Skip to contents

The package ships the results of its publication-grade simulation panel as compressed `.rds` files under `inst/extdata/`. The vignettes read these so they never re-run a simulation at build time. Regenerate them with the scripts in `data-raw/` (see `data-raw/run_all_data.R`).

Usage

lcda_data(name = NULL)

Arguments

name

dataset name (without the `.rds` extension). If `NULL` (default), returns a tibble listing the available datasets.

Value

If `name` is `NULL`, a tibble of available datasets. Otherwise the `list(results, meta)` bundle for that dataset, or `NULL` with a message if it has not been generated yet.

Details

Available datasets (name -> contents):

`repro_benchmarks`, `repro_summary`

Benchmark reproduction on the five networks vs Louvain/Leiden and literature best-known Q.

`doe_screening`, `doe_rsm`

Design-of-experiments runs: factorial screening and a central composite design in `(alpha_c, alpha_s)`.

`lfr_robustness`

Q and ARI across the LFR-like mixing sweep.

`prop6_summary`, `prop6_trajectories`

Reactive-update concentration.

`pool_sensitivity`

`(m, y, B)` grid for LCDA-GR.

`vnmi_nprime`

VNMI candidate-subset size `n'` sweep: modularity and local-search time vs `n'` across density regimes.

`degeneracy`

Near-best partition counts and pairwise NMI.

`eda_replicates`

Replicated `(Q, H)` for exploratory analysis.

`nce_alternatives`

Global vs community-conditioned NCE leaders.

`lcda_ecg`, `overlap_lcda_ecg`, `stability_pool`, `consensus_leader_test`

LCDA-ECG ensemble consensus: LFR recovery, overlapping/bridge nodes, pool-stability stopping rule, and the consensus-vs-central leader comparison.

`largescale`

Recovery and runtime up to n = 5e4 (synthetic LFR).

`realnet_amazon`, `realnet_coauthor`

Large real networks with ground truth: modularity Q and recovery (NMI/ARI vs labels) on Amazon-Computers (co-purchase) and Coauthor-Physics (co-authorship).

`openalex_leaders`

Leader validation against an external citation signal on an OpenAlex co-authorship graph.

`gnn_baseline`

Graph-auto-encoder baseline (true-k and auto-k) vs classical methods on LFR.

`weighted_demo`

Weighted-graph demonstration (weights aid recovery).

`blogs_table9`, `blogs_timing`

Weighted Political Blogs reproduction (mean/max Q, timings).

`doe_lfr_recovery`

DoE screening/RSM scored by LFR recovery.

`leader_utility`, `leader_vs_twostage`, `leader_vs_twostage_lfr`

Downstream leader utility (IC/LT spread, coverage) vs two-stage detect-then-centrality pipelines.

Examples

lcda_data()                       # list what is available
#> # A tibble: 28 × 1
#>    dataset              
#>    <chr>                
#>  1 blogs_table9         
#>  2 blogs_timing         
#>  3 consensus_leader_test
#>  4 degeneracy           
#>  5 doe_lfr_recovery     
#>  6 doe_rsm              
#>  7 doe_screening        
#>  8 eda_replicates       
#>  9 gnn_baseline         
#> 10 largescale           
#> # ℹ 18 more rows
if (FALSE) { # \dontrun{
d <- lcda_data("repro_summary")
d$meta$limitations
head(d$results)
} # }