ReferentialGym package

Subpackages

Submodules

ReferentialGym.referential_game module

class ReferentialGym.referential_game.ReferentialGame(datasets, config, modules, pipelines)

Bases: object

train(nbr_epoch: int = 10, logger: tensorboardX.writer.SummaryWriter = None, verbose_period=None)

Module contents

ReferentialGym.make(config, dataset_args)

Create a ReferentialGame with all the different evalutation modes, that are specified by the dataset_args’s mode entry. :param config: Dict that specifies all the important hyperparameters of the game. :param dataset_args: Dict with the following expected entries:

  • dataset_class: None, ‘LabeledDataset’, or `’DualLabeledDataset’ is expected.

    It specifies the class of dataset decorator to use.

  • modes: Dict of training/evaluation mode as keys and corresponding datasets as values.

    ‘test’ and ‘train’ are mandatory.