ax.early_stopping¶

Base Strategies¶

class ax.early_stopping.strategies.base.BaseEarlyStoppingStrategy(metric_names: Optional[Iterable[str]] = None, seconds_between_polls: int = 300, min_progression: Optional[float] = None, min_curves: Optional[int] = None, trial_indices_to_ignore: Optional[List[int]] = None, true_objective_metric_name: Optional[str] = None, normalize_progressions: bool = False)[source]¶

Bases: ABC, Base

Interface for heuristics that halt trials early, typically based on early results from that trial.

is_eligible(trial_index: int, experiment: Experiment, df: pandas.DataFrame, map_key: str) → Tuple[bool, Optional[str]][source]¶

Perform a series of default checks for a specific trial trial_index and determines whether it is eligible for further stopping logic:

1. Check for ignored indices based on self.trial_indices_to_ignore

2. Check that df contains data for the trial trial_index

3. Check that the trial has reached self.min_progression

Returns two elements: a boolean indicating if all checks are passed and a str indicating the reason that early stopping is not applied (None if all checks pass).

is_eligible_any(trial_indices: Set[int], experiment: Experiment, df: pandas.DataFrame, map_key: Optional[str] = None) → bool[source]¶

Perform a series of default checks for a set of trials trial_indices and determine if at least one of them is eligible for further stopping logic:

1. Check that at least self.min_curves trials have completed`

2. Check that at least one trial has reached self.min_progression

Returns a boolean indicating if all checks are passed.

This is useful for some situations where if no trials are eligible for stopping, then we can skip costly steps, such as model fitting, that occur before individual trials are considered for stopping.

abstract should_stop_trials_early(trial_indices: Set[int], experiment: Experiment, **kwargs: Dict[str, Any]) → Dict[int, Optional[str]][source]¶

Decide whether to complete trials before evaluation is fully concluded.

Typical examples include stopping a machine learning model’s training, or halting the gathering of samples before some planned number are collected.

Parameters:

• trial_indices – Indices of candidate trials to stop early.

• experiment – Experiment that contains the trials and other contextual data.

Returns:

A dictionary mapping trial indices that should be early stopped to (optional) messages with the associated reason.

class ax.early_stopping.strategies.base.EarlyStoppingTrainingData(X: ndarray, Y: ndarray, Yvar: ndarray, arm_names: List[Optional[str]])[source]¶

Bases: object

Dataclass for keeping data arrays related to model training and arm names together.

Parameters:

• X – An n x d’ array of training features. d’ = d + m, where d is the dimension of the design space and m are the number of map keys. For the case of learning curves, m = 1 since we have only the number of steps as the map key.

• Y – An n x 1 array of training observations.

• Yvar – An n x 1 observed measurement noise.

• arm_names – A list of length n of arm names. Useful for understanding which data come from the same arm.

X: ndarray¶

Y: ndarray¶

Yvar: ndarray¶

arm_names: List[Optional[str]]¶

class ax.early_stopping.strategies.base.ModelBasedEarlyStoppingStrategy(metric_names: Optional[Iterable[str]] = None, seconds_between_polls: int = 300, min_progression: Optional[float] = None, min_curves: Optional[int] = None, trial_indices_to_ignore: Optional[List[int]] = None, true_objective_metric_name: Optional[str] = None, normalize_progressions: bool = False)[source]¶

Bases: BaseEarlyStoppingStrategy

A base class for model based early stopping strategies. Includes a helper function for processing MapData into arrays.

get_training_data(experiment: Experiment, map_data: MapData, max_training_size: Optional[int] = None, outcomes: Optional[Sequence[str]] = None, parameters: Optional[List[str]] = None) → EarlyStoppingTrainingData[source]¶

Processes the raw (untransformed) training data into arrays for use in modeling. The trailing dimensions of X are the map keys, in their originally specified order from map_data.

Parameters:

• experiment – Experiment that contains the data.

• map_data – The MapData from the experiment, as can be obtained by via _check_validity_and_get_data.

• max_training_size – Subsample the learning curve to keep the total number of data points less than this threshold. Passed to MapData’s subsample method as limit_total_rows.

Returns:

An EarlyStoppingTrainingData that contains training data arrays X, Y,

and Yvar + a list of arm names.

ax.early_stopping.strategies.base.get_transform_helper_model(experiment: Experiment, data: Data, transforms: Optional[List[Type[Transform]]] = None) → MapTorchModelBridge[source]¶

Constructs a TorchModelBridge, to be used as a helper for transforming parameters. We perform the default Cont_X_trans for parameters but do not perform any transforms on the observations.

Parameters:

• experiment – Experiment.

• data – Data for fitting the model.

Returns: A torch modelbridge.

Logical Strategies¶

class ax.early_stopping.strategies.logical.AndEarlyStoppingStrategy(left: BaseEarlyStoppingStrategy, right: BaseEarlyStoppingStrategy, seconds_between_polls: int = 300, true_objective_metric_name: Optional[str] = None)[source]¶

Bases: LogicalEarlyStoppingStrategy

should_stop_trials_early(trial_indices: Set[int], experiment: Experiment, **kwargs: Dict[str, Any]) → Dict[int, Optional[str]][source]¶

Decide whether to complete trials before evaluation is fully concluded.

Typical examples include stopping a machine learning model’s training, or halting the gathering of samples before some planned number are collected.

Parameters:

• trial_indices – Indices of candidate trials to stop early.

