Struct SemanticAnomalyDetector 

pub struct SemanticAnomalyDetector { /* private fields */ }

Detects anomalous embeddings using distance-from-centroid analysis.

Supports three detection methods:

• ZScore: z-score of distances to centroid
• IQR: interquartile range on distances
• DistanceBased: mean + multiplier * std_dev threshold on distances

Example

use ipfrs_semantic::anomaly_detector::{
    SemanticAnomalyDetector, AnomalyConfig, SemanticAnomalyMethod,
};

let config = AnomalyConfig::default();
let mut detector = SemanticAnomalyDetector::new(config);

// Add some normal embeddings
for i in 0..20 {
    detector.add_embedding(&format!("doc_{i}"), vec![0.1 * i as f64, 0.0, 0.0]);
}

// Detect anomalies
let results = detector.detect_all();

Implementations

impl SemanticAnomalyDetector

pub fn new(config: AnomalyConfig) -> Self

Create a new detector with the given configuration.

pub fn add_embedding(&mut self, doc_id: &str, embedding: Vec<f64>)

Add an embedding and incrementally recompute the centroid.

pub fn remove_embedding(&mut self, doc_id: &str) -> bool

Remove an embedding by doc_id. Returns true if found and removed.

Recomputes the centroid from scratch after removal.

pub fn detect_all(&mut self) -> Vec<SemanticAnomalyResult>

Run detection on all stored embeddings using the configured method.

pub fn detect_single(&self, embedding: &[f64]) -> SemanticAnomalyResult

Check if a single new embedding is anomalous against existing data.

Does not add the embedding to the detector.

pub fn compute_centroid(embeddings: &[(String, Vec<f64>)]) -> Vec<f64>

Compute the centroid (mean vector) of the given embeddings.

pub fn euclidean_distance(a: &[f64], b: &[f64]) -> f64

Euclidean distance between two vectors.

pub fn distances_to_centroid(&self) -> Vec<(String, f64)>

Compute (doc_id, distance) pairs for all stored embeddings to centroid.

pub fn embedding_count(&self) -> usize

Number of stored embeddings.

pub fn stats(&self) -> AnomalyDetectorStats

Get detector statistics.