leidenAlg (version 1.1.5)

find_partition_with_rep: Finds the optimal partition using the Leiden algorithm with replicate starts

Description

This function performs Leiden algorithm nrep times and returns the result from the run with the maximum quality.

Since Leiden algorithm has stochastic process, repeating stochastically may improve the result. However, users should be aware of whether there is indeed a community structure with exploration, rather than blindly trusting the returned result that comes with the highest quality value.

The random number generator (RNG) is not re-seeded at each new start of community detection, in order to keep the independence of each replicate. To get reproducible result, users can run set.seed() before calling these functions.

find_partition only performs the community detection once and the reproducibility can also be ensured with set.seed().

Usage

find_partition_with_rep(
  graph,
  edge_weights,
  resolution = 1,
  niter = 2,
  nrep = 10
)

Value

A vector of membership values

Arguments

graph

The igraph graph to define the partition on

edge_weights

Vector of edge weights. In weighted graphs, a real number is assigned to each (directed or undirected) edge. For an unweighted graph, this is set to 1. Refer to igraph, weighted graphs.

resolution

Numeric scalar, resolution parameter controlling communities detected (default=1.0) Higher resolutions lead to more communities, while lower resolutions lead to fewer communities.

niter

Number of iterations that the algorithm should be run for (default=2)

nrep

Number of replicate starts with random number being updated. (default=10) The result with the best quality will be returned.

Examples

library(igraph)

# To run 10 replicates and get the partitioning with the highest quality
membership <- find_partition_with_rep(exampleGraph, E(exampleGraph)$weight, nrep = 10)

# To get reprodicible result for every function call, do `set.seed()` right before calling
set.seed(233)
res1 <- find_partition_with_rep(exampleGraph, E(exampleGraph)$weight, resolution = 2)
# Here, no seed was set...
res2 <- find_partition_with_rep(exampleGraph, E(exampleGraph)$weight, resolution = 2)
set.seed(233)
res3 <- find_partition_with_rep(exampleGraph, E(exampleGraph)$weight, resolution = 2)
identical(res1, res2) # FALSE (usually), as no seed as set
identical(res1, res3) # TRUE (always), as set.seed() was used directly before the function call