Struct lib_dachshund::dachshund::beam::Beam

pub struct Beam<'a, TGraph> where
    TGraph: GraphBase,  {
    pub candidates: Vec<Candidate<'a, TGraph>>,
    pub graph: &'a TGraph,
    // some fields omitted
}

Used for (quasi-clique) detection. A singleton object that keeps state across the beam search. At any point this object considers a "beam" of candidates that is always kept under beam_size, to avoid exponential blowup of the search space.

Fields

candidates: Vec<Candidate<'a, TGraph>>

graph: &'a TGraph

Implementations

impl<'a, TGraph: GraphBase> Beam<'a, TGraph>

pub fn new(
    graph: &'a TGraph,
    clique_rows: &'a Vec<CliqueRow>,
    beam_size: usize,
    verbose: bool,
    non_core_types: &'a [String],
    num_non_core_types: usize,
    alpha: f32,
    global_thresh: Option<f32>,
    local_thresh: Option<f32>,
    graph_id: GraphId
) -> CLQResult<Beam<'a, TGraph>>

creates new beam for mining quasi-bicliques. The following parameters are required:

• graph: a reference to a TGraph object (typically constructed by a transformer`.
• clique_rows: a Vector of CliqueRow entries, which are used to initialize the search process with already-existing cliques.
• beam_size: the number of top candidates to maintain as potential future sources for expansion in the "beam" (i.e., the list of top candidates).
• verbose: used for debugging.
• non_core_types: list of string identifiers for non-core types.
• alpha: Scorer constructor parameter. Controls the contribution of density to the ``cliqueness'' score. Higher values means denser cliques are prefered, all else being equal.
• global_thresh: Scorer constructor parameter. If provided, candidates must be at least this dense to be considered valid (quasi-)cliques.
• local_thresh: Scorer constructor parameter. if provided, each node in the candidate must have at least local_thresh proportion of ties to other nodes in the candidate, for the candidate to be considered valid.
• graph_id: uniquely identifies the graph currently being processed.

pub fn run_search(
    &mut self,
    num_to_search: usize,
    beam_size: usize,
    num_epochs: usize,
    max_repeated_prior_scores: usize
) -> CLQResult<BeamSearchResult<'a, TGraph>>

runs one_step_search for num_epochs epochs, trying num_to_search expansion candidates for each candidate in the beam (the list of top candidates found so far). The beam is of beam_size. If the top score resulting from a one step search is repeated max_repeated_prior_scores times, the search is terminated early. (Note that the search has a stochastic component, which is why repeating the search may yield different results).