drwnDataset< XType, YType, WType > Class Template Reference

Implements a cacheable dataset containing feature vectors, labels and optional weights. More...

Public Member Functions
	drwnDataset ()
	default constructor
	drwnDataset (const drwnDataset< XType, YType, WType > &d)
	copy constructor
	drwnDataset (const char *filename)
	construct and load dataset from file
bool	empty () const
	return true if the dataset is empty
int	size () const
	return the number of samples in the dataset
bool	hasWeights () const
	return true if the dataset contains weighted samples
bool	hasIndexes () const
	return true if the dataset has external indices associated with each sample
bool	valid () const
	return true if the dataset is valid (e.g., number of targets equals number of feature vectors)
int	count (const YType &label) const
	returns the number of samples with a given target label
void	reserve (int reserveSize)
	pre-allocate memory for storing samples (feature vectors and targets)
int	numFeatures () const
	returns the number of features in the feature vector
YType	minTarget () const
	returns the minimum target value in the dataset
YType	maxTarget () const
	returns the maximum target value in the dataset
void	clear ()
	clears all data in the dataset
int	write (const char *filename, bool bAppend=false) const
	writes the current dataset to disk (optionally appending to an existing dataset)
int	write (const char *filename, int startIndx, int endIndx, bool bAppend=false) const
	writes a range of samples to disk (optionally appending to an existing dataset)
int	read (const char *filename, bool bAppend=false)
	reads a dataset from disk (optionally appending to the current dataset)
int	read (const char *filename, int startIndx, int endIndx, bool bAppend=false)
	reads a range of samples from disk (optionally appending to the current dataset)
int	append (const drwnDataset< XType, YType, WType > &d)
	appends the samples from another dataset to this dataset
int	append (const vector< XType > &x, const YType &y)
	appends a sample (feature vector and target) to the dataset
int	append (const vector< XType > &x, const YType &y, const WType &w)
	appends a weighted sample (feature vector and target) to the dataset
int	append (const vector< XType > &x, const YType &y, const WType &w, int indx)
	appends a weighted sample with associated external index to the dataset
int	subSample (int sampleRate, bool bBalanced=false)
	subsample a dataset (balanced is only valid for discrete target types) if bBalanced is true then sampleRate is applied to most abundant target More...

Static Public Member Functions
static int	size (const char *filename)
	returns the size of a dataset stored on disk
static int	numFeatures (const char *filename)
	returns the number of features in the feature vectors of a dataset stored on disk
static bool	hasWeights (const char *filename)
	returns true if the dataset stored on disk contains weighted samples
static bool	hasIndexes (const char *filename)
	returns true if the dataset stored on disk has indexes associated with each sample

Public Attributes
vector< vector< XType > >	features
	feature vectors
vector< YType >	targets
	target labels
vector< WType >	weights
	weights (optional)
vector< int >	indexes
	external indices (optional)

Detailed Description

template<typename XType, typename YType, typename WType>
class drwnDataset< XType, YType, WType >

Implements a cacheable dataset containing feature vectors, labels and optional weights.

The dataset can be used by various machine learning algorithms. The dataset is stored on disk as

<version, flags (int32)> <num features (int32)>
<label> <feature 1> ... <feature n> (<weight>)? (<index>)?
...

Warning

The API for drwnDataset is not stable.

Member Function Documentation

template<typename XType , typename YType , typename WType >

int drwnDataset< XType, YType, WType >::subSample	(	int	sampleRate,
		bool	bBalanced = false
	)

subsample a dataset (balanced is only valid for discrete target types) if bBalanced is true then sampleRate is applied to most abundant target

Todo:

consider sample weighting

Todo:

enforce balanced for discrete targets only

---

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

• drwnDataset.h