Module warm_start 

Warm-starting and transfer learning for Bayesian Optimization.

Provides mechanisms to seed a new Bayesian optimization run with knowledge from previous runs, related tasks, or meta-learned initialization strategies.

Strategies

1. Direct warm-start: inject prior observations directly into the GP.
2. Scaled transfer: align source observations to the target domain via min-max rescaling and inject them with a down-weighted noise level.
3. Multi-task BO: maintain separate GPs per task and combine acquisition values using task-similarity weights.
4. Meta-learning: estimate good GP hyperparameter initialization from observed task features (warm-starting the surrogate model itself).

Example

use scirs2_optimize::bayesian::warm_start::{
    WarmStartBo, WarmStartConfig, PriorRun, MetaLearner,
};
use scirs2_core::ndarray::{Array1, Array2};

// A previous run on a related problem:
let prior = PriorRun {
    x: Array2::from_shape_vec((4, 1), vec![0.0, 1.0, 2.0, 3.0]).expect("shape"),
    y: Array1::from_vec(vec![1.0, 0.5, 0.1, 0.6]),
    bounds: vec![(0.0_f64, 4.0_f64)],
    weight: 0.8,
};

let config = WarmStartConfig {
    prior_runs: vec![prior],
    n_initial: 3,
    seed: Some(42),
    ..Default::default()
};

let mut bo = WarmStartBo::new(vec![(0.0_f64, 4.0_f64)], config).expect("create");
let result = bo.optimize(|x: &[f64]| (x[0] - 1.5_f64).powi(2), 10).expect("opt");
println!("Best x: {:?}  f: {:.4}", result.x_best, result.f_best);

Structs

MetaLearner

Meta-learner that estimates good initial GP hyperparameters from prior tasks.

MultiTaskBo

Multi-task Bayesian optimizer.

MultiTaskBoConfig

Configuration for multi-task Bayesian optimization.

PriorRun

A record of a previous optimization run that can be used to warm-start a new run.

Task

A task descriptor for multi-task BO.

WarmStartBo

Bayesian optimizer with warm-starting from prior runs.

WarmStartConfig

Configuration for warm-start Bayesian optimization.

WarmStartObs

A single observation recorded during optimization.

WarmStartResult

Result of a warm-start Bayesian optimization run.

Enums

BlendStrategy

Strategy for blending prior observations into the current run.

Functions

warm_start_optimize

Run Bayesian optimization with warm-starting from prior runs.