OptunaTuner

class lightautoml.ml_algo.tuning.optuna.OptunaTuner(timeout=1000, n_trials=100, direction='maximize', fit_on_holdout=True, random_state=42, fail_tolerance=0.5)[source]

Bases: ParamsTuner

Wrapper for optuna tuner.

Parameters:

timeout (Optional[int]) – Maximum learning time.
n_trials (Optional[int]) – Maximum number of trials.
direction (Optional[str]) – Direction of optimization. Set minimize for minimization and maximize for maximization.
fit_on_holdout (bool) – Will be used holdout cv-iterator.
random_state (int) – Seed for optuna sampler.
fail_tolerance (float) – the maximum allowed percentage of failed tuner trials. Exception will be thrown after crossing the threshold value.

fit(ml_algo, train_valid_iterator=None)[source]

Tune model.

Parameters:

ml_algo (TypeVar(TunableAlgo, bound= MLAlgo)) – Algo that is tuned.
train_valid_iterator (Optional[TrainValidIterator]) – Classic cv-iterator.

Return type:

Tuple[Optional[TypeVar(TunableAlgo, bound= MLAlgo)], Optional[LAMLDataset]]

Returns:

Tuple (None, None) if an optuna exception raised or fit_on_holdout=True and train_valid_iterator is not HoldoutIterator. Tuple (MlALgo, preds_ds) otherwise.

plot()[source]: Plot optimization history of all trials in a study.