README.md

Clustering Algorithms

Overview

Features \ Algs	DAOC	SCP	Louvain	Oslom2	GANXiS	pSCAN	CGGCi_RG	SCD
Hierarchical	+		+	+
Multi-scale	+	+	+	+	+
Deterministic	+	+				?		?
With Overlaps	+	+		+	+	+	*
With Weights	+	+	+	*	+
Parameter-Free	+*		+	*	*		*	*
Consensus/Ensemble	+			+			+

• With Overlaps
  * means non-overlapping clusters are produced but the algorithm can be modified to output the overlapping clusters.
• With Weights
  * means weighted networks are supported.
• Parameter-Free
  * means availability of default values for all parameters,
  +* means parameter-free clustering algorithm with optional parameters for the data preprocessing or output post-processing.

Origins

• DAOC (former and fully redesigned HiReCS);
• SCP (Sequential algorithm for fast clique percolation), paper: A sequential algorithm for fast clique percolation;
• Louvain (original and igraph implementations), paper: Fast unfolding of communities in large networks;
• Oslom2, paper: Finding Statistically Significant Communities in Networks;
• GANXiS/SLPA (not uploaded into the repository, because it was provided by the author Jerry Xie for "academic use only"), paper: Extension of Modularity Density for Overlapping Community Structure;

  GANXiS requires preliminary created output directory if it is specified in the options, but GANXiS always creates also default "./output/" directory, which is empty if the custom one is used.

• pSCAN binaries provided by the author, paper: pSCAN : Fast and Exact Structural Graph Clustering;
• CGGCi_RG, paper: An Ensemble Learning Strategy for Graph Clustering;
• SCD, paper: High Quality, Scalable and Parallel Community Detection for Large Real Graphs;
• randcommuns generates random communities (clusters) having the following properties: the number of nodes in each cluster and the number of clusters are taken from the ground-truth.