Struct VectorAnomalyDetector 

pub struct VectorAnomalyDetector { /* private fields */ }

Detects anomalous vectors relative to a sliding-window reference distribution.

Example

use ipfrs_semantic::anomaly_detector::{
    AnomalyMethod, DetectorConfig, VectorAnomalyDetector,
};

let config = DetectorConfig::with_method(AnomalyMethod::ZScore);
let mut detector = VectorAnomalyDetector::new(config);

// Add reference vectors (all roughly [0.0, 0.0, 0.0])
for _ in 0..12 {
    detector.add_reference(vec![0.0_f32, 0.0, 0.0]);
}

// A normal vector should not be flagged
let normal = detector.detect(1, &[0.1, -0.1, 0.05]);
assert!(normal.is_some());

// A large outlier should be flagged
let outlier = detector.detect(2, &[100.0, 100.0, 100.0]);
assert!(outlier.map_or(false, |r| r.is_anomaly));

Implementations

impl VectorAnomalyDetector

pub fn new(config: DetectorConfig) -> Self

Create a new detector with the given configuration.

pub fn add_reference(&mut self, vec: Vec<f32>)

Add a reference vector.

If the reference set has reached max_reference, the oldest vector is evicted before the new one is appended.

pub fn compute_mean_std(&self) -> (Vec<f32>, Vec<f32>)

Compute per-dimension mean and standard deviation over the reference set.

Returns (means, stds) where each slice has length equal to the dimensionality of the reference vectors. The std is clamped to a minimum of 1e-6 to avoid division by zero.

Panics if the reference set is empty.

pub fn detect(&mut self, vector_id: u64, vec: &[f32]) -> Option<AnomalyResult>

Check a single vector against the reference distribution.

Returns None if the reference set is smaller than min_samples.

pub fn stats(&self) -> &DetectorStats

Access running statistics.