vblpcmfit {VBLPCM}

fit the variational model through EM type iterations

Description

Perform optimisation of the variational parameters of the variational approximation to the posterior for the latent position cluster model for network data.

Usage

vblpcmfit(variational.start, STEPS = 50, maxiter = 100, tol=1e-6, NC=NULL, 
           seed=NaN, d_vector=rep(TRUE,9))

Arguments

variational.start

The starting configuration; use vblpcmstart() to generate this.

STEPS

Maximum number of iterations in the main VBEM loop.

maxiter

Maximum number of iterations for the internal univariate optimisation loops.

tol

tolerance of change in variational parameter updates below which the algorithm is deemed to have converged for that parameter.

NC

Number of non-links sampled in the case-control type sampler. Results in a speedup but loss of accuracy.

seed

Optional seed for the random number generator. Supplying NaN is equivalent to not supplying it. Supply a value so that results may be replicated.

d_vector

Optional logical vector specifying which sets of variational parameters are to be updated. See Details for more information.

Details

d_vector is a logical vector of length 9 that can be used to select which variational parameters are held fixed and which are updated. The parameters are in the following order: z (latent positions), sigma2 (variance of latent positions), lambda (membership probability matrix), eta (cluster centres), omega2 (cluster variances), alpha (cluster specific variance of nodes), nu (Dirichlet parameter for marginal cluster probabilities), xi (likelihood intercept term mean), psi2 (likelihood intercept term variance).

Value

A v.params list containing the fitted variational parameters for the latent positions, clustering membership probabilities, etc. conv indicated whether convergence was obtained within the specified number of iterations.

Author(s)

Michael Salter-Townshend

References

Michael Salter-Townshend and Thomas Brendan Murphy (2009). "Variational Bayesian Inference for the Latent Position Cluster Model." Workshop on Analyzing Networks and Learning with Graphs. Neural Information Processing Systems.

See Also

vblpcmstart, latentnet::ergmm