AutoMFLES (Automated Multi-Feature Locally Exponential Smoothing)

This module provides an automated, parallelized grid-search wrapper for the MFLES forecasting engine. It automatically identifies the optimal hyperparameters using time-series cross-validation.

AutoMFLES

auto_mfles.AutoMFLES · inherits BaseForecaster

Automated MFLES wrapper with parallelized grid-search hyperparameter optimization.

Inherits from BaseForecaster, providing the standard fit() / predict() / forecast() interface. Internally wraps an MFLES base-estimator, automatically selecting optimal hyperparameters via time-series cross-validation.

Attributes

Attribute	Type	Description
alias	str	Custom system tracking ID.
model_	Optional[Dict[str, Any]]	The fitted internal MFLES model and its fitted values.

__init__(self, test_size, season_length=None, n_windows=2, config=None, step_size=None, metric='smape', verbose=False, prediction_intervals=None, alias='AutoMFLES', n_jobs=4)

Initializes the AutoMFLES wrapper class.

Parameter	Type	Default	Description
test_size	int	-	Primary step horizon to evaluate internal cross validation.
season_length	Optional[Union[int, List[int]]]	None	Structural repetition frequency.
n_windows	int	2	Allowed number of cross validation iterations.
config	Optional[List[Dict[str, Any]]]	None	Hardcoded overrides.
step_size	Optional[int]	None	Steps separating CV windows. Defaults to test_size.
metric	str	"smape"	Assessed target loss metric.
verbose	bool	False	Reporting status flag.
prediction_intervals	Optional[Any]	None	Settings dictating conformal bound output.
alias	str	"AutoMFLES"	Custom system tracking ID.
n_jobs	int	4	Authorized CPU Thread limits.

Raises: - ValueError: If test_size or n_windows are <= 0.

fit(self, y, X=None) -> Self

Fits the AutoMFLES engine to the given time series and regressors.

Accepts training time series data, conducts parallelized grid optimization, applies required structural scaling, and fits the base system.

Parameter	Type	Default	Description
y	Union[np.ndarray, jnp.ndarray]	-	Vector array of historical values.
X	Optional[jnp.ndarray]	None	Structural feature regressor inputs.

Returns: Self (A reference mapping back to itself to permit operation chaining. Sets self.model_).

predict(self, h, X=None, level=None) -> Dict[str, Any]

Calculates out-of-sample forward observations utilizing parameterized state mapping.

Parameter	Type	Default	Description
h	int	-	Out-of-sample target evaluation step count.
X	Optional[jnp.ndarray]	None	Expected out-of-sample features array.
level	Optional[List[int]]	None	Percentage integer bounds (e.g. 90, 95).

Returns: dict (Dictionary keys mapping "mean", and conditionally bound arrays.) Keys include {"mean": jnp.ndarray} and optionally {"lo-{lv}": jnp.ndarray, "hi-{lv}": jnp.ndarray} if level is provided.

Raises: - RuntimeError: Tripped if action executed without preceding fit procedure. - ValueError: Tripped if inference attempts feature mapping absent historical features.

forecast(self, y, h, X=None, X_future=None, level=None, fitted=False) -> dict

Stateless fit+predict in one call.

Parameter	Type	Default	Description
y	jnp.ndarray	-	Input time series.
h	int	-	Forecast horizon.
X	jnp.ndarray or None	None	In-sample exogenous variables.
X_future	jnp.ndarray or None	None	Future exogenous variables.
level	list[int \| float] or None	None	Confidence levels for prediction intervals.
fitted	bool	False	Whether to return in-sample fitted values.

Returns: dict (Keys: 'mean', and optionally 'lo-{lv}', 'hi-{lv}', 'fitted'.)