Struct Sampler

pub struct Sampler<E, P>
where
    E: Env,
    P: StepProcessor<E>,
{ /* private fields */ }

Encapsulates sampling steps. Specifically it does the followint steps:

1. Samples an action from the Agent, apply to the Env and takes Step.
2. Convert Step into a transition (typically a batch) with StepProcessor.
3. Pushes the trainsition to ReplayBufferBase.
4. Count episode length and pushes to Record.

TODO: being able to set interval_env_record

Implementations

impl<E, P> Sampler<E, P>

where E: Env, P: StepProcessor<E>,

pub fn new(env: E, step_processor: P) -> Self

Creates a sampler.

pub fn sample_and_push<A, R, R_>( &mut self, agent: &mut A, buffer: &mut R_, ) -> Result<Record>

where A: Agent<E, R>, R: ExperienceBufferBase<Item = P::Output> + ReplayBufferBase, R_: ExperienceBufferBase<Item = R::Item>,

Samples transitions and pushes them into the replay buffer.

The replay buffer R_, to which samples will be pushed, has to accept Item that are the same with Agent::R.

pub fn fps(&mut self) -> f32

Returns frames (environment steps) per second, then resets the internal counter.

A frame involves taking action, applying it to the environment, producing transition, and pushing it into the replay buffer.

pub fn reset_fps_counter(&mut self)

Reset stats for computing FPS.