Trait border_core::core::base::Agent

pub trait Agent<E: Env>: Policy<E> {
    fn train(&mut self);
    fn eval(&mut self);
    fn is_train(&self) -> bool;
    fn observe(&mut self, step: Step<E>) -> Option<Record>;
    fn push_obs(&self, obs: &E::Obs);
    fn save<T: AsRef<Path>>(&self, path: T) -> Result<()>;
    fn load<T: AsRef<Path>>(&mut self, path: T) -> Result<()>;
}

Represents a trainable policy on an environment.

Required methods

fn train(&mut self)

Set the policy to training mode.

fn eval(&mut self)

Set the policy to evaluation mode.

fn is_train(&self) -> bool

Return if it is in training mode.

fn observe(&mut self, step: Step<E>) -> Option<Record>

Observe a crate::core::base::Step object. The agent is expected to do training its policy based on the observation.

If an optimization step was performed, it returns Some(crate::core::record::Record), otherwise None.

fn push_obs(&self, obs: &E::Obs)

Push observation to the agent. This method is used when resetting the environment.

fn save<T: AsRef<Path>>(&self, path: T) -> Result<()>

Save the agent in the given directory. This method commonly creates a number of files consisting the agent into the given directory. For example, DQN agent in border_tch_agent crate saves two Q-networks corresponding to the original and target networks.

fn load<T: AsRef<Path>>(&mut self, path: T) -> Result<()>

Load the agent from the given directory.

Implementors