The KGraph index. More...

#include <kgraph.h>

Classes
struct	IndexInfo
	Information and statistics of the indexing algorithm. More...

struct	IndexParams
	Indexing parameters. More...

struct	SearchInfo
	Information and statistics of the search algorithm. More...

struct	SearchParams
	Search parameters. More...

Public Types
enum	{ FORMAT_DEFAULT = 0, FORMAT_NO_DIST = 1 }

Public Member Functions
virtual void	load (char const *path)=0
	Load index from file. More...

virtual void	save (char const *path, int format=FORMAT_DEFAULT) const =0
	Save index to file. More...

virtual void	build (IndexOracle const &oracle, IndexParams const &params, IndexInfo *info=0)=0
	Build the index.

virtual void	prune (IndexOracle const &oracle, unsigned level)=0
	Prune the index. More...

unsigned	search (SearchOracle const &oracle, SearchParams const &params, unsigned ids, SearchInfo info=0) const
	Online k-NN search. More...

virtual unsigned	search (SearchOracle const &oracle, SearchParams const &params, unsigned ids, float dists, SearchInfo *info) const =0
	Online k-NN search. More...

virtual void	get_nn (unsigned id, unsigned nns, unsigned M, unsigned *L) const
	Get offline computed k-NNs of a given object. More...

virtual void	get_nn (unsigned id, unsigned nns, float dists, unsigned M, unsigned L) const =0
	Get offline computed k-NNs of a given object. More...

virtual void	reverse (int)=0

Static Public Member Functions
static KGraph *	create ()
	Constructor.

static char const *	version ()
	Returns version string.

Detailed Description

The KGraph index.

This is an abstract base class. Use KGraph::create to create an instance.

Member Function Documentation

virtual void kgraph::KGraph::get_nn	(	unsigned	id,
		unsigned *	nns,
		unsigned *	M,
		unsigned *	L
	)		const

inlinevirtual

Get offline computed k-NNs of a given object.

See the full version of get_nn.

virtual void kgraph::KGraph::get_nn	(	unsigned	id,
		unsigned *	nns,
		float *	dists,
		unsigned *	M,
		unsigned *	L
	)		const

pure virtual

Get offline computed k-NNs of a given object.

The user must provide space to save IndexParams::L values. The actually returned L could be smaller than IndexParams::L, and M <= L is the number of neighbors KGraph thinks could be most useful for online search, and is usually < L. If the index has been pruned, the returned L could be smaller than IndexParams::L used to construct the index.

id Object ID whose neighbor information are returned. nns Neighbor IDs, must have space to save IndexParams::L values. dists Distance values, must have space to save IndexParams::L values. M Useful number of neighbors, output only. L Actually returned number of neighbors, output only.

virtual void kgraph::KGraph::load ( char const * path )

pure virtual

Load index from file.

Parameters

path	Path to the index file.

virtual void kgraph::KGraph::prune	(	IndexOracle const &	oracle,
		unsigned	level
	)

pure virtual

Prune the index.

Pruning makes the index smaller to save memory, and makes online search on the pruned index faster. (The cost parameters of online search must be enlarged so accuracy is not reduced.)

Currently only two pruning levels are supported:

PRUNE_LEVEL_1 = 1: Only reduces index size, fast.
PRUNE_LEVEL_2 = 2: For improve online search speed, slow.

No pruning is done if level = 0.

virtual void kgraph::KGraph::save	(	char const *	path,
		int	format = `FORMAT_DEFAULT`
	)		const

pure virtual

Save index to file.

Parameters

path	Path to the index file.

unsigned kgraph::KGraph::search	(	SearchOracle const &	oracle,
		SearchParams const &	params,
		unsigned *	ids,
		SearchInfo *	info = `0`
	)		const

inline

Online k-NN search.

Search results are guaranteed to be ranked in ascending order of distance.

Parameters

ids	Pointer to the memory where neighbor IDs are stored, must have space to save params.K ids.

virtual unsigned kgraph::KGraph::search	(	SearchOracle const &	oracle,
		SearchParams const &	params,
		unsigned *	ids,
		float *	dists,
		SearchInfo *	info
	)		const

pure virtual

Online k-NN search.

Search results are guaranteed to be ranked in ascending order of distance.

Parameters

ids	Pointer to the memory where neighbor IDs are stored, must have space to save params.K values.
dists	Pointer to the memory where distances are stored, must have space to save params.K values.

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

kgraph.h

Classes

Public Types

Public Member Functions

Static Public Member Functions

Detailed Description

Member Function Documentation