Select Optimal Embedding and EGA Model Based on NMI

Select Optimal Embedding and EGA Model Based on NMI

Usage

select_optimal_embedding(
  embedding_matrix,
  sparse_matrix,
  true_communities,
  model = NULL,
  algorithm = "walktrap",
  uni.method = "louvain",
  corr = "auto"
)

Arguments

embedding_matrix

A numeric matrix (columns = items). The full (dense) representation.

sparse_matrix

A numeric matrix (columns = items) giving the sparse representation, aligned to embedding_matrix (same items and column order). This is computed once on the pre-UVA pool and then subset to the post-UVA items in the AI-GENIE pipeline; passing it in (rather than recomputing inside) preserves the pre-UVA quantile thresholds.

true_communities

A named list of known communities.

model

Character. One of "glasso", "TMFG", or NULL (to test both).

algorithm

Community detection algorithm (e.g., "walktrap").

uni.method

Unidimensionality method (e.g., "louvain").

corr

Character. Correlation method. Default "auto" uses EGAnet's automatic detection.

Value

A list with best embedding, model, communities, NMI, and comparison log.

Details

Full embeddings are evaluated before sparse embeddings. Therefore, exact within-model NMI ties retain the full representation. When model = NULL, exact cross-model NMI ties prefer TMFG.