README.md

Solving Clustering Problems

• Solving Clustering Problems
  • Algorithms
    • K-means
    • ISODATA
    • Hierarchical clustering
    • First come clustering
  • Auxiliary interfaces
    • Problem interface IClusteringData
    • Algorithm interface IClustering
    • Clustering result CClusteringResult

NeoML library provides several methods for clustering data.

Algorithms

K-means

K-means method is the most popular clustering algorithm. It assigns each object to the cluster with the nearest center. Implemented by the CKMeansClustering class.

ISODATA

ISODATA clustering algorithm is based on geometrical proximity of the data points. The clustering result will depend greatly on the initial settings. Implemented by the CIsoDataClusteringclass.

Hierarchical clustering

The library provides a "naive" implementation of upward hierarchical clustering. First, it creates a cluster per element, then merges clusters on each step until the final cluster set is achieved. Implemented by the CHierarchialClustering class.

First come clustering

A simple clustering algorithm that creates a new cluster for each new vector that is far enough from the clusters already existing. Implemented by the CFirstComeClustering class.

Auxiliary interfaces

Problem interface IClusteringData

The input data to be split into clusters is passed to any of the algorithms as a pointer to the object that implements the IClusteringData interface:

class IClusteringData : public virtual IObject {
public:
	// The number of vectors
	virtual int GetVectorCount() const = 0;

	// The number of features
	virtual int GetFeaturesCount() const = 0;

	// Gets all input vectors as a matrix of size GetVectorCount() x GetFeaturesCount()
	virtual CFloatMatrixDesc GetMatrix() const = 0;

	// Gets the vector weight
	virtual double GetVectorWeight( int index ) const = 0;
};

Algorithm interface IClustering

Every clustering algorithm implements the IClustering interface.

class IClustering {
public:
	virtual ~IClustering() {};

	// Clusterizes the input data 
	// and returns true if successful with the given parameters
	virtual bool Clusterize( IClusteringData* data, CClusteringResult& result ) = 0;
};

Clustering result CClusteringResult

The clustering result is described by the CClusteringResult structure.

class NEOML_API CClusteringResult {
public:
	int ClusterCount;
	CArray<int> Data;
	CArray<CClusterCenter> Clusters;
};

• ClusterCount — the number of clusters
• Data — the array of cluster numbers for each of the input data elements (the clusters are numbered from 0 to ClusterCount - 1)
• Clusters — the cluster centers