noether.training.runners.hydra_runner

Attributes

logger

Classes

HydraRunner

Runs an experiment using @hydra.main as entry point.

Module Contents

noether.training.runners.hydra_runner.logger
class noether.training.runners.hydra_runner.HydraRunner

Runs an experiment using @hydra.main as entry point.

run(hydra_config)

Runs an experiment. Given CLI args which contain a path to a hyperparameter file.

Parameters:

  • config – hydra config as a YAML.

  • hydra_config (dict)

Return type:

None

static derive_run_name(name)

Derives a run name by appending hydra overrides to the given name.

Parameters:

name (str)

static main(device, config, static_config)

Main method called from each GPU main process after being spawned and initialized for communication.

Parameters:

  • device (torch.device) – device of the current process.

  • config (noether.core.schemas.schema.ConfigSchema) – configuration of the experiment.

  • static_config (noether.core.configs.StaticConfig)

Return type:

None

static setup_experiment(device, config, static_config, initializer_config_class=ResumeInitializerConfig)

Sets up the experiment objects (datasets, trainer, model, etc.).

Parameters:

  • device (torch.device)

  • config (noether.core.schemas.schema.ConfigSchema)

  • static_config (noether.core.configs.StaticConfig)

  • initializer_config_class (type[noether.core.schemas.initializers.ResumeInitializerConfig] | type[noether.core.schemas.initializers.PreviousRunInitializerConfig])

Return type:

tuple[noether.training.trainers.BaseTrainer, noether.core.models.ModelBase, noether.core.trackers.base.BaseTracker, noether.core.utils.logging.base.MessageCounter]