Struct MLPLeafModel 

pub struct MLPLeafModel { /* private fields */ }

Available on crate feature alloc only.

Single hidden layer MLP leaf model with ReLU activation.

Learns a nonlinear function via backpropagation with Newton-scaled learning rate. Weights are lazily initialized on the first update call using a deterministic xorshift64 PRNG so results are reproducible.

Optional exponential weight decay (decay) gives the model a finite memory horizon for non-stationary streams.

Implementations

impl MLPLeafModel

pub fn new( hidden_size: usize, learning_rate: f64, seed: u64, decay: Option<f64>, ) -> Self

Create a new MLP leaf model with the given hidden layer size, learning rate, seed, and optional decay.

The seed controls deterministic weight initialization. Different seeds produce different initial weights, which is critical for ensemble diversity when multiple MLP leaves share the same hidden_size.

Trait Implementations

impl LeafModel for MLPLeafModel

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

Produce a prediction given input features.

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

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

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

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

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

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