• experiment – Experiment that contains the trials and other contextual data.

Returns:

A dictionary mapping trial indices that should be early stopped to (optional) messages with the associated reason.

class ax.early_stopping.strategies.logical.LogicalEarlyStoppingStrategy(left: BaseEarlyStoppingStrategy, right: BaseEarlyStoppingStrategy, seconds_between_polls: int = 300, true_objective_metric_name: Optional[str] = None)[source]¶

Bases: BaseEarlyStoppingStrategy

class ax.early_stopping.strategies.logical.OrEarlyStoppingStrategy(left: BaseEarlyStoppingStrategy, right: BaseEarlyStoppingStrategy, seconds_between_polls: int = 300, true_objective_metric_name: Optional[str] = None)[source]¶

Bases: LogicalEarlyStoppingStrategy

classmethod from_early_stopping_strategies(strategies: Sequence[BaseEarlyStoppingStrategy]) → BaseEarlyStoppingStrategy[source]¶

should_stop_trials_early(trial_indices: Set[int], experiment: Experiment, **kwargs: Dict[str, Any]) → Dict[int, Optional[str]][source]¶

Decide whether to complete trials before evaluation is fully concluded.

Typical examples include stopping a machine learning model’s training, or halting the gathering of samples before some planned number are collected.

Parameters:

• trial_indices – Indices of candidate trials to stop early.

• experiment – Experiment that contains the trials and other contextual data.

Returns:

A dictionary mapping trial indices that should be early stopped to (optional) messages with the associated reason.

PercentileEarlyStoppingStrategy¶

class ax.early_stopping.strategies.percentile.PercentileEarlyStoppingStrategy(metric_names: Optional[Iterable[str]] = None, seconds_between_polls: int = 300, percentile_threshold: float = 50.0, min_progression: Optional[float] = 10, min_curves: Optional[int] = 5, trial_indices_to_ignore: Optional[List[int]] = None, true_objective_metric_name: Optional[str] = None, normalize_progressions: bool = False)[source]¶

Bases: BaseEarlyStoppingStrategy

Implements the strategy of stopping a trial if its performance falls below that of other trials at the same step.

should_stop_trial_early(trial_index: int, experiment: Experiment, df: pandas.DataFrame, df_raw: pandas.DataFrame, map_key: str, minimize: bool) → Tuple[bool, Optional[str]][source]¶

Stop a trial if its performance is in the bottom percentile_threshold of the trials at the same step.

Parameters:

• trial_index – Indices of candidate trial to stop early.

• experiment – Experiment that contains the trials and other contextual data.

• df – Dataframe of partial results after applying interpolation, filtered to objective metric.

• df_raw – The original MapData dataframe (before interpolation).

• map_key – Name of the column of the dataset that indicates progression.

• minimize – Whether objective value is being minimized.

Returns:

A tuple (should_stop, reason), where should_stop is True iff the trial should be stopped, and reason is an (optional) string providing information on why the trial should or should not be stopped.

should_stop_trials_early(trial_indices: Set[int], experiment: Experiment, **kwargs: Dict[str, Any]) → Dict[int, Optional[str]][source]¶

Stop a trial if its performance is in the bottom percentile_threshold of the trials at the same step.

Parameters:

• trial_indices – Indices of candidate trials to consider for early stopping.

• experiment – Experiment that contains the trials and other contextual data.

Returns:

A dictionary mapping trial indices that should be early stopped to (optional) messages with the associated reason. An empty dictionary means no suggested updates to any trial’s status.

ThresholdEarlyStoppingStrategy¶

class ax.early_stopping.strategies.threshold.ThresholdEarlyStoppingStrategy(metric_names: Optional[Iterable[str]] = None, seconds_between_polls: int = 300, metric_threshold: float = 0.2, min_progression: Optional[float] = 10, min_curves: Optional[int] = 5, trial_indices_to_ignore: Optional[List[int]] = None, true_objective_metric_name: Optional[str] = None, normalize_progressions: bool = False)[source]¶

Bases: BaseEarlyStoppingStrategy

Implements the strategy of stopping a trial if its performance doesn’t reach a pre-specified threshold by a certain progression.

should_stop_trial_early(trial_index: int, experiment: Experiment, df: pandas.DataFrame, map_key: str, minimize: bool) → Tuple[bool, Optional[str]][source]¶

Stop a trial if its performance doesn’t reach a pre-specified threshold by min_progression.

Parameters:

• trial_index – Indices of candidate trial to stop early.

• experiment – Experiment that contains the trials and other contextual data.

• df – Dataframe of partial results for the objective metric.

• map_key – Name of the column of the dataset that indicates progression.

• minimize – Whether objective value is being minimized.

Returns:

A tuple (should_stop, reason), where should_stop is True iff the trial should be stopped, and reason is an (optional) string providing information on why the trial should or should not be stopped.

should_stop_trials_early(trial_indices: Set[int], experiment: Experiment, **kwargs: Dict[str, Any]) → Dict[int, Optional[str]][source]¶

Stop a trial if its performance doesn’t reach a pre-specified threshold by min_progression.

Parameters:

• trial_indices – Indices of candidate trials to consider for early stopping.

• experiment – Experiment that contains the trials and other contextual data.

Returns:

A dictionary mapping trial indices that should be early stopped to (optional) messages with the associated reason. An empty dictionary means no suggested updates to any trial’s status.