Module constrained_bo 

Constrained Bayesian Optimization.

Extends the standard Bayesian optimization framework to handle black-box constraints that are expensive to evaluate (e.g. safety constraints, physical feasibility conditions, computational limits).

Constraint Handling Strategies

Method	Description
Expected Feasible Improvement (EFI)	EI weighted by probability of satisfying all constraints
Probability of Feasibility (PoF)	Pure probability that a point is feasible
Constrained EI (cEI)	EI penalised multiplicatively by PoF for each constraint
Augmented Lagrangian BO	Adds constraint violation penalties to the GP posterior

Example

use scirs2_optimize::bayesian::constrained_bo::{
    ConstrainedBo, ConstrainedBoConfig, BlackBoxConstraint,
};
use scirs2_core::ndarray::ArrayView1;

// Minimize x[0]^2 + x[1]^2 subject to x[0] + x[1] >= 1.
let obj = |x: &[f64]| x[0].powi(2) + x[1].powi(2);

// Constraint g(x) <= 0 means "feasible". So g(x) = 1 - x[0] - x[1].
let con = BlackBoxConstraint {
    name: "sum_geq_1".into(),
    // returns positive => infeasible, <= 0 => feasible.
    evaluate: Box::new(|x: &[f64]| 1.0 - x[0] - x[1]),
};

let config = ConstrainedBoConfig {
    n_initial: 5,
    seed: Some(42),
    ..Default::default()
};

let mut cbo = ConstrainedBo::new(
    vec![(-2.0_f64, 2.0_f64), (-2.0_f64, 2.0_f64)],
    vec![con],
    config,
).expect("create");

let result = cbo.optimize(obj, 20).expect("opt");
println!("Best feasible x: {:?}  f: {:.4}", result.x_best, result.f_best);

Structs

BlackBoxConstraint

A black-box constraint function.

ConstrainedBo

Bayesian optimizer with support for expensive black-box constraints.

ConstrainedBoConfig

Configuration for constrained Bayesian optimization.

ConstrainedBoResult

Result of constrained Bayesian optimization.

ConstrainedObservation

A single evaluated point in constrained BO.

ExpectedFeasibleImprovement

Expected Feasible Improvement: EI weighted by joint probability of feasibility.

ProbabilityOfFeasibility

Probability of Feasibility for a single constraint GP.

Enums

ConstrainedAcquisitionStrategy

Strategy for handling constraints in the acquisition function.

Functions

constrained_optimize

Run constrained Bayesian optimization with the default EFI strategy.