Trait relearn::agents::Agent

pub trait Agent<O, A>: Actor<O, A> {
    fn update(&mut self, step: Step<O, A>, logger: &mut dyn TimeSeriesLogger);
}

A learning agent.

Takes actions in a reinforcement learning environment and updates based on the result (including reward).

Required methods

fn update(&mut self, step: Step<O, A>, logger: &mut dyn TimeSeriesLogger)

Update the agent based on the most recent action.

Must be called immediately after the corresponding call to Actor::act, before any other calls to act are made. This allows the agent to internally cache any information used in selecting the action that would also be useful for updating on the result.

Args

• step: The environment step resulting from the most recent call to Actor::act.

Implementations on Foreign Types

impl<T, O, A> Agent<O, A> for Box<T> where
    T: Agent<O, A> + ?Sized, 

fn update(&mut self, step: Step<O, A>, logger: &mut dyn TimeSeriesLogger)

Implementors

impl<AC, OS, AS> Agent<<ProductSpace<(OptionSpace<OS>, OptionSpace<ProductSpace<(AS, IntervalSpace<f64>)>>, BooleanSpace)> as Space>::Element, <AS as Space>::Element> for ResettingMetaAgent<AC, OS, AS> where
    AC: BuildAgent<OS, AS>,
    OS: Space + Clone,
    OS::Element: Clone,
    AS: SampleSpace + Clone,
    AS::Element: Clone, 

impl<O, A, T, HB> Agent<O, A> for BatchUpdateAgent<T, HB> where
    T: Actor<O, A> + BatchUpdate<O, A>,
    HB: HistoryBuffer<O, A>, 

impl<O, AS: SampleSpace> Agent<O, <AS as Space>::Element> for RandomAgent<AS>

impl<OS, AS, P, PU, C, CU> Agent<<OS as Space>::Element, <AS as Space>::Element> for ActorCriticAgent<OS, AS, P, PU, C, CU> where
    OS: FeatureSpace<Tensor> + BatchFeatureSpace<Tensor>,
    AS: ReprSpace<Tensor> + ParameterizedDistributionSpace<Tensor>,
    P: Policy,
    PU: UpdatePolicy<AS>,
    C: Critic,
    CU: UpdateCritic, 

impl<T: Agent<O, A>, O, A> Agent<O, A> for MutexAgentWorker<T>