pub trait Solver<O, I: State> {
// Required methods
fn name(&self) -> &str;
fn next_iter(
&mut self,
problem: &mut Problem<O>,
state: I
) -> Result<(I, Option<KV>), Error>;
// Provided methods
fn init(
&mut self,
_problem: &mut Problem<O>,
state: I
) -> Result<(I, Option<KV>), Error> { ... }
fn terminate_internal(&mut self, state: &I) -> TerminationStatus { ... }
fn terminate(&mut self, _state: &I) -> TerminationStatus { ... }
}Expand description
The interface all solvers are required to implement.
Every solver needs to implement this trait in order to function with the Executor.
It handles initialization (init), each iteration of the solver
(next_iter), and termination of the algorithm
(terminate and terminate_internal).
Only next_iter is mandatory to implement, all others provide default implementations.
A Solver should be (de)serializable in order to work with checkpointing.
§Example
use argmin::core::{
ArgminFloat, Solver, IterState, CostFunction, Error, KV, Problem, TerminationReason, TerminationStatus
};
#[derive(Clone)]
struct OptimizationAlgorithm {}
impl<O, P, G, J, H, R, F> Solver<O, IterState<P, G, J, H, R, F>> for OptimizationAlgorithm
where
O: CostFunction<Param = P, Output = F>,
P: Clone,
F: ArgminFloat
{
fn name(&self) -> &str { "OptimizationAlgorithm" }
fn init(
&mut self,
problem: &mut Problem<O>,
state: IterState<P, G, J, H, R, F>,
) -> Result<(IterState<P, G, J, H, R, F>, Option<KV>), Error> {
// Initialize algorithm, update `state`.
// Implementing this method is optional.
Ok((state, None))
}
fn next_iter(
&mut self,
problem: &mut Problem<O>,
state: IterState<P, G, J, H, R, F>,
) -> Result<(IterState<P, G, J, H, R, F>, Option<KV>), Error> {
// Compute single iteration of algorithm, update `state`.
// Implementing this method is required.
Ok((state, None))
}
fn terminate(&mut self, state: &IterState<P, G, J, H, R, F>) -> TerminationStatus {
// Check if stopping criteria are met.
// Implementing this method is optional.
TerminationStatus::NotTerminated
}
}Required Methods§
sourcefn next_iter(
&mut self,
problem: &mut Problem<O>,
state: I
) -> Result<(I, Option<KV>), Error>
fn next_iter( &mut self, problem: &mut Problem<O>, state: I ) -> Result<(I, Option<KV>), Error>
Computes a single iteration of the algorithm and has access to the optimization problem
definition and the internal state of the solver.
Returns an updated state and optionally a KV which holds key-value pairs used in
Observers.
Provided Methods§
sourcefn init(
&mut self,
_problem: &mut Problem<O>,
state: I
) -> Result<(I, Option<KV>), Error>
fn init( &mut self, _problem: &mut Problem<O>, state: I ) -> Result<(I, Option<KV>), Error>
Initializes the algorithm.
Executed before any iterations are performed and has access to the optimization problem
definition and the internal state of the solver.
Returns an updated state and optionally a KV which holds key-value pairs used in
Observers.
The default implementation returns the unaltered state and no KV.
sourcefn terminate_internal(&mut self, state: &I) -> TerminationStatus
fn terminate_internal(&mut self, state: &I) -> TerminationStatus
Checks whether basic termination reasons apply.
Terminate if
- algorithm was terminated somewhere else in the Executor
- iteration count exceeds maximum number of iterations
- best cost is lower than or equal to the target cost
This can be overwritten; however it is not advised. It is recommended to implement other
stopping criteria via (terminate.
sourcefn terminate(&mut self, _state: &I) -> TerminationStatus
fn terminate(&mut self, _state: &I) -> TerminationStatus
Used to implement stopping criteria, in particular criteria which are not covered by
(terminate_internal.
This method has access to the internal state and returns an TerminationReason.