Struct optimization_engine::core::panoc::PANOCOptimizer

pub struct PANOCOptimizer<'a, GradientType, ConstraintType, CostType>
where
    GradientType: Fn(&[f64], &mut [f64]) -> FunctionCallResult,
    CostType: Fn(&[f64], &mut f64) -> FunctionCallResult,
    ConstraintType: Constraint,
{ /* private fields */ }

Optimizer using the PANOC algorithm

Implementations

impl<'a, GradientType, ConstraintType, CostType> PANOCOptimizer<'a, GradientType, ConstraintType, CostType>

where GradientType: Fn(&[f64], &mut [f64]) -> FunctionCallResult, CostType: Fn(&[f64], &mut f64) -> FunctionCallResult, ConstraintType: Constraint,

pub fn new( problem: Problem<'a, GradientType, ConstraintType, CostType>, cache: &'a mut PANOCCache ) -> Self

Constructor of PANOCOptimizer

Arguments

• problem: definition of optimization problem
• cache: cache object constructed once

Panic

Does not panic

pub fn with_tolerance(self, tolerance: f64) -> Self

Sets the tolerance on the norm of the fixed-point residual

The algorithm will exit if the form of gamma*FPR drops below this tolerance

Panics

The method panics if the specified tolerance is not positive

pub fn with_akkt_tolerance(self, akkt_tolerance: f64) -> Self

Specify the tolerance $\epsilon$ related to the AKKT condition

$$ \Vert{}\gamma^{-1}(u-u^+) + \nabla f(u) - \nabla f(u^+){}\Vert \leq \epsilon $$

Arguments

• akkt_tolerance: the AKKT-specific tolerance

Returns

Returns the current mutable and updated instance of the provided object

Panics

The method panics if the provided value of the AKKT-specific tolerance is not positive.

pub fn with_max_iter(self, max_iter: usize) -> Self

Sets the maximum number of iterations

Panics

Panics if the provided number of iterations is equal to zero

pub fn with_max_duration(self, max_duation: Duration) -> Self

Sets the maximum solution time, useful in real-time applications

Trait Implementations

impl<'life, GradientType, ConstraintType, CostType> Optimizer for PANOCOptimizer<'life, GradientType, ConstraintType, CostType>

where GradientType: Fn(&[f64], &mut [f64]) -> FunctionCallResult + 'life, CostType: Fn(&[f64], &mut f64) -> FunctionCallResult, ConstraintType: Constraint + 'life,

fn solve(&mut self, u: &mut [f64]) -> Result<SolverStatus, SolverError>

solves a given problem and updates the initial estimate u with the solution