Struct TimeDecayedMean 

pub struct TimeDecayedMean { /* private fields */ }

An exponentially time-decayed weighted mean.

Each observation (t_i, v_i) contributes v_i · exp(-decay · (t_now - t_i)) to the weighted sum. The mean is Σ(v_i · w_i) / Σ(w_i). Older observations receive exponentially smaller weights, making the statistic responsive to recent changes.

Time complexity per update: O(1). Space complexity: O(1).

Examples

use rill_ml::drift::TimeDecayedMean;

let mut m = TimeDecayedMean::new(0.1).unwrap();
m.update(0.0, 10.0).unwrap();
m.update(1.0, 20.0).unwrap();
m.update(2.0, 30.0).unwrap();
// The mean should be closer to 30 than to the simple average (20)
// because recent observations are weighted higher.
let v = m.value().unwrap();
assert!(v > 20.0, "recent data should dominate: {}", v);

Implementations

impl TimeDecayedMean

pub fn new(decay: f64) -> Result<Self, RillError>

Create a new time-decayed mean with the given decay rate.

decay must be finite and strictly positive. Larger values cause faster forgetting.

pub const fn decay(&self) -> f64

The configured decay rate.

pub fn update(&mut self, time: f64, value: f64) -> Result<(), RillError>

Update with a new observation at time t with value v.

t must be finite and must not decrease (i.e., t >= last_time). v must be finite.

pub fn value(&self) -> Option<f64>

The current decayed mean, or None if no observations have been seen.

pub const fn weight_total(&self) -> f64

The total weight accumulated so far.

pub const fn last_time(&self) -> Option<f64>

The last observation’s timestamp, or None if no observations yet.

pub fn reset(&mut self)

Reset to the initial (no-data) state.

Trait Implementations

impl Clone for TimeDecayedMean

fn clone(&self) -> TimeDecayedMean

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TimeDecayedMean

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

Formats the value using the given formatter.