Struct optimization::NumericalDifferentiation

pub struct NumericalDifferentiation<F: Function> { /* fields omitted */ }

Wraps a function for which to provide numeric differentiation.

Uses simple one step forward finite difference with step width h = √εx.

Examples

let square = NumericalDifferentiation::new(Func(|x: &[f64]| {
    x[0] * x[0]
}));

assert!(square.gradient(&[0.0])[0] < 1.0e-3);
assert!(square.gradient(&[1.0])[0] > 1.0);
assert!(square.gradient(&[-1.0])[0] < 1.0);

Methods

impl<F: Function> NumericalDifferentiation<F>

pub fn new(function: F) -> Self

Creates a new differentiable function by using the supplied function in combination with numeric differentiation to find the derivatives.

Trait Implementations

impl<F: Function + Default> Default for NumericalDifferentiation<F>

fn default() -> Self

Returns the "default value" for a type.

impl<F: Function> Function for NumericalDifferentiation<F>

fn value(&self, position: &[f64]) -> f64

Computes the objective function at a given position x, i.e., f(x) = y.

impl<F: Function> Function1 for NumericalDifferentiation<F>

fn gradient(&self, position: &[f64]) -> Vec<f64>

Computes the gradient of the objective function at a given position x, i.e., ∀ᵢ ∂/∂xᵢ f(x) = ∇f(x).

impl<F: Problem> Problem for NumericalDifferentiation<F>

fn dimensions(&self) -> usize

Returns the dimensionality of the input domain.

fn domain(&self) -> Vec<(f64, f64)>

Returns the input domain of the function in terms of upper and lower, respectively, for each input dimension.

fn minimum(&self) -> (Vec<f64>, f64)

Returns the position as well as the value of the global minimum.

fn random_start(&self) -> Vec<f64>

Generates a random and feasible position to start a minimization.

fn is_legal_position(&self, position: &[f64]) -> bool

Tests whether the supplied position is legal for this function.