Package index • lcdaGRASP

Algorithms

The LCDA variants proposed in the paper, plus the ensemble-consensus variant.

lcda_grasp(): LCDA-GRASP: fixed-parameter variant.
lcda_gr(): LCDA-GR: Reactive variant with self-tuning of (alpha_c, alpha_s). Implements Algorithm 4 of the paper.
lcda_ecg(): LCDA-ECG: ensemble-consensus community and leader detection.

Building blocks of a single GRASP iteration (construct, repair, local search).

lcda_construct(): LCDA construction - variant 1 (centrality computed once) or 2 (adaptive).
lcda_repair(): Repair: ensure every community has exactly one leader by recomputing centrality within the community and designating the top-scoring node.
lcda_local_search(): First-improvement local search, with automatic VNMI dispatch for n > 300.

Everything the paper’s tables report, computed from a fitted result: one tidy metric table, plus the per-community and per-leader breakdowns.

Modularity, the NCE leader score (global and community-conditioned), and the lexicographic objective.

Leader-selection centralities.

centrality_eigen(): Eigenvector centrality (own implementation, O(m log n)).
centrality_betweenness(): Betweenness centrality - delegates to igraph::betweenness for now. Tagged for a future native Rcpp implementation (Brandes 2001).
centrality_closeness(): Closeness centrality - delegates to igraph::closeness. On a disconnected graph, plain closeness is ill-defined across components (nodes in tiny components score spuriously high, distorting leader selection); we fall back to harmonic centrality, the robust generalisation that handles unreachable pairs (Boldi & Vigna 2014). Note the two are NOT identical even on a connected graph – closeness inverts the mean distance, harmonic averages the inverse distances – so on disconnected inputs (e.g. PolBlogs, and the frequently-disconnected induced subgraphs of variant 2) the selected leader may differ from a per-component closeness; this is a deliberate robustness choice, recorded here so benchmark results are interpreted accordingly.

The non-parametric procedure used in the paper’s experiments.

kruskal_then_permute(): Two-stage non-parametric comparison: Kruskal-Wallis omnibus followed by pairwise permutation tests with Bonferroni correction.
permutation_test(): Two-sample permutation test for a difference in means. Robust to ties and to zero-variance pools (as discussed in section 5.1).

The community-and-leader map, plus the search-diagnostic figures.

lcda_plot_communities(): Detect communities and leaders on a graph, then plot them
plot_partition(): Plot a community partition with its leaders highlighted
plot(<lcda_grasp_result>) plot(<lcda_gr_result>) plot(<lcda_ecg_result>): Plot a fitted LCDA result as a community-and-leader map
autoplot(<lcda_grasp_result>) autoplot(<lcda_gr_result>) autoplot(<lcda_ecg_result>): Community-and-leader map as a ggplot object
plot_grasp_trajectory(): Plot the Q trajectory across GRASP iterations, with running maximum.
plot_reactive_pk(): Plot the evolution of selection probabilities p_k across iterations.

as_csr(): Convert an igraph graph to the CSR representation used by the kernels
as_graph(): Coerce a variety of inputs to igraph.
graph_from_edgelist_simple(): Build an igraph from an integer edgelist (1-based), undirected simple.
lcda_data(): Precomputed simulation datasets
lcda_provenance(): Provenance of a shipped dataset (versions, date, checksum)