#include <indexer.h>

Collaboration diagram for MultiIndexer< Record >:

Public Member Functions
	MultiIndexer (const int multiplicity=2)
void	BuildMultiIndex (const string &points_filename, const string &metainfo_filename, const int points_count, const vector< Centroids > &coarse_vocabs, const vector< Centroids > &fine_vocabs, const RerankMode &mode, const bool build_coarse_quantization, const string &files_prefix, const string &coarse_quantization_filename="")
Private Member Functions
void	PrepareCoarseQuantization (const string &points_filename, const int points_count, const vector< Centroids > &coarse_vocabs)
void	GetCoarseQuantizationsForSubset (const string &points_filename, const int start_pid, const int subset_size, const vector< Centroids > &coarse_vocabs, vector< vector< ClusterId > > *transposed_coarse_quantizations)
void	SerializeCoarseQuantizations (const vector< vector< ClusterId > > &transposed_coarse_quantizations, const string &filename)
void	SerializeMultiIndexFiles ()
void	ConvertPointsInCellsCountToCellEdges ()
void	FillMultiIndex (const string &points_filename, const int points_count, const vector< Centroids > &coarse_vocabs, const vector< Centroids > &fine_vocabs, const RerankMode &mode)
void	FillMultiIndexForSubset (const string &points_filename, const PointId start_pid, const int points_count, const vector< Centroids > &coarse_vocabs, const vector< Centroids > &fine_vocabs, const RerankMode &mode, Multitable< int > *points_written_in_index)
void	GetPointCoarseQuantization (const PointId pid, const string &filename, vector< ClusterId > *coarse_quantization)
void	FillPointRerankInfo (const Point &point, const PointId pid, const vector< Centroids > &fine_vocabs)
void	RestorePointsInCellsCountFromCourseQuantization (const string &points_filename, const int points_count, const vector< Centroids > &coarse_vocabs)
int	GetInputCoordSizeof ()
void	ReadPoint (ifstream &input, Point *point)
void	InitBlasStructures (const vector< Centroids > &coarse_vocabs)
Private Attributes
string	files_prefix_
string	coarse_quantization_filename_
int	multiplicity_
Multitable< int >	point_in_cells_count_
MultiIndex< Record >	multiindex_
boost::mutex	cell_counts_mutex_
vector< float * >	coarse_vocabs_matrices_
vector< vector< float > >	coarse_centroids_norms_

Detailed Description

template<class Record>
class MultiIndexer< Record >

This is the main class for creating multiindex for a set of points in a multidimensional space. Clusterization and vocabs learning happen outside of this class, multiindexer receives prepared vocabs in input

Constructor & Destructor Documentation

template<class Record >

MultiIndexer< Record >::MultiIndexer ( const int multiplicity = 2 )

This is the simple MultiIndexer constructor

Parameters:

multiplicity how many parts input points will be divide on

Member Function Documentation

template<class Record >

void MultiIndexer< Record >::BuildMultiIndex	(	const string &	points_filename,
		const string &	metainfo_filename,
		const int	points_count,
		const vector< Centroids > &	coarse_vocabs,
		const vector< Centroids > &	fine_vocabs,
		const RerankMode &	mode,
		const bool	build_coarse_quantization,
		const string &	files_prefix,
		const string &	coarse_quantization_filename = `""`
	)

This is the main function of MultiIndexer

Parameters:

points_filename	file with points in .fvecs or .bvecs format
points_count	how many points should we index
coarse_vocabs	vocabularies for coarse quantization
fine_vocabs	vocabularies for fine quantization for reranking
mode	determines the way of rerank info calculating
build_coarse_quantization	should we get coarse quantization or not
files_prefix	all index filenames will have this prefix
coarse_quantization_filename	file with coarse quantization (if exists)

template<class Record >

void MultiIndexer< Record >::ConvertPointsInCellsCountToCellEdges ( ) [private]

This function converts counts of points in cells to cell edges

template<class Record >

void MultiIndexer< Record >::FillMultiIndex	(	const string &	points_filename,
		const int	points_count,
		const vector< Centroids > &	coarse_vocabs,
		const vector< Centroids > &	fine_vocabs,
		const RerankMode &	mode
	)		`[private]`

This function fills multiindex data structures.

Parameters:

points_filename	file with points in .fvecs or .bvecs format
points_count	how many points should we index
coarse_vocabs	vocabularies for coarse quantization
fine_vocabs	vocabularies for fine quantization for reranking
mode	determines the way of rerank info calculating

template<class Record >

void MultiIndexer< Record >::FillMultiIndexForSubset	(	const string &	points_filename,
		const PointId	start_pid,
		const int	points_count,
		const vector< Centroids > &	coarse_vocabs,
		const vector< Centroids > &	fine_vocabs,
		const RerankMode &	mode,
		Multitable< int > *	points_written_in_index
	)		`[private]`

This function fills multiindex data structures.

