Enum Problem 

pub enum Problem {
    Cartpole,
    Rastrigin,
    Ackley,
    Rosenbrock,
    Himmelblau,
    ShiftedQuadratic,
    Custom(Arc<Landscape>),
}

A minimization problem.

Variants

Cartpole

CartPole-v1 balanced by a linear policy that Augmented Random Search trains (D113): x = log10 step size and y = log10 exploration noise on [-2.5, 0]², and the value is the run’s mean shortfall from 500 steps over its crate::training::STEPS epochs (the area above its learning curve). 0 is a lower bound; no location of it is claimed.

Rastrigin

The 2-D Rastrigin function on [-5.12, 5.12]²; global minimum 0 at the origin, surrounded by a regular grid of local minima.

Ackley

The 2-D Ackley function on [-5, 5]²; global minimum 0 at the origin, in a narrow funnel on a nearly flat, rippled plateau.

Rosenbrock

The Rosenbrock function on [-2, 2] × [-1, 3]; global minimum 0 at (1, 1), at the end of a long, flat, curved valley.

Himmelblau

Himmelblau’s function on [-5, 5]²; four global minima of value 0.

ShiftedQuadratic

(x - 1.25)² on [-5, 5]; minimum 0 at x = 1.25. A one-dimensional regression fixture, not shown in the browser.

Custom(Arc<Landscape>)

A user formula on [-5, 5]² (D94).

Implementations

impl Problem

pub const SHOWN: [Self; 4]

The test functions the browser offers, in display order, after Cart-pole.

pub const BUILT_IN: [Self; 5]

Every built-in problem the browser offers, in display order: cart-pole (the default), then the test functions.

pub const CARTPOLE_SLUG: &'static str = "cartpole"

The slug of Self::Cartpole.

pub const CUSTOM_SLUG: &'static str = "custom"

The slug of Self::Custom.

pub fn new_with_custom(landscape: Landscape) -> Self

The custom function of landscape.

pub const fn slug(&self) -> &'static str

The stable identifier used in URLs and form values.

pub fn landscape(&self) -> Option<&Arc<Landscape>>

The custom function’s landscape.

pub const fn is_learner(&self) -> bool

Whether a trial trains a policy (Self::Cartpole) rather than evaluating a formula.

pub fn from_slug(slug: &str) -> Option<Self>

Parses a Self::slug of a built-in problem.

pub const fn label(&self) -> &'static str

The display name.

pub const fn name(&self) -> &'static str

The study name used for this problem.

pub fn formula(&self) -> Cow<'static, str>

The closed form, as the page prints it; for cart-pole, what a value is.

pub const fn expression(&self) -> Option<&'static str>

The closed form in the custom formula language, so a built-in function can seed a custom one; None for cart-pole, the custom function itself and the one-dimensional fixture.

pub fn description(&self) -> Cow<'static, str>

One factual sentence on the function’s shape and minima; for Cart-pole, on what a trial trains and what its value is.

pub const fn bounds(&self) -> &'static [Bound]

The declared parameters, in suggestion order.

pub const fn optimum_is_known(&self) -> bool

Whether Self::optimum is the exact global minimum value, rather than the custom function’s lowest scanned value. For cart-pole the value 0 is exact but Self::minimizers is empty: no location is claimed (D113).

pub fn optimum(&self) -> f64

The known global minimum value; for the custom function, the lowest scanned value.

pub fn minimizers(&self) -> &[[f64; 2]]

Every global minimizer of a two-parameter test function, or the custom function’s lowest scanned point; empty for cart-pole, whose best value is reached in a region of training runs the page does not claim, and for the one-dimensional fixture.

pub fn ceiling(&self) -> f64

An upper bound of the function over its box: the loss a simulated training run starts from (training), so every curve descends. For a built-in function it is analytic (cart-pole’s value is at most 500); for the custom function it is a margin above the highest scanned value, and the controller still starts a trial no lower than its own value.

pub fn shade(&self, value: f64) -> f64

Compresses a value for the heatmap’s color ramp. Rosenbrock and Himmelblau span three orders of magnitude, so they take a logarithm; the others (cart-pole included) a square root (of the height above the lowest scanned value, for the custom function), which keeps the shallow basins near the optimum distinguishable from the high ridges at the box edges.

pub fn evaluate(&self, params: &[f64]) -> f64

Evaluates the objective at params (one value per Self::bounds entry, in order). For cart-pole this is a whole fixed-seed training run (cartpole::objective), a millisecond or so.

Trait Implementations

impl Clone for Problem

fn clone(&self) -> Problem

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Problem

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for Problem

fn eq(&self, other: &Problem) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for Problem