In bdabbs13/CIDnetworks: Generative Models for Complex Networks with Conditionally Independent Dyadic Structure

Fitting SBMs Using CIDnetworks

Loading Our Data

We will go through the example we just saw in the slides

library(igraph)
library(CIDnetworks)
load("samplenet.Rdata")

plot(graph.adjacency(samplenet), vertex.size=12, vertex.frame.color=NA, vertex.color="gold", edge.arrow.size=0.5)

Fitting a 5 block SBM

## Before we fit the blockmodel, we need to specify a few things
## We need to specify what we think the mean and variance values would be for the B matrix
## It's not required but we get better model fits this way
## The other confusing thing is this is on the probit scale
M5 <- matrix(-2, 5, 5) 
diag(M5) <- 0.67
V5 <- matrix(0.0625, 5, 5)
diag(V5) <- 0.09

fit5= CID.Gibbs(samplenet, components = c(SBM(5, block.matrix.m = M5, block.matrix.v = V5)), draws = 400, burnin=1000, thin = 20)

summary(fit5)
plot(fit5)

summary(sbm.mod)[[6]]$modal.membership

est.blocks=summary(fit5)[[6]]$modal.membership
plot(graph.adjacency(samplenet), vertex.size=12, vertex.frame.color=NA, vertex.color=est.blocks, edge.arrow.size=0.5)

Fitting a 6, 7, or 8 block SBM

M6 <- matrix(-2, 6, 6) 
diag(M6) <- 0.67
V6 <- matrix(0.0625, 6, 6)
diag(V6) <- 0.09

fit6= CID.Gibbs(samplenet, components = c(SBM(6, block.matrix.m = M6, block.matrix.v = V6)), draws = 400, burnin=1000, thin = 20)

summary(fit6)
plot(fit6)

est.blocks=summary(fit6)[[6]]$modal.membership
plot(graph.adjacency(samplenet), vertex.size=12, vertex.frame.color=NA, vertex.color=est.blocks, edge.arrow.size=0.5)

Generating our own network data to play with

## I wrote a function to generate network data for you
source("GenerateSBM.r")

## The three inputs of this function are
## n.nodes, n.blocks, bmatrix or vector(diagonal entries, off diagonal entries)

mySBMdata=generateSBMdata(20, 4, c(0.7, 0.1))
mynetwork=mySBMdata$network
true.mem=mySBMdata$trueblock
plot(graph.adjacency(mynetwork), vertex.color=true.mem)


## Fitting a SBM
M4 <- matrix(-2, 4, 4) 
diag(M4) <- 0.67
V4 <- matrix(0.0625, 4, 4)
diag(V4) <- 0.09

mySBM= CID.Gibbs(mynetwork, components = c(SBM(4, block.matrix.m = M4, block.matrix.v = V4)), draws = 400, burnin=1000, thin = 20)

summary(mySBM)
plot(mySBM)

est.blocks=summary(mySBM)[[6]]$modal.membership
plot(graph.adjacency(mynetwork), vertex.size=12, vertex.frame.color=NA, vertex.color=est.blocks, edge.arrow.size=0.5)

On Your Own

## Trying this out on your own
# Generating a network with 30 nodes and 4 blocks
# Within block probability of a tie is 0.6
# Between block probability of a tie is 0.05
simdata=generateSBMdata(30, 4, b=c(0.6, 0.05))
plot(graph.adjacency(simdata$network), vertex.size=8, vertex.label=NA, vertex.frame.color=NA, vertex.color=simdata$trueblock, edge.arrow.size=0.5)

mynetwork=simdata$network

M4 <- matrix(-2, 4, 4) 
diag(M4) <- 0.67
V4 <- matrix(0.0625, 4, 4)
diag(V4) <- 0.09

mySBM= CID.Gibbs(mynetwork, components = c(SBM(4, block.matrix.m = M4, block.matrix.v = V4)), draws = 400, burnin=1000, thin = 20)

summary(mySBM)
plot(mySBM)

est.blocks=summary(mySBM)[[6]]$modal.membership
plot(graph.adjacency(mynetwork), vertex.size=12, vertex.frame.color=NA, vertex.color=est.blocks, edge.arrow.size=0.5)

bdabbs13/CIDnetworks documentation built on Nov. 15, 2019, 2:41 a.m.