u-metaheur

Domain-agnostic metaheuristic optimization framework

Overview

u-metaheur provides generic implementations of common metaheuristic algorithms. It contains no domain-specific concepts — scheduling, nesting, routing, etc. are defined by the user through trait implementations.

Algorithms

| Module | Algorithm | Description | |--------|-----------|-------------| | ga | Genetic Algorithm | Population-based evolutionary optimization with pluggable selection, crossover, and mutation operators | | brkga | BRKGA | Biased Random-Key GA — user implements only a decoder; all evolutionary mechanics are handled generically | | sa | Simulated Annealing | Single-solution trajectory optimization with pluggable cooling schedules | | alns | ALNS | Adaptive Large Neighborhood Search — destroy/repair operators with adaptive weight selection | | cp | Constraint Programming | Domain-agnostic modeling layer for constrained optimization with interval, integer, and boolean variables | | dispatching | Dispatching | Generic priority rule composition engine for multi-rule item ranking |

Key Traits

// GA — implement these for your domain
trait Chromosome: Clone + Send + Sync {
    fn fitness(&self) -> f64;
}
trait Crossover<C: Chromosome> {
    fn crossover(&self, parent1: &C, parent2: &C, rng: &mut Rng) -> C;
}
trait Mutation<C: Chromosome> {
    fn mutate(&self, chromosome: &mut C, rng: &mut Rng);
}

// BRKGA — implement only the decoder
trait BrkgaDecoder: Send + Sync {
    type Solution;
    fn decode(&self, keys: &[f64]) -> Self::Solution;
    fn fitness(&self, solution: &Self::Solution) -> f64;
}

// ALNS — implement destroy and repair operators
trait DestroyOperator<S> {
    fn destroy(&self, solution: &S, rng: &mut Rng) -> S;
}
trait RepairOperator<S> {
    fn repair(&self, solution: &S, rng: &mut Rng) -> S;
}

Features

• serde — Enable serde serialization for algorithm parameters

Quick Start

[dependencies]
u-metaheur = { git = "https://github.com/iyulab/u-metaheur" }

# with serde support
u-metaheur = { git = "https://github.com/iyulab/u-metaheur", features = ["serde"] }

Build & Test

cargo build
cargo test
cargo bench  # criterion benchmarks

Dependencies

• u-numflow — Mathematical primitives (statistics, RNG)
• rand 0.9 — Random number generation
• rayon 1.10 — Parallel computation
• serde 1.0 — Serialization (optional)

License

MIT License — see LICENSE.

Related

• u-numflow — Mathematical primitives
• u-geometry — Computational geometry
• u-schedule — Scheduling framework
• u-nesting — 2D/3D nesting and bin packing