Struct QuantileRegressorSGBT 

pub struct QuantileRegressorSGBT { /* private fields */ }

Available on crate feature alloc only.

Non-crossing multi-quantile SGBT regressor.

Maintains K independent SGBT<QuantileLoss> models – one per quantile level – and applies the Pool Adjacent Violators Algorithm (PAVA) to predictions to guarantee non-crossing quantile estimates.

Non-crossing guarantee

For quantile levels tau_1 < tau_2 < … < tau_K, the predicted values q_hat(tau_1) <= q_hat(tau_2) <= … <= q_hat(tau_K) after PAVA enforcement. This makes the output suitable for prediction intervals, conditional density estimation, and risk quantification.

Implementations

impl QuantileRegressorSGBT

pub fn new(config: SGBTConfig, quantiles: &[f64]) -> Result<Self>

Create a new multi-quantile regressor.

Quantile levels are automatically sorted. Each level must be in (0, 1) and there must be at least one level.

Errors

Returns IrithyllError::InvalidConfig if:

• quantiles is empty
• any quantile is not in (0, 1)
• duplicate quantile levels exist

pub fn train_one(&mut self, sample: &impl Observation)

Train all quantile models on a single observation.

Each model independently learns from the same sample, targeting its respective quantile level.

pub fn train_batch<O: Observation>(&mut self, samples: &[O])

Train on a batch of observations.

pub fn predict(&self, features: &[f64]) -> Vec<f64>

Predict quantile values with PAVA non-crossing enforcement.

Returns a Vec<f64> of length n_quantiles where:

• result[i] is the predicted quantile at level quantiles[i]
• result[0] <= result[1] <= ... <= result[K-1] (guaranteed)

pub fn predict_raw(&self, features: &[f64]) -> Vec<f64>

Predict raw quantile values WITHOUT non-crossing enforcement.

This may produce crossing predictions. Use predict for the PAVA-enforced version.

pub fn predict_interval(&self, features: &[f64]) -> (f64, f64, f64)

Predict a symmetric prediction interval at coverage level 1 - alpha.

Returns (lower, median, upper) if the model was constructed with quantile levels that include alpha/2, 0.5, and 1 - alpha/2.

If the exact levels are not present, returns the closest available quantile predictions with PAVA enforcement applied.

pub fn predict_batch(&self, feature_matrix: &[Vec<f64>]) -> Vec<Vec<f64>>

Batch prediction with PAVA enforcement.

pub fn n_quantiles(&self) -> usize

Number of quantile levels.

pub fn quantiles(&self) -> &[f64]

The sorted quantile levels.

pub fn n_samples_seen(&self) -> u64

Total samples seen.

pub fn model(&self, idx: usize) -> &SGBT<QuantileLoss>

Access the model for a specific quantile index.

Panics

Panics if idx >= n_quantiles.

pub fn models(&self) -> &[SGBT<QuantileLoss>]

Access all quantile models.

pub fn reset(&mut self)

Reset all quantile models.

Trait Implementations

impl Clone for QuantileRegressorSGBT

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantileRegressorSGBT

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

Formats the value using the given formatter.