gnn_tracking_hpo.tune#

Module Contents#

Classes#

Dispatcher

For most arguments, see corresponding command line interface.

Functions#

add_common_options(parser)

get_timeout_stopper(→ ray.tune.Stopper | None)

Interpret timeout string as seconds.

simple_run_without_tune(→ None)

Simple run without tuning for testing purposes.

main(→ ray.tune.ResultGrid)

Dispatch with ray tune Arguments see Dispater.__call__.
gnn_tracking_hpo.tune.add_common_options(parser: argparse.ArgumentParser)#
gnn_tracking_hpo.tune.get_timeout_stopper(timeout: str | None = None) → ray.tune.Stopper | None#

Interpret timeout string as seconds.

gnn_tracking_hpo.tune.simple_run_without_tune(trainable, suggest_config: Callable) → None#

Simple run without tuning for testing purposes.

class gnn_tracking_hpo.tune.Dispatcher(*, test=False, cpu=False, restore=None, enqueue: None | list[str] = None, only_enqueued=False, fixed: None | str = None, timeout: None | str = None, tags=None, group=None, note=None, fail_fast=False, dname: str | None = None, metric='trk.double_majority_pt1.5', no_tune=False, num_samples: None | int = None, no_scheduler=False, local=False, grace_period=3, no_improvement_patience=10, additional_stoppers=None)#

For most arguments, see corresponding command line interface.

Parameters:

  • grace_period – Grace period for ASHA scheduler.

  • no_improvement_patience – Number of iterations without improvement before stopping

__call__(trainable: type[ray.tune.Trainable], suggest_config: Callable) → ray.tune.ResultGrid#
Parameters:

  • trainable – The trainable to run.

  • suggest_config – A function that returns a config dictionary.

Returns:

get_resources() → dict[str, int]#
get_tuner(trainable: type[ray.tune.Trainable], suggest_config: Callable) → ray.tune.Tuner#
get_no_improvement_stopper() → rt_stoppers_contrib.NoImprovementTrialStopper | None#
get_stoppers() → list[ray.tune.Stopper]#
get_wandb_callbacks() → list[ray.tune.Callback]#
get_callbacks() → list[ray.tune.Callback]#
points_to_evaluate() → list[dict[str, Any]]#
get_optuna_sampler()#
get_num_samples() → int#

Return number of samples/trials to run

get_scheduler() → None | ray.tune.schedulers.ASHAScheduler#
get_tune_config(suggest_config: Callable) → ray.tune.TuneConfig#
get_checkpoint_config() → ray.air.CheckpointConfig#
get_run_config() → ray.air.RunConfig#
gnn_tracking_hpo.tune.main(trainable, suggest_config, *args, **kwargs) → ray.tune.ResultGrid#

Dispatch with ray tune Arguments see Dispater.__call__.