Struct DBStream 

pub struct DBStream { /* private fields */ }

Density-based streaming clustering.

Maintains a set of weighted micro-clusters and a shared-density graph. Points that fall within radius of an existing micro-cluster are merged into it; otherwise a new micro-cluster is created. Shared density between pairs of micro-clusters accumulates when a single point is within range of both. Macro-clusters are formed by finding connected components in the shared-density graph.

Example

use irithyll::clustering::dbstream::{DBStreamConfig, DBStream};

let config = DBStreamConfig::builder(1.0).build().unwrap();
let mut db = DBStream::new(config);

db.train_one(&[0.0, 0.0]);
db.train_one(&[0.1, 0.1]);
db.train_one(&[10.0, 10.0]);

assert_eq!(db.n_micro_clusters(), 2);

Implementations

impl DBStream

pub fn new(config: DBStreamConfig) -> Self

Create a new DBStream instance from the given configuration.

pub fn train_one(&mut self, features: &[f64])

Process a single sample, updating micro-clusters and shared density.

This is the core DBSTREAM algorithm:

1. Find all micro-clusters within radius of the point.
2. If at least one is in range, merge into the nearest and update shared density for all in-range pairs.
3. Otherwise, create a new micro-cluster at the point’s location.
4. Apply exponential decay to all weights and shared densities.
5. Periodically clean up micro-clusters below min_weight.

pub fn predict(&self, features: &[f64]) -> usize

Assign the given point to the nearest micro-cluster.

Returns the index into micro_clusters().

Panics

Panics if there are no micro-clusters.

pub fn predict_or_noise( &self, features: &[f64], noise_radius: f64, ) -> Option<usize>

Assign the given point to the nearest micro-cluster, or return None if the point is further than noise_radius from all micro-clusters.

pub fn micro_clusters(&self) -> &[MicroCluster]

Return a reference to the current micro-clusters.

pub fn n_micro_clusters(&self) -> usize

Number of active micro-clusters.

pub fn macro_clusters(&self) -> Vec<Vec<usize>>

Group micro-clusters into macro-clusters via connected components in the shared-density graph.

Two micro-clusters i and j are connected if their shared density exceeds shared_density_threshold * (weight_i + weight_j). Returns a Vec of groups, where each group is a Vec<usize> of micro-cluster indices.

pub fn n_clusters(&self) -> usize

Number of macro-clusters (connected components in the shared-density graph).

pub fn n_samples_seen(&self) -> u64

Total number of samples processed so far.

pub fn reset(&mut self)

Reset all state, removing all micro-clusters and shared density.

Trait Implementations

impl Clone for DBStream

fn clone(&self) -> DBStream

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DBStream

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

Formats the value using the given formatter.