the training targets as a constant array
const Array< double > &	getTestData () const
	access to the test data as a constant array
const Array< double > &	getTestTarget () const
	access to the test targets as a constant array
bool	Save (const char filename, bool training=true, bool test=true, const char format="ascii")
	Save the current Dataset to a file.
bool	SaveLIBSVM (const char *filename, bool training=true, bool test=true)
	Save the current Dataset in LIBSVM format.
void	NormalizeComponents ()
	component wise normalization of the dataset
void	NormalizeComponent (int d)
	normalizes a single component
Protected Member Functions
bool	ReadSplitFile (const char *filename, std::vector< unsigned int > &train, std::vector< unsigned int > &test)
bool	ReadLine (FILE file, char buffer, int bufferlength)
Protected Attributes
Array< double >	trainingData
Array< double >	trainingTarget
Array< double >	testData
Array< double >	testTarget

Detailed Description

The Dataset class encapsulates a realization of data from a DataSource.

: A Dataset consists of separate training and test data. It may be necessary to split the training data into sub blocks, e.g. for cross validation. However, the test data are assumed to be completely unknown during the whole training process.

Definition at line 203 of file Dataset.h.

Constructor & Destructor Documentation

Dataset::Dataset ( const Dataset & dataset )

Construction of a Dataset from another Dataset.

Definition at line 418 of file Dataset.cpp.

References getTestData(), getTestTarget(), getTrainingData(), getTrainingTarget(), testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	DataSource &	source,
		int	train,
		int	test
	)

Construction of a Dataset from a generic DataSource.

Definition at line 426 of file Dataset.cpp.

References DataSource::GetData(), testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	const char *	filename,
		int	train,
		int	test = `0`
	)

Construction of a Dataset from a single file.

Definition at line 434 of file Dataset.cpp.

References DataFile::GetData(), DataFile::getNumberOfExamples(), testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	const char *	filename,
		double	train
	)

Construction of a Dataset from a single file.

Definition at line 446 of file Dataset.cpp.

References DataFile::GetData(), DataFile::getNumberOfExamples(), testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	const char *	trainfile,
		const char *	testfile
	)

Construction of a Dataset from a pair of files.

Definition at line 457 of file Dataset.cpp.

References DataFile::GetData(), DataFile::getNumberOfExamples(), testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	const char *	trainfile,
		const char *	testfile,
		int	train
	)

Construction of a Dataset from a pair of files, but using a different data separation into training and test set.

Definition at line 467 of file Dataset.cpp.

References DataFile::GetData(), DataFile::getNumberOfExamples(), i, testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	const char *	datafile,
		const char *	splitfile,
		double	disambiguation
	)

Construction of a Dataset from a data file and a split file.

Definition at line 529 of file Dataset.cpp.

References DataFile::GetData(), DataSource::getDataDimension(), DataFile::getNumberOfExamples(), DataSource::getTargetDimension(), i, ReadSplitFile(), testData, testTarget, trainingData, and trainingTarget.

Dataset::Dataset	(	const Array< double > &	trainingData,
		const Array< double > &	trainingTarget,
		const Array< double > &	testData,
		const Array< double > &	testTarget
	)

Construction of a Dataset object from given arrays.

Definition at line 575 of file Dataset.cpp.

Member Function Documentation

const Array<double>& Dataset::getTestData ( ) const [inline]

access to the test data as a constant array

Definition at line 257 of file Dataset.h.

Referenced by Dataset().

const Array<double>& Dataset::getTestTarget ( ) const [inline]

access to the test targets as a constant array

Definition at line 263 of file Dataset.h.

Referenced by Dataset().

const Array<double>& Dataset::getTrainingData ( ) const [inline]

access to the training data as a constant array

Definition at line 245 of file Dataset.h.

Referenced by Dataset().

const Array<double>& Dataset::getTrainingTarget ( ) const [inline]

access to the training targets as a constant array

Definition at line 251 of file Dataset.h.

Referenced by Dataset().

void Dataset::NormalizeComponent ( int d )

normalizes a single component

: Normalizes a component of the dataset by an affine linear transformation such that afterwards the component has zero mean and unit variance.

Definition at line 887 of file Dataset.cpp.

References i, testData, and trainingData.

void Dataset::NormalizeComponents ( )

component wise normalization of the dataset

: Normalize each component of the dataset by an affine linear transformation such that afterwards the training set has zero mean and unit variance in every component.

Definition at line 853 of file Dataset.cpp.

References i, testData, and trainingData.

bool Dataset::ReadLine	(	FILE *	file,
		char *	buffer,
		int	bufferlength
	)			`[protected]`

Definition at line 923 of file Dataset.cpp.

Referenced by ReadSplitFile().

bool Dataset::ReadSplitFile	(	const char *	filename,
		std::vector< unsigned int > &	train,
		std::vector< unsigned int > &	test
	)			`[protected]`

Definition at line 939 of file Dataset.cpp.

References i, and ReadLine().

Referenced by Dataset().

bool Dataset::Save	(	const char *	filename,
		bool	training = `true`,
		bool	test = `true`,
		const char *	format = `"ascii"`
	)

Save the current Dataset to a file.

Parameters:

	filename	name of the file, must not exist
	training	include the training data?
	test	include the test data?
	format	see description

Returns:: The method returns true on success and false in case of failure.

The data can be saved in one of the following formats:

ascii: text file with one ascii encoded number per input and output
sparse: text file with sparse encoding, usefull for datasets containing many zeros as input
float: binary file with one IEEE float number per input and output
double: binary file with one IEEE double number per input and output
int8: binary file with a signed 8-bit-integer per input and output
int16: binary file with a signed 16-bit-integer per input and output
int32: binary file with a signed 32-bit-integer per input and output
uint8: binary file with an unsigned 8-bit-integer per input and output
uint16: binary file with an unsigned 16-bit-integer per input and output
uint32: binary file with an unsigned 32-bit-integer per input and output

Definition at line 665 of file Dataset.cpp.

References Dataset_WriteType, i, testData, testTarget, trainingData, and trainingTarget.