tasks – RL Components: Tasks

class pybrain.rl.environments.task.Task(environment)

A task is associating a purpose with an environment. It decides how to evaluate the observations, potentially returning reinforcement rewards or fitness values. Furthermore it is a filter for what should be visible to the agent. Also, it can potentially act as a filter on how actions are transmitted to the environment.

denormalize(actors)

The function scales the parameters from -1 and 1 to the given interval (min, max) for each actor.

getObservation()

A filtered mapping to getSample of the underlying environment.

getReward()

Compute and return the current reward (i.e. corresponding to the last action performed)

normalize(sensors)

The function scales the parameters to be between -1 and 1. e.g. [(-pi, pi), (0, 1), (-0.001, 0.001)]

performAction(action)

A filtered mapping towards performAction of the underlying environment.

setScaling(sensor_limits, actor_limits)

Expects scaling lists of 2-tuples - e.g. [(-3.14, 3.14), (0, 1), (-0.001, 0.001)] - one tuple per parameter, giving min and max for that parameter. The functions normalize and denormalize scale the parameters between -1 and 1 and vice versa. To disable this feature, use ‘None’.

class pybrain.rl.environments.episodic.EpisodicTask(environment)

Bases: pybrain.rl.environments.task.Task, pybrain.rl.environments.fitnessevaluator.FitnessEvaluator

A task that consists of independent episodes.

addReward()

A filtered mapping towards performAction of the underlying environment.

discount

Discount factor

f(x)

An episodic task can be used as an evaluation function of a module that produces actions from observations, or as an evaluator of an agent.

getTotalReward()

Return the accumulated reward since the start of the episode

isFinished()

Is the current episode over?

performAction(action)

Execute one action.

reset()

Re-initialize the environment