mknn_predefined_index.h File Reference

MetricKnn provides a set of pre-defined indexes identified. More...

#include "../metricknn_c.h"

Go to the source code of this file.

Functions
Help functions.
void	mknn_predefIndex_helpListIndexes ()
	Lists to standard output all pre-defined indexes.
void	mknn_predefIndex_helpPrintIndex (const char *id_index)
	Prints to standard output the help for a index. More...
bool	mknn_predefIndex_testIndexId (const char *id_index)
	Tests whether the given string references a valid pre-defined index. More...
Linear Scan.
MknnIndexParams *	mknn_predefIndex_LinearScan_indexParams ()
	Resolves searches using the linear scan algorithm. More...
MknnResolverParams *	mknn_predefIndex_LinearScan_resolverExactNearestNeighbors (int64_t knn, double range, int64_t max_threads)
	Creates a new resolver for exact search of the K nearest neighbors using the linear scan index. More...
MknnResolverParams *	mknn_predefIndex_LinearScan_resolverExactFarthestNeighbors (int64_t knn, int64_t max_threads)
	Creates a new resolver for exact search of the K farthest neighbors using the linear scan index. More...
LAESA.
MknnIndexParams *	mknn_predefIndex_Laesa_indexParams (int64_t num_pivots)
	Resolves searches using the LAESA algorithm. More...
MknnResolverParams *	mknn_predefIndex_Laesa_resolverExactNearestNeighbors (int64_t knn, double range, int64_t max_threads)
	Creates a new resolver for exact search using the LAESA index. More...
MknnResolverParams *	mknn_predefIndex_Laesa_resolverApproximateNearestNeighbors (int64_t knn, double range, int64_t max_threads, double approximation_percentage)
Snake Table.
MknnIndexParams *	mknn_predefIndex_SnakeTable_indexParams (int64_t num_pivots)
	Resolves searches using a Snake Table. More...
MknnResolverParams *	mknn_predefIndex_SnakeTable_resolverExactNearestNeighbors (int64_t knn, double range, int64_t max_threads)
	Creates a new resolver for exact search using the SnakeTable index. More...
FLANN Indexes.
MknnIndexParams *	mknn_predefIndex_FlannLinearScan_indexParams ()
	Resolves searches using the FLANN's implementation of linear scan algorithm. More...
MknnIndexParams *	mknn_predefIndex_FlannKdTree_indexParams (int64_t num_trees)
	Resolves searches using FLANN's implementation of kd-tree. More...
MknnIndexParams *	mknn_predefIndex_FlannKmeansTree_indexParams (int64_t branching, int64_t iterations)
	Resolves searches using FLANN's implementation of k-means tree. More...
MknnIndexParams *	mknn_predefIndex_FlannLSH_indexParams (int64_t table_number, int64_t key_size, int64_t multi_probe_level)
	Resolves searches using FLANN's implementation of Local Sensitive Hashing. More...
MknnIndexParams *	mknn_predefIndex_FlannAutoIndex_indexParams (double target_precision, double build_weight, double memory_weight, double sample_fraction)
	Resolves searches using FLANN's automatic selection of the best algorithm. More...
MknnResolverParams *	mknn_predefIndex_Flann_resolverExactNearestNeighbors (int64_t knn, int64_t max_threads)
	Creates a new resolver for exact searches using FLANN's indexes. More...
MknnResolverParams *	mknn_predefIndex_Flann_resolverApproximateNearestNeighbors (int64_t knn, int64_t max_threads, int64_t num_checks)
	Creates a new resolver for approximate searches using FLANN's indexes. More...

Detailed Description

MetricKnn provides a set of pre-defined indexes identified.

The complete list of pre-defined indexes provided by MetricKnn can be printed by calling mknn_predefIndex_helpListIndexes. The parameters supported by each index can be printed by calling mknn_predefIndex_helpPrintIndex.

Function Documentation

MknnResolverParams* mknn_predefIndex_Flann_resolverApproximateNearestNeighbors	(	int64_t	knn,
		int64_t	max_threads,
		int64_t	num_checks
	)

Creates a new resolver for approximate searches using FLANN's indexes.

Parameters

knn	number of nearest neighbors to return. A value < 1 means 1.
max_threads	the maximum number of threads to be used.
num_checks	how many leafs to check in one search. A value >= 1, where a value closer to 1 means faster (but lower quality) search.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

MknnResolverParams* mknn_predefIndex_Flann_resolverExactNearestNeighbors	(	int64_t	knn,
		int64_t	max_threads
	)

Creates a new resolver for exact searches using FLANN's indexes.

Parameters

knn	number of nearest neighbors to return. A value < 1 means 1.
max_threads	the maximum number of threads to be used.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

MknnIndexParams* mknn_predefIndex_FlannAutoIndex_indexParams	(	double	target_precision,
		double	build_weight,
		double	memory_weight,
		double	sample_fraction
	)

Resolves searches using FLANN's automatic selection of the best algorithm.

This method is just an interface to FLANN library. See FLANN's documentation http://www.cs.ubc.ca/research/flann/ .

Parameters

target_precision	precision desired
build_weight	build tree time weighting factor
memory_weight	index memory weigthing factor
sample_fraction	what fraction of the dataset to use for autotuning

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnIndexParams* mknn_predefIndex_FlannKdTree_indexParams

(

int64_t

num_trees

)

Resolves searches using FLANN's implementation of kd-tree.

