Graph Agglomerative Clustering (GAC) toolbox

A better alternative to conventional algorithms, such as k-means, spectral clustering and linkage

Gactoolbox is a summary of our research of agglomerative clustering on a graph. Agglomerative clustering, which iteratively merges small clusters, is commonly used for clustering because it is conceptually simple and produces a hierarchy of clusters. Classical agglomerative clustering algorithms, such as average linkage and DBSCAN, were widely used in many areas. Those algorithms, however, are not designed for clustering on a graph. This toolbox implements the following algorithms for agglomerative clustering on a directly graph.
1) Structural descriptor based algorithms (gacCluster.m). We define a cluster descriptor based on the graph structure, and each merging is determined by maximizes the increment of the descriptor. Two descriptors, including zeta function and path integral, are implemented. You can also design new descriptor (creating functions similar to gacPathEntropy.m and gacPathCondEntropy.m) and develop new algorithms with our code.
2) Graph degree linkage (gdlCluster.m). It is a simple and effective algorithm, with better performance than normalized cuts and spectral clustering, and is faster.