Skip to contents

Estimate a colSBM on a collection of networks

estimate_colSBM.Rd

Estimate a colSBM on a collection of networks

Usage

estimate_colSBM(
  netlist,
  colsbm_model,
  net_id = NULL,
  directed = NULL,
  distribution = "bernoulli",
  fit_sbm = NULL,
  nb_run = 3L,
  global_opts = list(),
  fit_opts = list(),
  Z_init = NULL,
  fit_init = NULL
)

Arguments

netlist

A list of matrices.

colsbm_model

Which colSBM to use, one of "iid", "pi", "delta", "deltapi".

net_id

A vector of string, the name of the networks.

directed

A boolean, are the networks directed or not.

distribution

A string, the emission distribution, either "bernoulli" (the default) or "poisson"

fit_sbm

A list of fitted model using the sbm package. Use to speed up the initialization.

nb_run

An integer, the number of run the algorithm do.

global_opts

Global options for the outer algorithm and the output

fit_opts

Fit options for the VEM algorithm

Z_init

An optional list of cluster memberships for q in 1:Q_max. Default to NULL.

fit_init

Do not use! Optional fit init from where initializing the algorithm.

Value

A bmpop object listing a collection of fitted models for the collection of networks

Details

The list of parameters global_opts essentially tunes the exploration process.

  • nb_cores integer for number of cores used for parallelization. Default is 1.

  • sbm_init boolean specifying wether {sbm} package should be used to initialize the algorithm.

  • spectral_init boolean specifying wether a spectral clustering should be used to initialize the algorithm.

  • nb_init the number of different inits to perform to start the algorithm fit. Default to 10.

  • verbosity integer for verbosity (0, 1, 2, 3, 4). Default is 1. 0 will disable completely the output of the function. modelisation is preferred

  • Q_max integer for the max size to explore. Default is computed with the following formula: floor(log(sum(sapply(netlist, function(A) nrow(A)))) + 2)

  • nb_models the number of models to keep for each values of Q. Default is 5.

  • depth specifies how far around the best model the exploration will fit models. Default is 3.

  • plot_details integer to control the display of the exploration process. Values are 0 or 1. Default is 1.

  • max_pass the maximum number of passes that will be executed. Default is 10.

  • backend the parallelization backend to use. Options available are "no_mc", "future", "parallel". Default is "future". Note: we plan to unify everything behind future.

See also

clusterize_unipartite_networks(), bmpop, fitSimpleSBMPop, browseVignettes("colSBM")

Examples

# Trivial example with Gnp networks:
Net <- lapply(
  list(.7, .7, .2, .2),
  function(p) {
    A <- matrix(0, 15, 15)
    A[lower.tri(A)][sample(15 * 14 / 2, size = round(p * 15 * 14 / 2))] <- 1
    A <- A + t(A)
  }
)
if (FALSE) { # \dontrun{
cl <- estimate_colSBM(Net,
  colsbm_model = "delta",
  directed = FALSE,
  distribution = "bernoulli",
  nb_run = 1
)
} # }