Parameters:

points_filename	file with points in .fvecs or .bvecs format
start_pid	identifier of the first point in subset
subset_size	points count in subset
coarse_vocabs	vocabularies for coarse quantization
fine_vocabs	vocabularies for fine quantization for reranking
mode	determines the way of rerank info calculating
points_written_in_index	auxillary structure for correct index filling

template<class Record >

void MultiIndexer< Record >::FillPointRerankInfo	(	const Point &	point,
		const PointId	pid,
		const vector< Centroids > &	fine_vocabs
	)		`[private]`

This function calculates rerank info for point

Parameters:

point	target point
pid	identifier of target point
fine_vocabs	vocabularies for rerank info calculation

template<class Record >

void MultiIndexer< Record >::GetCoarseQuantizationsForSubset	(	const string &	points_filename,
		const int	start_pid,
		const int	subset_size,
		const vector< Centroids > &	coarse_vocabs,
		vector< vector< ClusterId > > *	transposed_coarse_quantizations
	)		`[private]`

This function prepares for each point in subset its coarse quantization

Parameters:

points_filename	file with points in .fvecs or .bvecs format
start_pid	identifier of the first point in subset
subset_size	points count in subset
coarse_vocabs	vocabularies for coarse quantization
transposed_coarse_quantizations	result

template<class Record >

int MultiIndexer< Record >::GetInputCoordSizeof ( ) [private]

This simple function returns size of one coordinate of input point

template<class Record >

void MultiIndexer< Record >::GetPointCoarseQuantization	(	const PointId	pid,
		const string &	filename,
		vector< ClusterId > *	coarse_quantization
	)		`[private]`

This function reads point coarse quantization from file

Parameters:

pid	identifier of target point
filename	file with coarse quantizations
coarse_quantization	result

template<class Record >

void MultiIndexer< Record >::InitBlasStructures ( const vector< Centroids > & coarse_vocabs ) [private]

Initialize all structures for BLAS operations

Parameters:

coarse_vocabs coarse vocabularies

template<class Record >

void MultiIndexer< Record >::PrepareCoarseQuantization	(	const string &	points_filename,
		const int	points_count,
		const vector< Centroids > &	coarse_vocabs
	)		`[private]`

This function prepares for each point its coarse quantization

Parameters:

points_filename	file with points in .fvecs or .bvecs format
points_count	how many points should we handle
coarse_vocabs	vocabularies for coarse quantization

template<class Record >

void MultiIndexer< Record >::ReadPoint	(	ifstream &	input,
		Point *	point
	)		`[private]`

This simple function reads one point from input stream

Parameters:

input	input stream
point	result point

template<class Record >

void MultiIndexer< Record >::RestorePointsInCellsCountFromCourseQuantization	(	const string &	points_filename,
		const int	points_count,
		const vector< Centroids > &	coarse_vocabs
	)		`[private]`

This function restores counts of points from coarse quantizations

Parameters:

points_filename	file with points in .fvecs or .bvecs format
points_count	how many points should we index
coarse_vocabs	vocabularies for coarse quantization We need them to init counts table correctly

template<class Record >

void MultiIndexer< Record >::SerializeCoarseQuantizations	(	const vector< vector< ClusterId > > &	transposed_coarse_quantizations,
		const string &	filename
	)		`[private]`

This function serializes prepared coarse quantizations to file

Parameters:

transposed_coarse_quantizations	quantizations to serialize. They are transposed because of effective memory usage
filename	file we should serialize to

template<class Record >

void MultiIndexer< Record >::SerializeMultiIndexFiles ( ) [private]

This function saves index to files. All filenames start form the common files prefix

Member Data Documentation

template<class Record >

boost::mutex MultiIndexer< Record >::cell_counts_mutex_ [private]

Mutex for critical section in filling index stage

template<class Record >

vector<vector<float> > MultiIndexer< Record >::coarse_centroids_norms_ [private]

Struct for BLAS

template<class Record >

string MultiIndexer< Record >::coarse_quantization_filename_ [private]

Filename of file with coarse quantizations

template<class Record >

vector<float*> MultiIndexer< Record >::coarse_vocabs_matrices_ [private]

Struct for BLAS

template<class Record >

string MultiIndexer< Record >::files_prefix_ [private]

All index filenames will start from this prefix

template<class Record >

MultiIndex<Record> MultiIndexer< Record >::multiindex_ [private]

Multiindex

template<class Record >

int MultiIndexer< Record >::multiplicity_ [private]

Multiplicity (how many parts point space is divided on)

template<class Record >

Multitable<int> MultiIndexer< Record >::point_in_cells_count_ [private]

Table with number of points in each cell

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

indexer.h

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

template<class Record> class MultiIndexer< Record >

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation

template<class Record>
class MultiIndexer< Record >