canari.skf_optimizer

canari.skf_optimizer#

This module automates the search for optimal hyperparameters of a SKF instance by leveraging the Ray Tune external library.

class canari.skf_optimizer.SKFOptimizer(initialize_skf: Callable, model_param: dict, param_space: dict, data: dict, detection_threshold: float | None = 0.5, false_rate_threshold: float | None = 0.0, max_timestep_to_detect: int | None = None, num_synthetic_anomaly: int | None = 50, num_optimization_trial: int | None = 50, grid_search: bool | None = False, algorithm: str | None = 'default')[source]#

Bases: object

Optimize hyperparameters for SKF using the Ray Tune external library.

Parameters:

  • initialize_skf (Callable) – Function that returns an SKF instance given a configuration.

  • model_param (dict) – Serializable dictionary for Model obtained from get_dict().

  • param_space (dict) – Parameter search space: two-value lists [min, max] for defining the bounds of the optimization.

  • data (dict) – Input data for adding synthetic anomalies.

  • detection_threshold (float, optional) – Threshold for the target maximal anomaly detection rate. Defaults to 0.5.

  • false_rate_threshold (float, optional) – Threshold for the maximal false detection rate. Defaults to 0.0.

  • max_timestep_to_detect (int, optional) – Maximum number of timesteps to allow detection. Defaults to None (to the end of time series).

  • num_synthetic_anomaly (int, optional) – Number of synthetic anomalies to add. This will create as many time series, because one time series contains only one anomaly. Defaults to 50.

  • num_optimization_trial (int, optional) – Number of trials for optimizer. Defaults to 50.

  • grid_search (bool, optional) – If True, perform grid search. Defaults to False.

  • algorithm (str, optional) – Search algorithm: ‘default’ (OptunaSearch) or ‘parallel’ (ASHAScheduler).

  • 'OptunaSearch'. (Defaults to)

skf_optim#

Best SKF instance after optimization.

param_optim#

Best hyperparameter configuration.

Type:

dict

detection_threshold#

Threshold for detection rate for anomaly detection.

false_rate_threshold#

Threshold for false rate.

get_best_model()[source]#

Retrieves the SKF instance initialized with the best parameters.

Returns:

SKF instance corresponding to the optimal configuration.

Return type:

Any

get_best_param()[source]#

Retrieve the optimized parameters after running optimization.

Returns:

Best hyperparameter values.

Return type:

dict

optimize()[source]#

Run hyperparameter optimization over the defined search space.

Contents