ray.rllib.env.env_runner.EnvRunner#

class ray.rllib.env.env_runner.EnvRunner(*, config: AlgorithmConfig, **kwargs)[source]#

Bases: FaultAwareApply

Base class for distributed RL-style data collection from an environment.

The EnvRunner API’s core functionalities can be summarized as: - Gets configured via passing a AlgorithmConfig object to the constructor. Normally, subclasses of EnvRunner then construct their own environment (possibly vectorized) copies and RLModules/Policies and use the latter to step through the environment in order to collect training data. - Clients of EnvRunner can use the sample() method to collect data for training from the environment(s). - EnvRunner offers parallelism via creating n remote Ray Actors based on this class. Use ray.remote([resources])(EnvRunner) method to create the corresponding Ray remote class. Then instantiate n Actors using the Ray [ctor].remote(...) syntax. - EnvRunner clients can get information about the server/node on which the individual Actors are running.

PublicAPI (alpha): This API is in alpha and may change before becoming stable.

Methods

`__init__`	Initializes an EnvRunner instance.
`apply`	Calls the given function with this Actor instance.
`assert_healthy`	Checks that self.__init__() has been completed properly.
`get_metrics`	Returns metrics (in any form) of the thus far collected, completed episodes.
`get_spaces`	Returns a dict mapping ModuleIDs to 2-tuples of obs- and action space.
`make_env`	Creates the RL environment for this EnvRunner and assigns it to `self.env`.
`make_module`	Creates the RLModule for this EnvRunner and assigns it to `self.module`.
`ping`	Ping the actor.
`sample`	Returns experiences (of any form) sampled from this EnvRunner.
`sample_get_state_and_metrics`	Convenience method for fast, async algorithms.
`stop`	Releases all resources used by this EnvRunner.