DataFile Class Reference

The DataFile class is a DataSource based upon a file. More...

#include <Dataset.h>

Inheritance diagram for DataFile:

Public Member Functions
	DataFile (const char *filename)
	Constructor.
	~DataFile ()
	Destructor.
bool	GetData (Array< double > &data, Array< double > &target, int count)
	Deliver data from the file.
bool	GetData (Array< double > &training_data, Array< double > &training_target, int training, Array< double > &test_data, Array< double > &test_target, int test, bool shuffle=false)
	Deliver data from the file.
int	getNumberOfExamples ()
	Return the number of examples available.
Protected Member Functions
bool	ReadHeaderLine ()
bool	ReadExample (Array< double > &data, Array< double > &target, int number)
int	ReadToken (char buffer, int maxlength, const char separators)
int	DiscardUntil (const char *separators)
Protected Attributes
FILE *	file
	open file descriptor
int	numberOfExamples
	number of examples
int	format
	0=ascii, 1=sparse, 2=float, 3=double
int	currentExample
	number of examples already deliviered

Detailed Description

The DataFile class is a DataSource based upon a file.

ReClaM defines a simple file format for real valued datasets. The first line, starting with a doublecross ('#'), serves as a file header. It contains exactly 4 tokens separated by whitespace with the following meaning:

number of examples
input dimension
output/target dimension
data format: one of the keywords "ascii", "sparse", "float", "double", "int8", "int16", "int32", "uint8", "uint16" or "uint32"

: Generally, the data are organized sample by sample, and for each sample, the inputs are followed by the outputs.

: In ascii and sparse format, every line defines one example where a single space character serves as a separator. All other formats are binary, that is, the numbers are organized continuously without separators. For integers and unsigned integers little endian encoding and for floating point numbers IEEE float or double format is assumed.

: In sparse format, all numbers are assumed to be zero. Exceptions are indicated by pairs of the format "index:value", interpreted as "data(index) = value". A semicolon is used to separate the data from the targets. The targets are NOT sparse encoded, that is, they are in standard ascii format.

Definition at line 149 of file Dataset.h.

Constructor & Destructor Documentation

DataFile::DataFile ( const char * filename )

Constructor.

Definition at line 112 of file Dataset.cpp.

References file, and ReadHeaderLine().

DataFile::~DataFile ( )

Destructor.

Definition at line 125 of file Dataset.cpp.

References file.

Member Function Documentation

int DataFile::DiscardUntil ( const char * separators ) [protected]

Definition at line 395 of file Dataset.cpp.

References file.

Referenced by ReadExample(), and ReadHeaderLine().

bool DataFile::GetData	(	Array< double > &	training_data,
		Array< double > &	training_target,
		int	training,
		Array< double > &	test_data,
		Array< double > &	test_target,
		int	test,
		bool	shuffle = `false`
	)