vblpcmstart: Generate sensible starting configuration for the variational parameter set.

Description

Uses fast methods to generate sensible and coherent values for the parameters of the variational method. There are returned as a list and that list may be passed directly to vblpcmfit(). User specification of the configuration is recommended as tweaks to this list only.

Usage

vblpcmstart(g.network,G=1,d=2,LSTEPS=5e3,model="plain", CLUST=0, B=NULL,
           lcc=TRUE, edgecovs=NULL,sendcovs=NULL,receivecovs=NULL,
           socialcovs=NULL,START="FR", seed=NaN)

Value

A v.params list containing the latent positions, clustering membership probabilities, etc.

Arguments

g.network: a network object
G: Desired number of groups
d: Desired dimensionality of the latent space
LSTEPS: Number of steps in the log-likelihood forces algorithm
model: model specified as "plain", "rreceiver", "rsender" or "rsocial". See vblpcmcovs for details.
CLUST: degree of push to clustering at the start
B: default intercept value
lcc: logical indicator. TRUE => analyze largest connected component of g.network only FALSE => analyze the whole network.
edgecovs: optional edge covariates.
sendcovs: optional sender node covariates.
receivecovs: optional receiver node covariates.
socialcovs: optional sociality node covariates.
START: what to start the initial positions with. "FR" for Fruchterman-Reingold. "geodesic" for geodesic distances. "laplace" for using the Graph Laplacian. "random" for random.
seed: Optional seed for the random number generator in R. Equivalent to using set.seed(seed). The default NaN value does not call set.seed().

Author

Michael Salter-Townshend

Examples

Run this code

data(sampson)
### plot the mean posterior positions with initial estimations for variational parameters
plot(vblpcmstart(samplike,G=3),main="Sampson's Monks: VB Initial Values")
### plot the mean posterior positions with final estimations for variational parameters
plot(vblpcmfit(vblpcmstart(samplike,G=3)),main="Sampson's Monks: VB Solution")