label_spreading — label_spreading • Clonotrace

Label Propagation via Iterative Graph-Based Spreading

Usage

label_spreading(
  adj,
  labels,
  label_n = NULL,
  alpha = 0.9,
  max_iter = 100,
  tol = 0.001,
  epsilon = 0,
  verbose = TRUE
)

Arguments

adj: A square adjacency matrix (preferably sparse) representing the graph.
labels: An integer vector of length equal to the number of nodes. Use `NA` for unlabeled entries.
label_n: Optional. The number of label classes. If `NULL`, inferred as `max(labels, na.rm = TRUE)`.
alpha: Float in (0, 1). The propagation coefficient controlling the balance between prior and propagated labels (default: 0.9).
max_iter: Maximum number of iterations for propagation (default: 100).
tol: Convergence threshold (default: 1e-3).
epsilon: Small prior assigned to unlabeled entries (default: 0).
verbose: Logical. Whether to print progress and convergence status (default: TRUE).

Value

A matrix of size N x C where each row contains soft label probabilities for a node across `C` classes.

Details

Performs semi-supervised label propagation over a graph using a sparse or dense adjacency matrix. The algorithm propagates known labels across the graph structure, allowing soft label assignment for unlabeled nodes.

Examples

if (FALSE) {
set.seed(1)
adj <- Matrix::rsparsematrix(100, 100, density = 0.05)
labels <- rep(NA, 100)
labels[1:10] <- sample(1:3, 10, replace = TRUE)
prob_matrix <- label_spreading(adj, labels)
}