BCRunner¶
-
class
maze.train.trainers.imitation.bc_runners.
BCRunner
(state_dict_dump_file: str, dump_interval: Optional[int], spaces_config_dump_file: str, normalization_samples: int, dataset: omegaconf.DictConfig, eval_concurrency: int)¶ Dev runner for imitation learning. Loads the given trajectory data and trains a policy on top of it using supervised learning.
-
abstract classmethod
create_distributed_eval_env
(env_factory: Callable[], Union[maze.core.env.structured_env.StructuredEnv, maze.core.env.structured_env_spaces_mixin.StructuredEnvSpacesMixin]], eval_concurrency: int, logging_prefix: str) → maze.train.parallelization.vector_env.structured_vector_env.StructuredVectorEnv¶ The individual runners implement the setup of the distributed eval env
-
dataset
: omegaconf.DictConfig¶ Specify the Dataset class used to load the trajectory data for training.
-
run
(n_epochs: Optional[int] = None, evaluator: Optional[maze.train.trainers.common.evaluators.evaluator.Evaluator] = None, eval_every_k_iterations: Optional[int] = None) → None¶ (overrides
TrainingRunner
)Run the training master node. See
run()
. :param evaluator: Evaluator to use for evaluation rollouts :param n_epochs: How many epochs to train for :param eval_every_k_iterations: Number of iterations after which to run evaluation (in addition to evaluations at the end of each epoch, which are run automatically). If set to None, evaluations will run on epoch end only.
-
setup
(cfg: omegaconf.DictConfig) → None¶ (overrides
TrainingRunner
)See
setup()
.
-
abstract classmethod