find_hyperparameters

airt.keras.experiments.find_hyperparameters(dataset_name: str, *, monotonicity_indicator: Dict[str, int], final_activation: Union[str, Callable[[TensorLike, TensorLike], TensorLike]], loss: Union[str, Callable[[TensorLike, TensorLike], TensorLike]], metrics: Union[str, Callable[[TensorLike, TensorLike], TensorLike]], hp_params_f: Optional[Callable[[HyperParameters], Dict[str, Any]]] = None, max_trials: int = 100, max_epochs: int = 50, batch_size: int = 8, objective: Union[str, Objective], direction: str, dir_root: Union[Path, str] = 'tuner', seed: int = 42, executions_per_trial: int = 3, max_consecutive_failed_trials: int = 5, patience: int = 10) -> Tuner

Search for optimal hyperparameters

Parameters:

Name	Type	Description	Default
dataset_name	str	name of the dataset, one of "auto", "heart", compas", "blog", "loan"	required
monotonicity_indicator	Dict[str, int]	monotonicity indicator as used in MonoDense.__init__	required
final_activation	Union[str, Callable[[TensorLike, TensorLike], TensorLike]]	final activation of the neural network	required
loss	Union[str, Callable[[TensorLike, TensorLike], TensorLike]]	Tensorflow loss function	required
metrics	Union[str, Callable[[TensorLike, TensorLike], TensorLike]]	Tensorflow metrics function	required
hp_params_f	Optional[Callable[[HyperParameters], Dict[str, Any]]]	a function constructing sampling hyperparameters using Keras Tuner	None
max_trials	int	maximum number of trials	100
max_epochs	int	maximum number of epochs in each trial	50
batch_size	int	batch size	8
objective	Union[str, Objective]	objective, typically f"val_{metrics}"	required
direction	str	direction of the objective, either "min" or "max"	required
dir_root	Union[Path, str]	root directory for storing Keras Tuner data	'tuner'
seed	int	random seed used to guarantee reproducibility of results	42
executions_per_trial	int	number of executions per trial. Set it to number higher than zero for small datasets	3
max_consecutive_failed_trials	int	maximum number of failed trials as used in Keras Tuner	5
patience	int	number of epoch with worse objective before stopping trial early	10

Returns:

Type	Description
Tuner	An instance of Keras Tuner

Source code in airt/keras/experiments.py

def find_hyperparameters(
    dataset_name: str,
    *,
    monotonicity_indicator: Dict[str, int],
    final_activation: Union[str, Callable[[TensorLike, TensorLike], TensorLike]],
    loss: Union[str, Callable[[TensorLike, TensorLike], TensorLike]],
    metrics: Union[str, Callable[[TensorLike, TensorLike], TensorLike]],
    hp_params_f: Optional[Callable[[HyperParameters], Dict[str, Any]]] = None,
    max_trials: int = 100,
    max_epochs: int = 50,
    batch_size: int = 8,
    objective: Union[str, Objective],
    direction: str,
    dir_root: Union[Path, str] = "tuner",
    seed: int = 42,
    executions_per_trial: int = 3,
    max_consecutive_failed_trials: int = 5,
    patience: int = 10,
) -> Tuner:
    """Search for optimal hyperparameters

    Args:
        dataset_name: name of the dataset, one of "auto", "heart", compas", "blog", "loan"
        monotonicity_indicator: monotonicity indicator as used in `MonoDense.__init__`
        final_activation:  final activation of the neural network
        loss: Tensorflow loss function
        metrics: Tensorflow metrics function
        hp_params_f: a function constructing sampling hyperparameters using Keras Tuner
        max_trials: maximum number of trials
        max_epochs: maximum number of epochs in each trial
        batch_size: batch size
        objective: objective, typically f"val_{metrics}"
        direction: direction of the objective, either "min" or "max"
        dir_root: root directory for storing Keras Tuner data
        seed: random seed used to guarantee reproducibility of results
        executions_per_trial: number of executions per trial. Set it to number higher than zero for small datasets
        max_consecutive_failed_trials: maximum number of failed trials as used in Keras Tuner
        patience: number of epoch with worse objective before stopping trial early

    Returns:
        An instance of Keras Tuner

    """
    tf.keras.utils.set_random_seed(seed)

    train_df, test_df = get_train_n_test_data(dataset_name)
    train_ds, test_ds = df2ds(train_df), df2ds(test_df)

    oracle = _TestHyperModel(
        monotonicity_indicator=monotonicity_indicator,
        hp_params_f=hp_params_f,
        final_activation=final_activation,
        loss=loss,
        metrics=metrics,
        train_ds=train_ds,
        batch_size=batch_size,
    )

    tuner = BayesianOptimization(
        oracle,
        objective=Objective(objective, direction),
        max_trials=max_trials,
        seed=seed,
        directory=Path(dir_root),
        project_name=dataset_name,
        executions_per_trial=executions_per_trial,
        max_consecutive_failed_trials=max_consecutive_failed_trials,
    )

    stop_early = tf.keras.callbacks.EarlyStopping(monitor="val_loss", patience=patience)

    tuner.search(
        train_ds.shuffle(len(train_ds)).batch(batch_size).prefetch(2),
        validation_data=test_ds.batch(256),
        callbacks=[stop_early],
        epochs=max_epochs,
    )

    return tuner