Struct GaussianActor 

pub struct GaussianActor<P>
where
    P: SubModel1<Output = (Tensor, Tensor)>,
    P::Config: DeserializeOwned + Serialize + OutDim + Clone,
{ /* private fields */ }

Actor with Gaussian policy.

Implementations

impl<P> GaussianActor<P>

where P: SubModel1<Output = (Tensor, Tensor)>, P::Config: DeserializeOwned + Serialize + OutDim + Clone,

pub fn build( config: GaussianActorConfig<P::Config>, device: Device, ) -> Result<GaussianActor<P>>

Constructs GaussianActor.

pub fn forward(&self, x: &P::Input) -> (Tensor, Tensor)

Returns the parameters of Gaussian distribution given an observation.

The type of return values is (Tensor, Tensor). The shape of the both tensors is (batch_size, action_dimension).

pub fn logp<'a>(&self, obs: &P::Input, act: &Tensor) -> Result<Tensor>

Rerurns the log probabilities (densities) of the given actions

pub fn sample(&mut self, obs: &P::Input, train: bool) -> Result<Tensor>

Samples actions.

If train is true, actions are sampled from a Gaussian distribution. Otherwise, the mean of the Gaussian distribution is returned.

pub fn backward_step(&mut self, loss: &Tensor) -> Result<()>

pub fn save(&self, prefix: impl AsRef<Path>) -> Result<PathBuf>

Save variables to prefix + “.pt”.

pub fn load(&mut self, prefix: impl AsRef<Path>) -> Result<()>

Load variables from prefix + “.pt”.

Trait Implementations

impl<P> Clone for GaussianActor<P>

where P: SubModel1<Output = (Tensor, Tensor)>, P::Config: DeserializeOwned + Serialize + OutDim + Clone,

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.