This method is just an interface to FLANN library. See FLANN's documentation http://www.cs.ubc.ca/research/flann/ .

Parameters

num_trees

Amount of trees to build.

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnIndexParams* mknn_predefIndex_FlannKmeansTree_indexParams	(	int64_t	branching,
		int64_t	iterations
	)

Resolves searches using FLANN's implementation of k-means tree.

This method is just an interface to FLANN library. See FLANN's documentation http://www.cs.ubc.ca/research/flann/ .

Parameters

branching	branching factor. Number of centroids at each level.
iterations	max iterations to perform in one kmeans clustering.

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnIndexParams* mknn_predefIndex_FlannLinearScan_indexParams

(

)

Resolves searches using the FLANN's implementation of linear scan algorithm.

This method is just an interface to FLANN library. See FLANN's documentation http://www.cs.ubc.ca/research/flann/ .

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnIndexParams* mknn_predefIndex_FlannLSH_indexParams	(	int64_t	table_number,
		int64_t	key_size,
		int64_t	multi_probe_level
	)

Resolves searches using FLANN's implementation of Local Sensitive Hashing.

This method is just an interface to FLANN library. See FLANN's documentation http://www.cs.ubc.ca/research/flann/ .

Parameters

table_number	The number of hash tables to use
key_size	The length of the key in the hash tables
multi_probe_level	Number of levels to use in multi-probe LSH, 0 for standard LSH

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

void mknn_predefIndex_helpPrintIndex

(

const char *

id_index

)

Prints to standard output the help for a index.

Parameters

id_index

the unique identifier of a pre-defined index.

MknnIndexParams* mknn_predefIndex_Laesa_indexParams

(

int64_t

num_pivots

)

Resolves searches using the LAESA algorithm.

In order to obtain the exact nearest neighbors, the distance distance must satisfy the metric properties, otherwise the exact search will be incorrect.

Parameters

num_pivots

Amount of pivots to compute and store in memory.

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnResolverParams* mknn_predefIndex_Laesa_resolverApproximateNearestNeighbors	(	int64_t	knn,
		double	range,
		int64_t	max_threads,
		double	approximation_percentage
	)

Parameters

knn	number of nearest neighbors to return. A value < 1 means 1.
range	the search range. A value <= 0 or the constant DBL_MAX mean maximum range.
max_threads	the maximum number of threads to be used.
approximation_percentage	The amount of distance evaluation to compute. A value between 0 and 1, where a value closer to 0 means faster (but lower quality) search.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

MknnResolverParams* mknn_predefIndex_Laesa_resolverExactNearestNeighbors	(	int64_t	knn,
		double	range,
		int64_t	max_threads
	)

Creates a new resolver for exact search using the LAESA index.

Parameters

knn	number of nearest neighbors to return. A value < 1 means 1.
range	the search range. A value <= 0 or constant DBL_MAX mean maximum range.
max_threads	the maximum number of threads to be used.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

MknnIndexParams* mknn_predefIndex_LinearScan_indexParams

(

)

Resolves searches using the linear scan algorithm.

The algorithm compares each query object to each object in search_dataset sequentially and retrieves the nearest objects.

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnResolverParams* mknn_predefIndex_LinearScan_resolverExactFarthestNeighbors	(	int64_t	knn,
		int64_t	max_threads
	)

Creates a new resolver for exact search of the K farthest neighbors using the linear scan index.

Parameters

knn	number of farthest neighbors to return. A value < 1 means 1.
max_threads	the maximum number of threads to be used.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

MknnResolverParams* mknn_predefIndex_LinearScan_resolverExactNearestNeighbors	(	int64_t	knn,
		double	range,
		int64_t	max_threads
	)

Creates a new resolver for exact search of the K nearest neighbors using the linear scan index.

Parameters

knn	number of nearest neighbors to return. A value < 1 means 1.
range	the search range. A value <= 0 or constant DBL_MAX (defined in float.h) mean maximum range.
max_threads	the maximum number of threads to be used.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

MknnIndexParams* mknn_predefIndex_SnakeTable_indexParams

(

int64_t

num_pivots

)

Resolves searches using a Snake Table.

In order to obtain the exact nearest neighbors, the distance distance must satisfy the metric properties, otherwise the exact search will be incorrect.

Parameters

num_pivots

Amount of pivots to store in memory.

Returns

parameters to create an index (it must be released with mknn_indexParams_release or bound to the new index)

MknnResolverParams* mknn_predefIndex_SnakeTable_resolverExactNearestNeighbors	(	int64_t	knn,
		double	range,
		int64_t	max_threads
	)

Creates a new resolver for exact search using the SnakeTable index.

Parameters

knn	number of nearest neighbors to return. A value < 1 means 1.
range	the search range. A value <= 0 or constant DBL_MAX mean maximum range.
max_threads	the maximum number of threads to be used.

Returns

parameters to create a resolver for the given similarity search (it must be released with mknn_resolverParams_release or bound to the new resolver)

bool mknn_predefIndex_testIndexId

(

const char *

id_index

)

Tests whether the given string references a valid pre-defined index.

Parameters

id_index

the unique identifier of a pre-defined index.

Returns

true whether id_index corresponds to a pre-defined index, and false otherwise.