imitation.policies.interactive

Interactive policies that query the user for actions.

Classes

AtariInteractivePolicy(env, *args, **kwargs)	Interactive policy for Atari environments.
DiscreteInteractivePolicy(observation_space, ...)	Abstract class for interactive policies with discrete actions.
ImageObsDiscreteInteractivePolicy(...[, ...])	DiscreteInteractivePolicy that renders image observations.

class imitation.policies.interactive.AtariInteractivePolicy(env, *args, **kwargs)

Bases: ImageObsDiscreteInteractivePolicy

Interactive policy for Atari environments.

__init__(env, *args, **kwargs)

Builds AtariInteractivePolicy.

features_extractor: BaseFeaturesExtractor

optimizer: th.optim.Optimizer

training: bool

class imitation.policies.interactive.DiscreteInteractivePolicy(observation_space, action_space, action_keys_names, clear_screen_on_query=True)

Bases: NonTrainablePolicy, ABC

Abstract class for interactive policies with discrete actions.

For each query, the observation is rendered and then the action is provided as a keyboard input.

__init__(observation_space, action_space, action_keys_names, clear_screen_on_query=True)

Builds DiscreteInteractivePolicy.

Parameters

• observation_space (Space) – Observation space.

• action_space (Space) – Action space.

• action_keys_names (OrderedDict) – OrderedDict containing pairs (key, name) for every action, where key will be used in the console interface, and name is a semantic action name. The index of the pair in the dictionary will be used as the discrete, integer action.

• clear_screen_on_query (bool) – If True, console will be cleared on every query.

features_extractor: BaseFeaturesExtractor

optimizer: th.optim.Optimizer

training: bool

class imitation.policies.interactive.ImageObsDiscreteInteractivePolicy(observation_space, action_space, action_keys_names, clear_screen_on_query=True)

Bases: DiscreteInteractivePolicy

DiscreteInteractivePolicy that renders image observations.

features_extractor: BaseFeaturesExtractor

optimizer: th.optim.Optimizer

training: bool