wfg (version 0.1)

wfg: Weighted Fast Greedy Algorithm

Description

Implementation of weighted fast greedy algorithm for community detection in networks with mixed types of attributes.

Usage

wfg(net, attr=NULL, under.sample=FALSE, prioritize=FALSE)

Arguments

net

network for community detection

attr

data frame of attribute information. The default value is NULL, when no attribute information will be used. Under default this method is identical to fast greedy community detection algorithm.

under.sample

a boolean parameter. When it is TRUE, the vertex pairs without links will be under-sampled to have the same number as that of the linked pairs of vertices.

prioritize

a boolean parameter. When it is TRUE, a matrix of cummunity-specific coefficients will be returned, by which the communities can be prioritized.

Value

• beta
  Estimates of coefficients from logistic regression.
• beta.matrix
  Estimates of community specific coefficients.
• memb
  Community membership of vertices.

Details

Each column of attr data frame can be a vector with type of either numeric (continuous) or factor (categorical). The matrix of cummunity-specific coefficients gives estimates as to the relative homogeneity of each attribute within each community. Specifically, a negative beta with large absolute value indicates corresponding attribute is homogeneous.

References

Clauset, Aaron, Mark EJ Newman, and Cristopher Moore. "Finding community structure in very large networks." Physical review E 70.6 (2004): 066111.

Examples

##### implementation of wfg on a computer generated network with
##### structually relevant continuous attribute and irrelevant categorical attribute
set.seed(7)
##### set network properties
## four groups, each with 32 vertices
nv <- c(32,32,32,32)
l <- length(nv)
## obtain p.in and p.out from z.out
z.out <- 6
z.in <- 16-z.out
p.out <- z.out/96
p.in <- rep(z.in/31, l)

##### simulate network
net.simu <- network.simu(nv=nv, p.in=p.in, p.out=p.out, p.del=0)
net <- net.simu$net
group <- net.simu$group

##### simulate attributes
## separation of continuous attribute
delta <- 1
## p's for the multinomial distribution of categorical attributes
p1 <- 0.25
p2 <- (1-p1)/3
attr1 <- c(rnorm(nv[1],0), rnorm(nv[2],1*delta), rnorm(nv[3],2*delta), rnorm(nv[4],3*delta))
attr2 <- c(sample(c(1,2,3,4), size=nv[1], prob=c(p1, p2, p2, p2), replace=TRUE),
           sample(c(1,2,3,4), size=nv[2], prob=c(p2, p1, p2, p2), replace=TRUE),
           sample(c(1,2,3,4), size=nv[3], prob=c(p2, p2, p1, p2), replace=TRUE),
           sample(c(1,2,3,4), size=nv[4], prob=c(p2, p2, p2, p1), replace=TRUE))
attributes <- data.frame(attr1, attr2)

##### implementation of wfg
wfg.result <- wfg(net=net, attr=attributes, under.sample = FALSE, prioritize = TRUE)

##### plot network colored by wfg result
V(net)$size <- 7
V(net)$color <- wfg.result$memb
plot(net, vertex.label='')