mlos_bench.schedulers package¶

Interfaces and implementations of the optimization loop scheduling policies.

class mlos_bench.schedulers.Scheduler(*, config: Dict[str, Any], global_config: Dict[str, Any], environment: Environment, optimizer: Optimizer, storage: Storage, root_env_config: str)¶

Bases: object

Base class for the optimization loop scheduling policies.

Attributes:

max_trials: Gets the maximum number of trials to run for a given experiment, or -1 for no limit.
ran_trials: Get the list of trials that were run.
trial_config_repeat_count: Gets the number of trials to run for a given config.
trial_count: Gets the current number of trials run for the experiment.

Methods

`get_best_observation`()	Get the best observation from the optimizer.
`load_config`(config_id)	Load the existing tunable configuration from the storage.
`not_done`()	Check the stopping conditions.
`run_trial`(trial)	Set up and run a single trial.
`schedule_trial`(tunables)	Add a configuration to the queue of trials.
`start`()	Start the optimization loop.
`teardown`()	Tear down the environment.

get_best_observation() → Tuple[Dict[str, float] | None, TunableGroups | None]¶: Get the best observation from the optimizer.

load_config(config_id: int) → TunableGroups¶: Load the existing tunable configuration from the storage.

property max_trials: int¶: Gets the maximum number of trials to run for a given experiment, or -1 for no limit.

not_done() → bool¶

Check the stopping conditions.

By default, stop when the optimizer converges or max limit of trials reached.

property ran_trials: List[Trial]¶: Get the list of trials that were run.

abstract run_trial(trial: Trial) → None¶

Set up and run a single trial.

Save the results in the storage.

schedule_trial(tunables: TunableGroups) → None¶: Add a configuration to the queue of trials.

abstract start() → None¶: Start the optimization loop.

teardown() → None¶

Tear down the environment.

Call it after the completion of the .start() in the scheduler context.

property trial_config_repeat_count: int¶: Gets the number of trials to run for a given config.

property trial_count: int¶: Gets the current number of trials run for the experiment.

class mlos_bench.schedulers.SyncScheduler(*, config: Dict[str, Any], global_config: Dict[str, Any], environment: Environment, optimizer: Optimizer, storage: Storage, root_env_config: str)¶

Bases: Scheduler

A simple single-threaded synchronous optimization loop implementation.

Attributes:

max_trials: Gets the maximum number of trials to run for a given experiment, or -1 for no limit.
ran_trials: Get the list of trials that were run.
trial_config_repeat_count: Gets the number of trials to run for a given config.
trial_count: Gets the current number of trials run for the experiment.

Methods

`get_best_observation`()	Get the best observation from the optimizer.
`load_config`(config_id)	Load the existing tunable configuration from the storage.
`not_done`()	Check the stopping conditions.
`run_trial`(trial)	Set up and run a single trial.
`schedule_trial`(tunables)	Add a configuration to the queue of trials.
`start`()	Start the optimization loop.
`teardown`()	Tear down the environment.

run_trial(trial: Trial) → None¶

Set up and run a single trial.

Save the results in the storage.

start() → None¶: Start the optimization loop.

mlos_bench.schedulers package¶

Submodules¶