multitask.OptunaTask

multitask.OptunaTask(
    config=None,
    *,
    dataframe=None,
    data_test=None,
    cache_home=None,
    log_level=logging.INFO,
    **overrides,
)

Task 3 — Optuna Bayesian hyperparameter tuning.

Uses Optuna’s TPE sampler to search for optimal LightGBM hyperparameters, then re-fits with the best discovered parameters.

Examples

from spotforecast2.multitask import OptunaTask

task = OptunaTask(n_trials_optuna=5, predict_size=24)
print(f"Task: {task.TASK}")
print(f"Optuna trials: {task.config.n_trials_optuna}")

Task: optuna
Optuna trials: 5

Methods

Name	Description
run	Run Optuna Bayesian tuning for all targets.

run

multitask.OptunaTask.run(show=True, search_space=None, **kwargs)

Run Optuna Bayesian tuning for all targets.

Parameters

Name	Type	Description	Default
show	bool	If `True`, display prediction figures.	`True`
search_space	Optional[Callable]	Callable `(trial) -> dict`. `None` uses the built-in default.	`None`

Returns

Name	Type	Description
	Dict[str, Any]	Aggregated prediction package. Per-target packages are stored
	Dict[str, Any]	on `self.results["optuna"]`.

Examples

import warnings
from spotforecast2_safe.data.fetch_data import fetch_data, get_package_data_home
from spotforecast2.multitask import OptunaTask

data_home = get_package_data_home()
df = fetch_data(filename=str(data_home / "demo10.csv"))
tiny_df = df.iloc[:500][["A"]]

task = OptunaTask(
    n_trials_optuna=2,
    predict_size=24,
    auto_save_models=False,
    lags_consider=[1, 2, 24],
    number_folds=2,
    verbose=False,
)
task.prepare_data(demo_data=tiny_df)

with warnings.catch_warnings():
    warnings.simplefilter("ignore")
    result = task.run(show=False)

assert "future_pred" in result
assert result.get("validation_passed") is True
print("OptunaTask.run result keys:", sorted(result.keys()))

/tmp/ipykernel_3155/104901005.py:17: DeprecationWarning: Derived pipeline fields (start_download, end_download, data_start, data_end, cov_start, cov_end, end_train_ts, start_train_ts) have moved to task.run_state. Reading them from the config is deprecated and will stop working in the next major release. config.targets continues to hold the user input unchanged; read the resolved list from task.run_state.targets.
  task.prepare_data(demo_data=tiny_df)

OptunaTask.run result keys: ['forecaster', 'future_actual', 'future_pred', 'metrics_future', 'metrics_future_one_day', 'metrics_train', 'train_actual', 'train_pred', 'validation_passed']