tbp.monty.frameworks.experiments#

tbp.monty.frameworks.experiments.mode#

class ExperimentMode(value)[source]#

Bases: Enum

Experiment mode.

EVAL = 'eval'#

Evaluation mode.

TRAIN = 'train'#

Training mode.

tbp.monty.frameworks.experiments.monty_experiment#

tbp.monty.frameworks.experiments.object_recognition_experiments#

tbp.monty.frameworks.experiments.pretraining_experiments#

tbp.monty.frameworks.experiments.profile#

tbp.monty.frameworks.experiments.seed#

episode_seed(seed: int, mode: ExperimentMode, episode: int) → int[source]#

Generate a seed for an episode.

In some cases, for each episode, we want to deterministically modify the experiment’s random seed based on the experiment mode and episode. We don’t want to start with the same experiment random seed for each episode. For example, if we want to present objects in a random rotation for each episode, starting with the same random seed for each episode would result in the same rotation for each object in each episode. As another example, if we add noise during training, we want to add different noise during evaluation, even if the episode is the same.

Parameters:

  • seed (int) – The experiment’s random seed.

  • mode (ExperimentMode) – The experiment mode.

  • episode (int) – The episode number.

Return type:

int

Returns:

A seed for the episode in the range [0, 2**32).