Struct Bc 

pub struct Bc<E, P, R>
where
    P: SubModel1<Output = Tensor>,
    P::Config: DeserializeOwned + Serialize + OutDim + Debug + PartialEq + Clone,
{ /* private fields */ }

Behavior cloning (BC) agent implemented with candle.

P is the type parameter of the policy model.

Trait Implementations

impl<E, P, R> Agent<E, R> for Bc<E, P, R>

where E: Env, P: SubModel1<Output = Tensor>, R: ReplayBufferBase, E::Obs: Into<P::Input>, E::Act: From<P::Output>, P::Config: DeserializeOwned + Serialize + OutDim + Debug + PartialEq + Clone, R::Batch: TransitionBatch, <R::Batch as TransitionBatch>::ObsBatch: Into<P::Input>, <R::Batch as TransitionBatch>::ActBatch: Into<Tensor>,

fn train(&mut self)

For BC agent, this method does nothing.

fn eval(&mut self)

For BC agent, this method does nothing.

fn is_train(&self) -> bool

For BC agent, this method always returns false.

fn save_params(&self, path: &Path) -> Result<Vec<PathBuf>>

Save model parameters in the given directory.

The parameters of the policy_model are saved as policy_model.pt.

fn load_params(&mut self, path: &Path) -> Result<()>

Load model parameters in the given directory.

The parameters of the policy_model are loaded from policy_model.pt.

fn opt(&mut self, buffer: &mut R)

Performs a single optimization step using experiences from the replay buffer.

fn opt_with_record(&mut self, buffer: &mut R) -> Record

Performs an optimization step and returns training metrics.

fn as_any_ref(&self) -> &(dyn Any + 'static)

Returns a reference to the agent as a type-erased Any value.

fn as_any_mut(&mut self) -> &mut (dyn Any + 'static)

Returns a mutable reference to the agent as a type-erased Any value.

impl<E, P, R> Configurable for Bc<E, P, R>

where E: Env, P: SubModel1<Output = Tensor>, E::Obs: Into<P::Input>, E::Act: From<P::Output>, P::Config: DeserializeOwned + Serialize + OutDim + Debug + PartialEq + Clone,

fn build(config: Self::Config) -> Self

Constructs DQN agent.

type Config = BcConfig<P>

The configuration type for this object.

fn build_from_path(path: impl AsRef<Path>) -> Result<Self, Error>

where Self: Sized,

Builds a new instance from a YAML configuration file.

impl<E, P, R> Policy<E> for Bc<E, P, R>

where E: Env, P: SubModel1<Output = Tensor>, E::Obs: Into<P::Input>, E::Act: From<P::Output>, P::Config: DeserializeOwned + Serialize + OutDim + Debug + PartialEq + Clone,

fn sample(&mut self, obs: &E::Obs) -> E::Act

Sample an action.

When action_type is set to BcActionType::Discrete, this method returns the action corresponding to the argmax of the policy model’s output tensor. On the other hand, when action_type is set to BcActionType::Continuous, this method returns the output tensor as is.