fastLiclust: Hierarchical clustering on a linkage matrix

In meowcat/fastliclust: Fast linkage clustering

Description

Performs hierarchical clustering (currently only average linkage is supported) on a dataset which is only partially interconnected (i.e. between some vertices there is no edge.) This is in particular a memory-saving alternative on large sparse graphs which are only encoded in the linkage matrix, i.e. have no nxn distance matrix. Other use cases are sparsely populated matrices which can be extracted directly from ff objects on disk, so no nxn matrix has to be built in RAM.

Usage

fastLiclust(linkmat, sim, weights, disconnect = 1)

Arguments

linkmat	A nx2 integer matrix of vertices which are connected.
sim	A length n numeric vector of the distances between the connected edges.
weights	A n integer vector initialized with all 1, which will contain the number of vertices summarized under a cluster (necessary in the algorithm.)

Value

Nothing - the operation is performed in place on the matrix! Note this is highly unstandard R behavior. To get the results in a useful way, run first crop then toHclust.

meowcat/fastliclust documentation built on May 22, 2019, 6:51 p.m.