DeepARForecaster

chronax.forecaster.DeepARForecaster

High-level fit/predict API wrapping DeepAR_EncDec.

__init__(self, h, hidden_size=64, dropout_rate=0.1, min_sigma=0.02, seed=0)

Parameter	Type	Default	Description
h	int	-	Forecast horizon.
hidden_size	int	64	LSTM hidden dimension.
dropout_rate	float	0.1	Dropout probability.
min_sigma	float	0.02	Minimum σ floor.
seed	int	0	Master random seed.

fit(self, y_series, x_futr_list=None, x_stat_list=None, input_size=168, num_steps=1000, batch_size=32, learning_rate=0.001, weight_decay=1e-05, verbose=True) -> DeepARForecaster

Fit DeepAR to a collection of time series.

Parameter	Type	Default	Description
y_series	List[jnp.ndarray]	-	List of 1-D JAX arrays (one per series).
x_futr_list	Optional[List[Optional[jnp.ndarray]]]	None	Future exogenous per series [T+h, F] or None.
x_stat_list	Optional[List[Optional[jnp.ndarray]]]	None	Static features per series [S] or None.
input_size	int	168	History window length.
num_steps	int	1000	Training steps.
batch_size	int	32	Ignored (single-series batches currently).
learning_rate	float	1e-3	Learning rate.
weight_decay	float	1e-5	AdamW weight-decay.
verbose	bool	True	Print progress.

Returns: Self (the fitted forecaster).

predict(self, y_series, x_futr_hist=None, x_futr=None, x_stat=None, num_samples=100, seed=0) -> jnp.ndarray

Monte Carlo sample paths.

Parameter	Type	Default	Description
y_series	jnp.ndarray	-	Historical series [T].
x_futr_hist	Optional[jnp.ndarray]	None	(undocumented)
x_futr	Optional[jnp.ndarray]	None	Future exogenous for the forecast horizon [h, F] or None.
x_stat	Optional[jnp.ndarray]	None	Static features [S] or None.
num_samples	int	100	Number of MC paths.
seed	int	0	Random seed.

Returns: jnp.ndarray (Sample paths [num_samples, h]).

quantile(self, y_series, x_futr_hist=None, x_futr=None, x_stat=None, quantiles=(0.1, 0.5, 0.9), num_samples=1000, seed=0) -> Dict[float, jnp.ndarray]

Compute quantile forecasts.

Parameter	Type	Default	Description
y_series	jnp.ndarray	-	(undocumented)
x_futr_hist	Optional[jnp.ndarray]	None	(undocumented)
x_futr	Optional[jnp.ndarray]	None	(undocumented)
x_stat	Optional[jnp.ndarray]	None	(undocumented)
quantiles	Tuple[float, ...]	(0.1, 0.5, 0.9)	(undocumented)
num_samples	int	1000	(undocumented)
seed	int	0	(undocumented)

Returns: Dict[float, jnp.ndarray] (Dict mapping quantile level → [h] array).

forecast(self, y_series, x_futr_hist=None, x_futr=None, x_stat=None) -> Dict[str, jnp.ndarray]

Point forecast (median) with 10/90 uncertainty intervals.

Parameter	Type	Default	Description
y_series	jnp.ndarray	-	(undocumented)
x_futr_hist	Optional[jnp.ndarray]	None	(undocumented)
x_futr	Optional[jnp.ndarray]	None	(undocumented)
x_stat	Optional[jnp.ndarray]	None	(undocumented)

Returns: Dict[str, jnp.ndarray] (Dict with keys median, lower, upper, each [h]).