Detect variable communities in a catgraph using graph clustering algorithms

Applies a community detection algorithm to the weighted undirected graph of a catgraph object. Assigns cluster membership as a vertex attribute and returns the updated object. The Louvain algorithm (Blondel et al., 2008) is the default as it is fast, handles weighted edges, and requires no pre-specification of the number of clusters.

Usage

detect_clusters(
  x,
  method = c("louvain", "walktrap", "fast_greedy", "label_prop", "edge_betweenness"),
  resolution = 1,
  steps = 4L
)

Arguments

x: A catgraph object.
method: Character. Community detection algorithm to use. One of "louvain" (default), "walktrap", "fast_greedy", "label_prop", or "edge_betweenness". All methods use edge weights when available.
resolution: Numeric. Resolution parameter for the Louvain method (igraph::cluster_louvain resolution argument). Higher values favour smaller communities. Default 1.
steps: Integer. Number of random walk steps for the Walktrap method. Default 4L.

Value

The input catgraph object with two additions:

graph: Vertex attribute cluster (integer membership vector) and cluster_method (character) are added to the graph.
clustering: A named list with the raw igraph communities object (communities), the membership vector (membership), the number of communities (n_clusters), and the modularity score (modularity).

Details

Cluster detection is performed on the graph at the time of the call. If you want to cluster a pruned graph, call prune_edges first. This function clusters vertices of the variable-level association graph. Since vertices represent variables rather than factor levels or respondents, the detected communities should be read as blocks of variables that co-vary pairwise. For community detection at the modality level (factor-level co-association), see cluster_modalities.

The Louvain algorithm is non-deterministic; set a seed with set.seed() before calling if reproducibility is needed.

References

Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 2008(10), P10008. doi:10.1088/1742-5468/2008/10/P10008

Pons, P., & Latapy, M. (2005). Computing communities in large networks using random walks. In Computer and Information Sciences – ISCIS 2005 (pp. 284–293). Springer. doi:10.1007/11569596_31

Examples

df <- as.data.frame(Titanic)
df_exp <- df[rep(seq_len(nrow(df)), df$Freq), -5]
cg <- catgraph(df_exp)
set.seed(42)
cg <- detect_clusters(cg)
cg$clustering$n_clusters
#> [1] 1
cg$clustering$modularity
#> [1] 0