Struct QuantileForecaster 

pub struct QuantileForecaster { /* private fields */ }

Wraps a ForecastModel with empirical, horizon-aware quantile intervals.

Example

use chronix_analytics::forecast::{
    HoltWintersModel, QuantileConfig, QuantileForecaster,
};

let config = QuantileConfig::central(0.9, 24);

let mut qf = QuantileForecaster::new(
    || Box::new(HoltWintersModel::new(Some(0.3), Some(0.1), Some(0.1), Some(24), false)),
    config,
);
qf.fit(&timestamps, &values)?;

let forecast = qf.predict(24)?;
let p05 = forecast.level(0.05).expect("requested level");

Implementations

impl QuantileForecaster

pub fn new<F>(factory: F, config: QuantileConfig) -> Self

where F: Fn() -> Box<dyn ForecastModel> + Send + Sync + 'static,

Build a forecaster from a model factory.

factory must return a fresh, unfitted model on every call.

pub fn fit( &mut self, timestamps: &[i64], values: &[f64], ) -> Result<(), ForecastError>

Fit the underlying model and calibrate the residual distribution.

Errors

Returns ForecastError::InvalidInput if the configuration is invalid or the series is too short to produce a single rolling origin, or propagates the underlying model’s fit errors.

pub fn predict(&self, horizon: usize) -> Result<QuantileForecast, ForecastError>

Produce point forecasts plus calibrated quantiles for horizon steps.

Horizons beyond the calibrated range reuse the last calibrated step’s residual distribution, which is the most conservative available estimate rather than a silent extrapolation.

Errors

Returns ForecastError::NotFitted if fit has not been called, or propagates the underlying model’s prediction errors.

pub fn predict_interval( &self, horizon: usize, level: f64, ) -> Result<ForecastResult, ForecastError>

Rewrite a ForecastResult’s bounds using the calibrated empirical quantiles, keeping the existing API shape.

Errors

As predict.

pub fn calibration_size(&self, h: usize) -> usize

Number of calibration residuals collected for horizon step h (1-based). Returns 0 for uncalibrated horizons.

pub fn model(&self) -> &dyn ForecastModel

The wrapped model.