Crate optimal

Mathematical optimization and machine learning framework and algorithms.

Optimal provides a composable framework for mathematical optimization and machine learning from the optimization perspective, in addition to algorithm implementations.

The framework consists of runners, optimizers, and problems, with a chain of dependency as follows: runner -> optimizer -> problem. Most optimizers can support many problems and most runners can support many optimizers.

A problem defines a mathematical optimization problem. An optimizer defines the steps for solving a problem, usually as an infinite series of state transitions incrementally improving a solution. A runner defines the stopping criteria for an optimizer and may affect the optimization sequence in other ways.

Examples

Minimize the “count” problem using a derivative-free optimizer:

use optimal::{prelude::*, BinaryDerivativeFreeConfig};

println!(
    "{:?}",
    BinaryDerivativeFreeConfig::start_default_for(16, |point| {
        point.iter().filter(|x| **x).count() as f64
    })
    .argmin()
);

Minimize the “sphere” problem using a derivative optimizer:

use optimal::{prelude::*, RealDerivativeConfig};

println!(
    "{:?}",
    RealDerivativeConfig::start_default_for(
        2,
        std::iter::repeat(-10.0..=10.0).take(2),
        |point| point.iter().map(|x| x.powi(2)).sum(),
        |point| point.iter().map(|x| 2.0 * x).collect(),
    )
    .nth(100)
    .unwrap()
    .best_point()
);

For more control over configuration parameters, introspection of the optimization process, serialization, and specialization that may improve performance, see individual optimizer packages.

Modules

• prelude
  Useful traits, types, and functions unlikely to conflict with existing definitions.

Structs

• BinaryDerivativeFree
  A generic binary derivative-free optimizer.
• BinaryDerivativeFreeConfig
  Binary derivative-free optimizer configuration parameters.
• RealDerivative
  A generic binary derivative-free optimizer.
• RealDerivativeConfig
  Binary derivative-free optimizer configuration parameters.

Traits

• Optimizer
  Running optimizer methods independent of configuration and state.
• OptimizerExt
  An extension trait adding methods to StreamingIterator for optimization.
• StreamingIterator
  An interface for dealing with streaming iterators.
• StreamingIteratorExt
  An extension trait adding methods to StreamingIterator.