Control settings for `global_geo_align.hyperdesign()`

Usage

global_geo_align_control(
  knn = 15L,
  metric = c("euclidean", "cosine"),
  corr_edge_weight = 0,
  n_landmarks_total = 1000L,
  landmarks_per_dataset = NULL,
  landmark_corr_frac = 0.3,
  k_embed = NULL,
  scale_method = c("none", "pairwise_eta", "robust_median"),
  scale_graph_reg = 1e-04,
  scale_anchor_max = 128L,
  max_pairs_per_label = 100L,
  ridge_eps = 1e-04,
  ridge_relative = TRUE,
  ginv_method = c("auto", "woodbury", "direct"),
  eig_tol = 1e-08,
  embed_block_size = 2000L,
  seed = 1L,
  verbose = TRUE
)

Arguments

knn: Number of within-dataset nearest neighbors used to build each domain graph.
metric: Distance metric for within-dataset graph edges (`"euclidean"` or `"cosine"`).
corr_edge_weight: Weight assigned to correspondence edges.
n_landmarks_total: Total number of landmarks across all datasets.
landmarks_per_dataset: Optional fixed landmark count per dataset.
landmark_corr_frac: Fraction of landmarks reserved for correspondence nodes when available.
k_embed: Intermediate geometry rank used in landmark MDS. If `NULL`, defaults to `max(3 * ncomp, ncomp + 5)` inside the solver.
scale_method: Dataset scale harmonization strategy.
scale_graph_reg: Ridge penalty for global log-scale fitting.
scale_anchor_max: Maximum matched anchors per dataset pair used during scale estimation.
max_pairs_per_label: Maximum correspondences sampled per shared label when `y` is used to generate correspondences.
ridge_eps: Ridge regularization base value for linear map estimation.
ridge_relative: Whether to scale ridge by per-dataset feature energy.
ginv_method: Method for solving the regularized least squares problem.
eig_tol: Minimum eigenvalue threshold retained in spectral steps.
embed_block_size: Column block size used for landmark out-of-sample embedding.
seed: Random seed for landmark/correspondence sampling.
verbose: Logical toggle for progress messages.

Value

A list of class `global_geo_align_control`.

Examples

ctrl <- global_geo_align_control(knn = 10, verbose = FALSE)
ctrl$knn
#> [1] 10