Crate turboswarm_core 

turboswarm-core

Particle Swarm Optimization (PSO) core: extensible and modular. Pure Rust computation, with no FFI dependencies; the Python bindings live in the separate pso-py crate.

Core idea

The PSO loop (Pso) knows nothing about any concrete variant. What changes between variants lives behind three traits (SearchSpace, Velocity, Topology). Adding a variant = implementing a trait, without touching the core. Internally, all positions and velocities are Vec<f64>; the integer/real difference lives solely in SearchSpace::decode.

What it includes

• Spaces (spaces): ContinuousSpace (real) and IntegerSpace (integer, with configurable Discretization).
• Velocity variants (velocity): InertiaVelocity (Shi-Eberhart), ConstrictionVelocity (Clerc-Kennedy) and FipsVelocity (Fully Informed PSO).
• Topologies (topology): GlobalBest, Ring (lbest ring), VonNeumann (2D grid) and Random.
• Benchmarks (benchmarks): sphere, rastrigin, rosenbrock, ackley, griewank and schwefel, with their metadata (benchmarks::Benchmark).
• History (History): a complete trace of the swarm for visualization.

Minimal example

use turboswarm_core::prelude::*;

let space = ContinuousSpace::uniform(2, -5.12, 5.12);
let velocity = InertiaVelocity::new(0.729, 1.49445, 1.49445);
let params = PsoParams { seed: Some(42), ..Default::default() };

let pso = Pso::new(space, velocity, GlobalBest::new(), params);
let result = pso.minimize(|x| turboswarm_core::benchmarks::sphere(x));

assert!(result.best_value < 1e-3);

How to extend

For a new variant, implement Velocity (template: velocity/inertia.rs); for a new topology, implement Topology by defining Topology::neighbors. See the README.md for the full guide.

Modules

benchmarks

Standard test functions for optimization, with their known global optimum. They allow validating the optimizer and measuring the real error.

history

Record of the optimization process, for visualization and analysis.

mopso

Multi-objective PSO (MOPSO, Coello Coello & Lechuga, 2004).

params

Optimizer configuration parameters.

prelude

Convenient re-exports for common usage.

pso

The PSO optimizer. The loop knows nothing about any concrete variant: it receives a SearchSpace, a Velocity and a Topology via generics.

spaces

Search spaces: they define the domain and the integer/real distinction.

swarm

Swarm data structures.

topology

Social topologies: they define the neighborhood of each particle.

traits

Core traits of the framework.

velocity

Velocity update rules. Each one is a PSO variant.