VBLPCM-package: VBLPCM: Variational Bayes for the Latent Position Cluster...

In VBLPCM: Variational Bayes Latent Position Cluster Model for Networks

VBLPCM: Variational Bayes for the Latent Position Cluster Model for networks

Description

A faster approximate alternative to using latentnet. Interfaces C code to fit a Variational Bayes approximation to the posterior for the Latent Position Cluster Model for networks.

Details

Package:	VBLPCM
Type:	Package
Version:	2.4.9
Date:	2023-03-22
License:	GPL (>=2)
LazyLoad:	yes

This package is designed to be used as an alternative to the latentnet package when network size computationally prohibits latentnet. It uses a Variational Bayesian Expectation Maximisation algorithm to compute a closed-form approximation to the posterior that the ergmm function in latentnet samples from. It may be thought of as an intermediary approximation that is more accurate than the two-stage MLE fit provided by latentnet but a faster approximation to the MCMC sampler provided by latentnet. In fact, the VB iterations also converge quicker than the two-stage MLE.

VBLPCM can also take advantage of the stratified sampler of Adrian Raftery, Xiaoyue Niu, Peter Hoff and Ka Yee Yeung. This approximation to the (log)likelihood allows for even larger networks to be analysed (see tech report below). Rather than using a fixed number of "controls" per geodesic distance we set a probability of sampling each non-link at each level.

We also provide four choices of model; these are "plain" and three with random node-specific social effects. "rsender" for sender random effects, "rreceiver" for receiver random effects and "rsocial" for both. For undirected networks only "plain" or "rsocial" may be chosen.

References

Michael Salter-Townshend and Thomas Brendan Murphy (2013). "Variational Bayesian Inference for the Latent Position Cluster Model." Computational Statistics and Data Analysis, volume 57, number 1, pages 661-671. DOI=10.1016/j.csda.2012.08.004

Pavel N. Krivitsky and Mark S. Handcock (2008). "Fitting Latent Cluster Models for Social Networks with latentnet." Journal of Statistical Software, number 5, volume 24, pages 1-23.

Mark S. Handcock, Adrian E. Raftery and Jeremy Tantrum (2007). "Model-Based Clustering for Social Networks." Journal of the Royal Statistical Society: Series A (Statistics in Society), 170(2), 301-354.

Adrian Raftery, Xiaoyue Niu, Peter Hoff and Ka Yee Yeung (2012). "Fast Inference for the Latent Space Network Model Using a Case-Control Approximate Likelihood." Journal of Computational and Graphical Statistics. doi: 10.1080/10618600.2012.679240

Sucharita Gopal (2007). "The Evolving Social Geography of Blogs" Societies and Cities in the Age of Instant Access Berlin:Springer, 275–294

See Also

vblpcmstart vblpcmfit

Examples

### Sampson's monks with sender random effects ###
data(sampson,package="VBLPCM")
v.start<-vblpcmstart(samplike,G=3,model="rreceiver",LSTEPS=1e3)
v.fit<-vblpcmfit(v.start,STEPS=20)
### plot the mean posterior positions ###
plot(v.fit, R2=0.05,main="Sampson's Monks: VB with Receiver Random Effects")
### Who's in each group?  ###
vblpcmgroups(v.fit)

### Look at a goodness-of-fit plot ###
plot(gof(v.fit,GOF=~distance))

### create a matrix of link posterior probabilities given the fitted model ###
probs<-predict.vblpcm(v.fit)
### create a boxplot goodness-of-fit graphic ###
boxplot(split(probs,as.sociomatrix(samplike)))

### run a bigger example, using the likelihood sampler set to 0.1 ###
## Not run: 
data(aids,package="VBLPCM")
v.start<-vblpcmstart(aids.net,G=7,model="rsender",d=3)
use the case-control sampler with 10 controls per case
v.fit<-vblpcmfit(v.start,NC=10)
plot the mean posterior positions ###
plot(v.fit, R2=0.1,main="Aids Blogs with Sender Random Effects")

### Use ROC / AUC to get a measure of model fit to the data ###
vblpcmroc(v.fit)

## End(Not run)

VBLPCM documentation built on March 31, 2023, 9:21 p.m.