Struct IsolationForestDetector 

pub struct IsolationForestDetector { /* private fields */ }

Isolation Forest multivariate anomaly detector.

Recursively partitions data along random features and split values. Anomalous points are isolated in fewer splits (shorter path length). Score is normalized to [0, 1] where values close to 1 indicate anomalies.

Deterministic Seeding

The forest uses a fixed base seed (42) combined with the tree index to derive per-tree PRNG streams. This makes results fully deterministic for a given dataset, which is valuable for reproducibility and testing.

The seed does not incorporate data shape (n_series, n_timestamps) because the subsample selection and split decisions are already conditioned on the actual data values. Adding shape to the seed would change results when the same logical dataset is padded or truncated, without improving detection quality.

If caller-controlled seeding is needed (e.g. for ensemble diversity), a seed parameter can be added to the constructor in a future release.

Implementations

impl IsolationForestDetector

pub fn new( n_trees: Option<usize>, subsample_size: Option<usize>, threshold: Option<f64>, ) -> Self

Creates a new Isolation Forest detector with optional parameters.

pub fn with_seed(self, seed: u64) -> Self

Sets the seed for deterministic PRNG. Default is 42.

Trait Implementations

impl MultivariateAnomalyDetector for IsolationForestDetector

fn fit(&mut self, ctx: &MultiSeriesContext) -> Result<(), MultivariateError>

Fit the detector to the multi-series context.

fn detect( &self, ctx: &MultiSeriesContext, ) -> Result<Vec<MultivariateAnomalyScore>, MultivariateError>

Detect anomalies and return per-timestamp scores.