hikmeans.h File Reference

Hierarchical integer K-Means clustering. More...

#include "generic.h"
#include "ikmeans.h"

Data Structures
struct	VlHIKMNode
	HIKM tree node. More...
struct	VlHIKMTree
	HIKM tree. More...

Functions
Create and destroy
VlHIKMTree *	vl_hikm_new (int method)
	New HIKM tree.
void	vl_hikm_delete (VlHIKMTree *f)
	Delete HIKM tree.
Retrieve data and parameters
int	vl_hikm_get_ndims (VlHIKMTree const *f)
	Get data dimensionality.
int	vl_hikm_get_K (VlHIKMTree const *f)
	Get K.
int	vl_hikm_get_depth (VlHIKMTree const *f)
	Get depth.
int	vl_hikm_get_verbosity (VlHIKMTree const *f)
	Get verbosity level.
int	vl_hikm_get_max_niters (VlHIKMTree const *f)
	Get maximum number of iterations.
VlHIKMNode const *	vl_hikm_get_root (VlHIKMTree const *f)
	Get maximum number of iterations.
Set parameters
void	vl_hikm_set_verbosity (VlHIKMTree *f, int verb)
	Set verbosity level.
void	vl_hikm_set_max_niters (VlHIKMTree *f, int max_niters)
	Set maximum number of iterations.
Process data
void	vl_hikm_init (VlHIKMTree *f, int M, int K, int depth)
	Initialize HIKM tree.
void	vl_hikm_train (VlHIKMTree *f, vl_uint8 const *data, int N)
	Train HIKM tree.
void	vl_hikm_push (VlHIKMTree *f, vl_uint *asgn, vl_uint8 const *data, int N)
	Project data down HIKM tree.

---

Detailed Description

Hierarchical Integer K-Means Clustering.

Hierarchical integer K-Means clustering (HIKM) is a simple hierarchical version of integer K-Means (IKM). The algorithm recursively applies integer K-means to create more refined partitions of the data.

Create a tree with vl_hikm_new() and delete it with vl_hikm_delete(). Use vl_hikm_train() to build the tree from training data and vl_hikm_push() to project new data down a HIKM tree.

HIKM tree

The HIKM tree is represented by a VlHIKMTree structure, which contains a tree composed of VlHIKMNode. Each node is an integer K-means filter which partitions the data into K clusters.

Author:

Brian Fulkerson

---

Function Documentation

void vl_hikm_delete

(

VlHIKMTree *

f

)

Parameters:

f	HIKM tree.

int vl_hikm_get_depth ( VlHIKMTree const *  f )

inline

Parameters:

f	HIKM tree.

Returns:

depth.

int vl_hikm_get_K ( VlHIKMTree const *  f )

inline

Parameters:

f	HIKM tree.

Returns:

K.

int vl_hikm_get_max_niters ( VlHIKMTree const *  f )

inline

Parameters:

f	HIKM tree.

Returns:

maximum number of iterations.

int vl_hikm_get_ndims ( VlHIKMTree const *  f )

inline

Parameters:

f	HIKM tree.

Returns:

data dimensionality.

VlHIKMNode const * vl_hikm_get_root ( VlHIKMTree const *  f )

inline

Parameters:

f	HIKM tree.

Returns:

maximum number of iterations.

int vl_hikm_get_verbosity ( VlHIKMTree const *  f )

inline

Parameters:

f	HIKM tree.

Returns:

verbosity level.

void vl_hikm_init	(	VlHIKMTree *	f,
		int	M,
		int	K,
		int	depth
	)

Parameters:

f	HIKM tree.
M	Data dimensionality.
K	Number of clusters per node.
depth	Tree depth.

Remarks:

depth cannot be smaller than 1.

Returns:

a new HIKM tree representing the clustering.

VlHIKMTree* vl_hikm_new

(

int

method

)

Parameters:

method

clustering method.

Returns:

new HIKM tree.

void vl_hikm_push	(	VlHIKMTree *	f,
		vl_uint *	asgn,
		vl_uint8 const *	data,
		int	N
	)

Parameters:

f	HIKM tree.
asgn	Path down the tree (out).
data	Data to project.
N	Number of data.

The function writes to asgn the path of the data data down the HIKM tree f. The parameter asgn must point to an array of M by N elements, where M is the depth of the HIKM tree and N is the number of data point to process.

void vl_hikm_set_max_niters ( VlHIKMTree *  f, int  max_niters  )

inline

Parameters:

f	HIKM tree.
max_niters	maximum number of iterations.

void vl_hikm_set_verbosity ( VlHIKMTree *  f, int  verb  )

inline

Parameters:

f	HIKM tree.
verb	verbosity level.

void vl_hikm_train	(	VlHIKMTree *	f,
		vl_uint8 const *	data,
		int	N
	)

Parameters:

f	HIKM tree.
data	Data to cluster.
N	Number of data.