Struct GeneralGraphCrf 

pub struct GeneralGraphCrf {
    pub node_potentials: Vec<f64>,
    pub edge_potentials: Vec<f64>,
    pub edges: Vec<Edge>,
    /* private fields */
}

General-graph pairwise CRF with loopy belief propagation inference.

Fields

node_potentials: Vec<f64>

Node unary log-potentials [n_nodes × n_labels].

edge_potentials: Vec<f64>

Edge pairwise log-potentials [n_edges × n_labels × n_labels].

edges: Vec<Edge>

Edge list.

Implementations

impl GeneralGraphCrf

pub fn new(config: GraphCrfConfig, edges: Vec<Edge>) -> SeqResult<Self>

Create a new CRF with zero potentials and the given edges.

Validates that node indices in edges are within [0, n_nodes).

pub fn set_node_potential( &mut self, node: usize, lbl: usize, val: f64, ) -> SeqResult<()>

Set the unary log-potential for node node, label lbl.

pub fn set_edge_potential( &mut self, e_idx: usize, li: usize, lj: usize, val: f64, ) -> SeqResult<()>

Set the pairwise log-potential for edge e_idx, labels (li, lj).

pub fn sum_product_marginals(&self) -> SeqResult<Vec<f64>>

Run sum-product loopy belief propagation.

Returns marginal log-beliefs [n_nodes × n_labels] (normalised so that Σ_label exp(beliefs[node * n_labels + label]) = 1).

pub fn map_decode(&self) -> SeqResult<Vec<usize>>

Run max-product loopy BP to compute approximate MAP assignments.

Returns [n_nodes] with the argmax label for each node.

pub fn n_nodes(&self) -> usize

Number of nodes.

pub fn n_edges(&self) -> usize

Number of edges.

pub fn n_labels(&self) -> usize

Number of labels.

Trait Implementations

impl Clone for GeneralGraphCrf

fn clone(&self) -> GeneralGraphCrf

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GeneralGraphCrf

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

Formats the value using the given formatter.