Fit GRASP on a pair of domains — fit_pair.grasp

GRASP-specific method for pairwise alignment using spectral bases and descriptor alignment with an orthogonal rotation matrix.

Usage

# S3 method for class 'grasp_aligner'
fit_pair(
  algo,
  X_i,
  X_j,
  links = NULL,
  ncomp = 30,
  q_descriptors = 100,
  sigma = 0.73,
  lambda = 0.1,
  use_laplacian = TRUE,
  solver = c("linear", "hungarian", "auction"),
  ...
)

Arguments

algo: A grasp_aligner object
X_i: First domain data matrix (samples x features)
X_j: Second domain data matrix (samples x features)
links: Optional correspondence links (unused by GRASP)
ncomp: Number of spectral components to compute
q_descriptors: Number of descriptor vectors for alignment
sigma: Bandwidth parameter for descriptor weighting
lambda: Regularization parameter for rotation alignment
use_laplacian: Logical; if TRUE use graph Laplacian, else adjacency
solver: Assignment solver: "linear", "hungarian", or "auction"
...: Additional arguments (unused)

Value

An object of class grasp_pair_fit containing rotation (kxk matrix), mapping_matrix (assignment matrix), assignment vector, loss (numeric), and k (latent dimension).

Examples

# \donttest{
set.seed(123)
X1 <- matrix(rnorm(100), 50, 2)
X2 <- matrix(rnorm(100), 50, 2)
algo <- grasp_aligner()
fit <- fit_pair(algo, X1, X2, ncomp = 5, q_descriptors = 10)
class(fit)
#> [1] "grasp_pair_fit"
# }