Struct AdwinTest 

pub struct AdwinTest<A: Float + Send + Sync> { /* private fields */ }

ADWIN (ADaptive WINdowing).

Maintains a single window of recent observations and, on every update, searches every valid split point for a pair of sub-windows whose means differ by more than the Hoeffding cut eps_cut = R * sqrt(ln(4/delta) / (2m)), where m is the harmonic combination 1/(1/n0 + 1/n1) of the sub-window sizes and R is the observed range of the window (the practical adaptation used for unbounded metrics — the textbook bound assumes observations in [0, 1]). When a cut is found the older sub-window is dropped, which is ADWIN’s defining behaviour: the window shrinks to the most recent stationary segment.

ADWIN is a single-stream detector, so it consumes only the newly arrived observations; the caller’s reference sample is not used (that is what makes it genuinely different from the two-sample tests in this module).

Implementations

impl<A: Float + Send + Sync> AdwinTest<A>

pub fn new(sensitivity: f64, significance_level: f64) -> Result<Self, String>

Creates an ADWIN detector. sensitivity is interpreted as ADWIN’s delta confidence parameter (smaller = more conservative).

Trait Implementations

impl<A: Float + Default + Clone + Send + Sync + Sum> StatisticalTest<A> for AdwinTest<A>

fn test_for_drift( &mut self, _reference: &[A], current: &[A], ) -> Result<DriftTestResult<A>, String>

Performs the statistical test for drift

fn update_parameters(&mut self, performance_feedback: A) -> Result<(), String>

Updates test parameters based on historical performance

fn reset(&mut self)

Resets the test state