Struct argmin::solver::simulatedannealing::SimulatedAnnealing

pub struct SimulatedAnnealing<F, R> { /* private fields */ }

Simulated Annealing

Simulated Annealing (SA) is a stochastic optimization method which imitates annealing in metallurgy. Parameter vectors are randomly modified in each iteration, where the degree of modification depends on the current temperature. The algorithm starts with a high temperature (a lot of modification and hence movement in parameter space) and continuously cools down as the iterations progress, hence narrowing down in the search. Under certain conditions, reannealing (increasing the temperature) can be performed. Solutions which are better than the previous one are always accepted and solutions which are worse are accepted with a probability proportional to the cost function value difference of previous to current parameter vector. These measures allow the algorithm to explore the parameter space in a large and a small scale and hence it is able to overcome local minima.

The initial temperature has to be provided by the user as well as the a initial parameter vector (via configure of Executor.

The cooling schedule can be set with SimulatedAnnealing::with_temp_func. For the available choices please see SATempFunc.

Reannealing can be performed if no new best solution was found for N iterations (SimulatedAnnealing::with_reannealing_best), or if no new accepted solution was found for N iterations (SimulatedAnnealing::with_reannealing_accepted) or every N iterations without any other conditions (SimulatedAnnealing::with_reannealing_fixed).

The user-provided problem must implement Anneal which defines how parameter vectors are modified. Please see the Simulated Annealing example for one approach to do so for floating point parameters.

Requirements on the optimization problem

The optimization problem is required to implement CostFunction.

References

Wikipedia

S Kirkpatrick, CD Gelatt Jr, MP Vecchi. (1983). “Optimization by Simulated Annealing”. Science 13 May 1983, Vol. 220, Issue 4598, pp. 671-680 DOI: 10.1126/science.220.4598.671

Implementations

impl<F> SimulatedAnnealing<F, Xoshiro256PlusPlus>

where F: ArgminFloat,

pub fn new(initial_temperature: F) -> Result<Self, Error>

Construct a new instance of SimulatedAnnealing

Takes the initial temperature as input, which must be >0.

Uses the Xoshiro256PlusPlus RNG internally. For use of another RNG, consider using SimulatedAnnealing::new_with_rng.

Example

let sa = SimulatedAnnealing::new(100.0f64)?;

impl<F, R> SimulatedAnnealing<F, R>

where F: ArgminFloat,

pub fn new_with_rng(init_temp: F, rng: R) -> Result<Self, Error>

Construct a new instance of SimulatedAnnealing

Takes the initial temperature as input, which must be >0. Requires a RNG which must implement rand::Rng (and serde::Serialize if the serde1 feature is enabled).

Example

let sa = SimulatedAnnealing::new_with_rng(100.0f64, my_rng)?;

pub fn with_temp_func(self, temperature_func: SATempFunc<F>) -> Self

Set temperature function

The temperature function defines how the temperature is decreased over the course of the iterations. See SATempFunc for the available options. Defaults to SATempFunc::TemperatureFast.

Example

let sa = SimulatedAnnealing::new(100.0f64)?.with_temp_func(SATempFunc::Boltzmann);

pub fn with_stall_accepted(self, iter: u64) -> Self

If there are no accepted solutions for iter iterations, the algorithm stops.

Defaults to u64::MAX.

Example

let sa = SimulatedAnnealing::new(100.0f64)?.with_stall_accepted(1000);

pub fn with_stall_best(self, iter: u64) -> Self

If there are no new best solutions for iter iterations, the algorithm stops.

Defaults to u64::MAX.

Example

let sa = SimulatedAnnealing::new(100.0f64)?.with_stall_best(2000);

pub fn with_reannealing_fixed(self, iter: u64) -> Self

Set number of iterations after which reannealing is performed

Every iter iterations, reannealing (resetting temperature to its initial value) will be performed. This may help in overcoming local minima.

Defaults to u64::MAX.

Example

let sa = SimulatedAnnealing::new(100.0f64)?.with_reannealing_fixed(5000);

pub fn with_reannealing_accepted(self, iter: u64) -> Self

Set the number of iterations that need to pass after the last accepted solution was found for reannealing to be performed.

If no new accepted solution is found for iter iterations, reannealing (resetting temperature to its initial value) is performed. This may help in overcoming local minima.

Defaults to u64::MAX.

Example

let sa = SimulatedAnnealing::new(100.0f64)?.with_reannealing_accepted(5000);

pub fn with_reannealing_best(self, iter: u64) -> Self

Set the number of iterations that need to pass after the last best solution was found for reannealing to be performed.

If no new best solution is found for iter iterations, reannealing (resetting temperature to its initial value) is performed. This may help in overcoming local minima.

Defaults to u64::MAX.

Example

let sa = SimulatedAnnealing::new(100.0f64)?.with_reannealing_best(5000);

Trait Implementations

impl<F: Clone, R: Clone> Clone for SimulatedAnnealing<F, R>

fn clone(&self) -> SimulatedAnnealing<F, R>

Returns a copy of the value.

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

Performs copy-assignment from source.

impl<'de, F, R> Deserialize<'de> for SimulatedAnnealing<F, R>

where F: Deserialize<'de>, R: Deserialize<'de>,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl<F, R> Serialize for SimulatedAnnealing<F, R>

where F: Serialize, R: Serialize,

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl<O, P, F, R> Solver<O, IterState<P, (), (), (), (), F>> for SimulatedAnnealing<F, R>

where O: CostFunction<Param = P, Output = F> + Anneal<Param = P, Output = P, Float = F>, P: Clone, F: ArgminFloat, R: Rng,

fn next_iter( &mut self, problem: &mut Problem<O>, state: IterState<P, (), (), (), (), F>, ) -> Result<(IterState<P, (), (), (), (), F>, Option<KV>), Error>

Perform one iteration of SA algorithm

fn name(&self) -> &str

Name of the solver. Mainly used in Observers.

fn init( &mut self, problem: &mut Problem<O>, state: IterState<P, (), (), (), (), F>, ) -> Result<(IterState<P, (), (), (), (), F>, Option<KV>), Error>

Initializes the algorithm.

fn terminate( &mut self, _state: &IterState<P, (), (), (), (), F>, ) -> TerminationStatus

Used to implement stopping criteria, in particular criteria which are not covered by (terminate_internal.

fn terminate_internal(&mut self, state: &I) -> TerminationStatus

Checks whether basic termination reasons apply.