Enum HessianFallbackPolicy 

pub enum HessianFallbackPolicy {
    Error,
    FiniteDifference,
}

What NewtonTrustRegion / Arc should do when an objective returns SecondOrderSample { hessian: None } (i.e. no analytic Hessian was supplied for this evaluation).

The default is FiniteDifference for backward compatibility with opt 0.2.x. Callers with exact analytic Hessians should set Error: a single None then surfaces as a fatal evaluation error instead of silently triggering O(n) extra gradient probes per iteration. This matters most when each gradient probe is itself expensive (nested solves, biobank-scale outer iterations).

Variants

Error

Treat a missing Hessian as a fatal evaluation error. Use this when the caller guarantees the objective always supplies an analytic Hessian — a None then indicates a routing/contract mismatch and should fail loudly rather than be papered over by finite differences.

FiniteDifference

Estimate the Hessian by finite-differencing the gradient when it is missing. This is the historical default. Step size comes from with_fd_hessian_step.

Trait Implementations

impl Clone for HessianFallbackPolicy

fn clone(&self) -> HessianFallbackPolicy

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for HessianFallbackPolicy

impl Debug for HessianFallbackPolicy

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

Formats the value using the given formatter.

impl Default for HessianFallbackPolicy

fn default() -> Self

Returns the “default value” for a type.

impl Eq for HessianFallbackPolicy

impl PartialEq for HessianFallbackPolicy

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for HessianFallbackPolicy