Struct QuadraticObjective 

pub struct QuadraticObjective<T: Float + Debug + Send + Sync + 'static> { /* private fields */ }

An anisotropic quadratic bowl: f(x) = ½ Σᵢ cᵢ (xᵢ − x*ᵢ)².

The Hessian is diag(c), the unique minimum is at x* with f = 0, and the condition number is max(c) / min(c) — so a family of these with different c vectors is a controllable benchmark for whether a learned optimizer has picked up anything about curvature.

Implementations

impl<T: Float + Debug + Send + Sync + 'static> QuadraticObjective<T>

pub fn from_curvature(curvature: Vec<f64>, start_value: f64) -> Result<Self>

Build a bowl from an explicit per-coordinate curvature vector, with the optimum at the origin and every start coordinate at start_value.

Errors

Returns OptimError::InvalidConfig for an empty or non-positive curvature vector, or a non-finite start_value.

pub fn isotropic( dimension: usize, curvature: f64, start_value: f64, ) -> Result<Self>

A dimension-dimensional bowl with uniform curvature.

Errors

As Self::from_curvature.

pub fn with_optimum(self, optimum: Vec<f64>) -> Result<Self>

Move the optimum away from the origin.

Errors

Returns OptimError::InvalidConfig if optimum has the wrong length or a non-finite entry.

pub fn labelled(self, label: impl Into<String>) -> Self

Attach a diagnostic label.

pub fn condition_number(&self) -> f64

max(c) / min(c), the condition number of the Hessian.

Trait Implementations

impl<T: Clone + Float + Debug + Send + Sync + 'static> Clone for QuadraticObjective<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug + Float + Debug + Send + Sync + 'static> Debug for QuadraticObjective<T>

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

Formats the value using the given formatter.

impl<T: Float + Debug + Send + Sync + 'static> MetaObjective<T> for QuadraticObjective<T>

fn dimension(&self) -> usize

Number of parameters the task optimizes over.

fn initial_parameters(&self) -> Array1<T>

The point a rollout starts from. Must have length Self::dimension.

fn loss_and_gradient(&self, params: &Array1<T>) -> Result<(T, Array1<T>)>

The loss and its gradient at params.

fn name(&self) -> &str

A short human-readable name, for diagnostics.