Struct KnnLabPropGraph 

pub struct KnnLabPropGraph<T> {
    pub offsets: Vec<usize>,
    pub neighbours: Vec<usize>,
    pub weights: Vec<T>,
}

Structure to store KNN graphs and do label propagation

Fields

offsets: Vec<usize>

Stores the offsets for node i’s neighbours

neighbours: Vec<usize>

Flat array with neighbour indices

weights: Vec<T>

Normalised edge weights

Implementations

impl<T> KnnLabPropGraph<T>

where T: BixverseFloat,

pub fn from_edge_list(edges: &[usize], n_nodes: usize, symmetrise: bool) -> Self

Generate the KnnLabelPropGraph

Params

• edges - edge list in form of [node_1, node_2, node_3, …] which indicates alternating pairs (node_1, node_2), etc in terms of edges.
• n_nodes - Number of nodes in the graph
• symmetrise - Shall the graph be made symmetric.

Returns

Self with the data stored in the structure.

pub fn from_node_pairs( from: &[usize], to: &[usize], n_nodes: usize, symmetrise: bool, ) -> Self

Generate KnnLabPropGraph from node pairs

Params

• from - Index of the from node
• to - Index of the to node
• n_nodes - Number of nodes in the graph
• symmetrise - Shall the graph we symmetrised

Returns

Initialised class

pub fn from_weighted_node_pairs( from: &[usize], to: &[usize], weights: &[T], n_nodes: usize, symmetrise: Option<SymmetryWeightStrategy>, ) -> Self

Build the KnnLabPropGraph from node pairs (with weights)

Params

• from - Index of the from node
• to - Index of the to node
• weights - The weights between the nodes
• n_nodes - Number of nodes
• symmetrise - Symmetrisation strategy

Returns

Initialised self

pub fn from_weighted_edge_list( edges: &[usize], weights: &[T], n_nodes: usize, symmetrise: Option<SymmetryWeightStrategy>, ) -> Self

Generate the KnnLabelPropGraph with weights

Params

• edges - edge list in form of [node_1, node_2, node_3, …] which indicates alternating pairs (node_1, node_2), etc in terms of edges.
• weights - the weights between the two nodes in the graph.
• n_nodes - Number of nodes in the graph
• symmetrise - Which symmetry strategy to use. If None, no symmetrisation is performed.

Returns

Self with the data stored in the structure.

pub fn label_spreading( &self, labels: &[Vec<T>], mask: &[bool], alpha: T, iterations: usize, tolerance: T, max_hops: Option<usize>, ) -> Vec<Vec<T>>

Label spreading algorithm

Spreads labels (one-hot encoded categorical data) over the graph. The input needs to be of structure:

class 1 -> [1.0, 0.0, 0.0, 0.0]

class 2 -> [0.0, 1.0, 0.0, 0.0]

unlabelled -> [0.0, 0.0, 0.0, 0.0]

Labelled nodes are anchored to their original labels and act as sources, while unlabelled nodes receive purely propagated distributions.

Params

• labels - One-hot encoded group membership. All zeroes == unlabelled.
• mask - Boolean slice where true indicates an unlabelled node.
• alpha - Anchoring strength for labelled nodes, typically 0.9 to 0.95. Higher values anchor more strongly to the original label; lower values allow more influence from neighbours.
• iterations - Maximum number of spreading iterations.
• tolerance - Convergence threshold. Stops early if the maximum change across all nodes and classes falls below this value.
• max_hops - If Some(k), restricts spreading to nodes within k hops of any labelled node. Nodes beyond this limit are never updated and remain as all-zeroes. If None, spreading is unrestricted.

Returns

A Vec<Vec<T>> of length n_nodes where each entry is the probability distribution over classes for that node.

Trait Implementations

impl<T: Clone> Clone for KnnLabPropGraph<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T: Debug> Debug for KnnLabPropGraph<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.