experiments – RL Components: Experiments

class pybrain.rl.experiments.Experiment(task, agent)

An experiment matches up a task with an agent and handles their interactions.

doInteractions(number=1)

The default implementation directly maps the methods of the agent and the task. Returns the number of interactions done.

class pybrain.rl.experiments.EpisodicExperiment(task, agent)

Bases: pybrain.rl.experiments.experiment.Experiment

The extension of Experiment to handle episodic tasks.

doEpisodes(number=1)

Do one episode, and return the rewards of each step as a list.

class pybrain.rl.experiments.ContinuousExperiment(task, agent)

Bases: pybrain.rl.experiments.experiment.Experiment

The extension of Experiment to handle continuous tasks.

doInteractionsAndLearn(number=1)

Execute a number of steps while learning continuously. no reset is performed, such that consecutive calls to this function can be made.