Skip to contents

kNN label propagation

rs_knn_label_propagation.Rd

The function is a helper function to do kNN label propagation. This can be useful for semi-supervised tasks. It implements the label spreading method.

Usage

rs_knn_label_propagation(
  from,
  to,
  one_hot_encoding,
  label_mask,
  weights,
  label_prop_params
)

Arguments

from

Integer vector. Source node indices for each edge.

to

Integer vector. Target node indices for each edge. Must be the same length as from.

one_hot_encoding

Integer matrix. Each row represents a sample, the columns the one-hot encodings. Everything 0 denotes the unlabelled data.

label_mask

Boolean vector. Which of the samples do not have a label. Needs to be same length as nrow(one_hot_encoding).

weights

Optional numeric vector. Edge weights for each pair in from/to. Must have the same length as from. If NULL, all edges are treated as unweighted.

label_prop_params

List. Named list of parameters with the following optional fields (defaults in parentheses):

  • alpha numeric, spreading strength (0.9)

  • iter integer, max iterations (100)

  • tolerance numeric, convergence threshold (1e-6)

  • symmetrise logical, symmetrise the graph (FALSE)

  • symmetry_strategy character, one of "average", "min", "max" ("average")

  • max_hops integer, restrict spreading radius (unrestricted)

Value

The matrix with the probabilities of being of a certain class.