Skip to main content

LeafModel

irithyll_core::tree::leaf_model

Trait LeafModel 

Source 
pub trait LeafModel: Send + Sync {
    // Required methods
    fn predict(&self, features: &[f64]) -> f64;
    fn update(
        &mut self,
        features: &[f64],
        gradient: f64,
        hessian: f64,
        lambda: f64,
    );
    fn clone_fresh(&self) -> Box<dyn LeafModel>;

    // Provided method
    fn clone_warm(&self) -> Box<dyn LeafModel> { ... }
}
Available on crate feature alloc only.
Expand description

A trainable prediction model that lives inside a decision tree leaf.

Implementations must be Send + Sync so trees can be shared across threads.

Required Methods§

Source

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

Produce a prediction given input features.

Source

fn update(&mut self, features: &[f64], gradient: f64, hessian: f64, lambda: f64)

Update model parameters given a gradient, hessian, and regularization lambda.

Source

fn clone_fresh(&self) -> Box<dyn LeafModel>

Create a fresh (zeroed / re-initialized) clone of this model’s architecture.

Provided Methods§

Source

fn clone_warm(&self) -> Box<dyn LeafModel>

Create a warm clone preserving learned weights but resetting optimizer state.

Used when splitting a leaf: child leaves inherit the parent’s learned function as a starting point, converging faster than starting from scratch. Defaults to clone_fresh for models where warm-starting is not meaningful (e.g. ClosedFormLeaf).

Implementors§