mlpy.experiments.task.EpisodicTask

class mlpy.experiments.task.EpisodicTask(initial_states, terminal_states, env=None)

Bases: mlpy.experiments.task.Task

The episodic task description base class.

This class automatically identifies the task as an episodic task. An episodic task has a set of actions that transitions the agent into a terminal state. Once a terminal state is reached the task is complete.

Parameters:

initial_states : str or State or list[str or State]

List of possible initial states.

terminal_states : str or State or list[str or State]

List of terminal states.

env : Environment, optional

The environment in which the agent performs the task.

Notes

Every deriving class must overwrite the methods _configure_state and _configure_action to configure the classes State and Action, respectively.

For both State and Action the appropriate class variables can be set by calling the following functions:

• set_nfeatures
• set_dtype
• set_description
• set_discretized
• set_minmax_features
• set_states_per_dim

Overwrite the following State and Action methods to allow for more readable descriptions:

• encode
• decode

Additionally, the State class provides a method to check a state’s validity. Overwrite this function to specify valid states:

• is_valid

Attributes

event_delay	Event delay.
is_episodic	Identifies if the task is episodic or not.

Methods

get_reward(state, action)	Retrieve the reward.
is_complete()	Check if the task has completed.
random_initial_state()	Return a random initial state.
request_termination(value)	Request termination of the task.
reset(t, **kwargs)	Reset the task.
sensation(**kwargs)	Gather the state feature information.
terminate(value)	Set the termination flag.
termination_requested()	Check if termination was requested.