Crate proximal_optimize

A hill-climbing optimizer that works by systematically testing nearby candidates.

This optimizer works even when the objective function (for which a maximum or minimum value is sought) is not differentiable, so that a gradient magnitude cannot be calculated. Any function may be optimized, provided its parameters are (or can be converted from) a &[f64] and its output implments PartialOrd.

Here's an example optimization, using the Rosenbrock function.

use proximal_optimize::ProximalOptimizer;

let mut po = ProximalOptimizer::new(2);
po.iterations(10000);
let initial_position = vec![-1.2, 1.0];
let optimized = po.optimize(&initial_position, |x: &[f64]| {
      ((1.0 - x[0]) * (1.0 - x[0])
    + 100.0 * (x[1] - x[0] * x[0]) * (x[1] - x[0] * x[0]))
}).unwrap();
println!("Optimized values is: {:?}", &optimized);
assert_eq!(optimized, vec![0.999208314861111, 0.998416214890118]);

Structs

ProximalOptimizer

A hill-climbing optimizer that works by systematically testing nearby candidates.

Enums

ProximalOptimizerErr

Constants

DEFAULT_COMPRESSION_RATIO
DEFAULT_EXPANSION_RATIO
DEFAULT_INITIAL_STEP_SIZE
DEFAULT_NUM_ITERATIONS