#include <clizSorter.h>

Public Member Functions
	Sorter (SorterHandle iHandle, std::vector< Save * > &iInputSaveVec, std::vector< Save * > &iOutputSaveVec, std::vector< IndicesStruct > iIndicesOut)

void	calcDist (float &oDist, unsigned int cluster1, unsigned int cluster2)

int	fusionPossible (unsigned int idMaster, unsigned int idSlave, float diffCrit)

void	feedSpikeFeatures (float iFeatures, uint64_t iTimestamps, int iNspk)

void	clusteringLoop (unsigned int newClusters)

void	fuseCluster (unsigned int idMaster, unsigned int idSlave)

void	killCluster (unsigned int iClusterIndex)

void	killTheList ()

void	getClusterCount (int &oClusterCount)

void	threadFunc ()

	~Sorter ()

Detailed Description

Modules Except for the Accumulator and the Communicator, all Modules share the same structure. For initialization, each Module needs a parameter handle, an std::vector of Saves for their input, an std::vector of Saves for their output and an std::vector of IndicesStructs to know which output Saves require which output. During initialization a worker thread is spawned that tries to receive input from all input Saves. Once one ChunkPtrStruct from each input Save was obtained, functions are called that perform the respective jobs of the module. For each output Save, based on the respective IndicesStruct, the data obtained from these functions is copied and a ChunkPtrStruct that holds a pointer to that data is pushed in the output Save.Module: Sorter For each new spike, a new cluster is generated holding just 1 spike. For each cluster pair, its euclidean distance (after normalization and weighting) is determined. From the closest to the most distant pair, they are tested for fusion using specific criteria: Distance < thr_min => Fuse (thr_min = fusionCriteriaParams[0]) Distance > thr_max => Don't fuse and don't test more distant pairs (thr_max = fusionCriteriaParams[1]) max(abs(AC-CC)) < thr_accc => Don't fuse (thr_accc = fusionCriteriaParams[2]) As we pick different features, their correlation might be too weak. Thus, to improve the correlation-based criterium, selected features of the greater cluster's representative are first ordered by their value (and features of the smaller cluster's representative are sorted accordingly (so that the order of feature indices is equal for both cluster's representatives). This increases correlation and makes differences appear more clearly. Each time 2 clusters are fused, the sorting loop starts again, thereby only fusing the closest possible pairs Sorting stops when no more clusters can be fused or the number of clusters reached min_clusters

Constructor & Destructor Documentation

cliz::Sorter::Sorter	(	SorterHandle	iHandle,
		std::vector< Save * > &	iInputSaveVec,
		std::vector< Save * > &	iOutputSaveVec,
		std::vector< IndicesStruct >	iIndicesOut
	)

Module: Sorter Constructor

Here is the call graph for this function:

cliz::Sorter::~Sorter ( )

inline

Here is the call graph for this function:

Member Function Documentation

void cliz::Sorter::calcDist	(	float &	oDist,
		unsigned int	cluster1,
		unsigned int	cluster2
	)

inline

Module: Sorter Calculating the distance between two cluster representatives

Parameters

[out]	oDist	Distance between the representatives of cluster1 and cluster2 (considering only the selected features)
[in]	cluster1	Array position of cluster1
[in]	cluster2	Array position of cluster2

Here is the caller graph for this function:

void cliz::Sorter::clusteringLoop ( unsigned int newClusters )

Module: Sorter Main clustering loop

Parameters

[in] newClusters Number of new clusters

Here is the call graph for this function:

Here is the caller graph for this function:

void cliz::Sorter::feedSpikeFeatures	(	float *	iFeatures,
		uint64_t *	iTimestamps,
		int	iNspk
	)

Module: Sorter Feeding the Sorter with new spikes

Parameters

[in]	iFeatures	features of new spikes(iNspk*numAllFeatures)
[in]	iTimestamps	timestamps of new spikes
[in]	iNspk	number of new spikes

Here is the call graph for this function:

Here is the caller graph for this function:

void cliz::Sorter::fuseCluster	(	unsigned int	idMaster,
		unsigned int	idSlave
	)

Module: Sorter Fusing two clusters into one. The new cluster will have the ID of the greater cluster

Parameters

[in]	idMaster	Array position of the greater cluster
[in]	idSlave	Array position of the smaller cluster

Here is the call graph for this function:

Here is the caller graph for this function:

int cliz::Sorter::fusionPossible	(	unsigned int	idMaster,
		unsigned int	idSlave,
		float	diffCrit
	)

Module: Sorter Determining if two clusters should be fused

Parameters

[in]	diffCrit	Distance between the representatives of Master and Slave (considering only the selected features)
[in]	idMaster	Array position of the greater cluster
[in]	idSlave	Array position of the smaller cluster