chronax.models.nbeats.forecaster

`NBEATSForecaster`

chronax.models.NBEATSForecaster · inherits BaseForecaster

High-level fit/forecast wrapper around :class:~chronax.models.nbeats.model.NBEATS.

init(self, h, input_size=-1, stack_types=('identity', 'trend', 'seasonality'), n_blocks=(1, 1, 1), mlp_units=512, mlp_layers=4, n_harmonics=2, n_basis=2, basis='polynomial', activation='relu', shared_weights=False, dropout_prob=0.0, layer_norm=True, *, max_steps=1000, learning_rate=0.001, batch_size=1024, random_seed=0, alias='NBEATS', loss='mae', scale=False, grad_clip=0.0, weight_decay=0.0, use_lr_schedule=False, num_lr_decays=3, window_sampling=True, val_fraction=0.1, val_check_steps=100, early_stop_patience_steps=-1)

(No prose summary provided in docstring.)

Parameter	Type	Default	Description
`h`	`int`	-	Forecast horizon.
`input_size`	`int`	`-1`	History window length; -1 (default) uses `3 * h`.
`stack_types`	`Tuple[str, ...]`	`('identity', 'trend', 'seasonality')`	forwarded to :class:`NBEATSConfig`.
`n_blocks`	`Tuple[int, ...]`	`(1, 1, 1)`	forwarded to :class:`NBEATSConfig`.
`mlp_units`	`int`	`512`	forwarded to :class:`NBEATSConfig`.
`mlp_layers`	`int`	`4`	forwarded to :class:`NBEATSConfig`.
`n_harmonics`	`int`	`2`	forwarded to :class:`NBEATSConfig`.
`n_basis`	`int`	`2`	forwarded to :class:`NBEATSConfig`.
`basis`	`str`	`'polynomial'`	forwarded to :class:`NBEATSConfig`.
`activation`	`str`	`'relu'`	forwarded to :class:`NBEATSConfig`.
`shared_weights`	`bool`	`False`	forwarded to :class:`NBEATSConfig`.
`dropout_prob`	`float`	`0.0`	forwarded to :class:`NBEATSConfig`.
`layer_norm`	`bool`	`True`	forwarded to :class:`NBEATSConfig`.
`max_steps`	`int`	`1000`	Total optimiser steps.
`learning_rate`	`float`	`0.001`	Adam(W) peak learning rate.
`batch_size`	`int`	`1024`	Windows sampled per training step.
`random_seed`	`int`	`0`	PRNG seed for parameter init and window sampling.
`alias`	`str`	`'NBEATS'`	Display name for external reporting.
`loss`	`Union[str, LossFn]`	`'mae'`	Registered name (`"mae"`, `"mse"`) from :mod:`chronax.models.nbeats.loss` or a callable `(y, y_hat) -> scalar`.
`scale`	`bool`	`False`	Apply per-window `RobustScaler` during training and inference. Strongly recommended; matches NeuralForecast's `scaler_type`.
`grad_clip`	`float`	`0.0`	Global gradient-norm clip (0 = disabled).
`weight_decay`	`float`	`0.0`	AdamW weight decay (0 = plain Adam).
`use_lr_schedule`	`bool`	`False`	Use warmup-cosine LR schedule; when False uses constant `learning_rate`.
`num_lr_decays`	`int`	`3`	(undocumented)
`window_sampling`	`bool`	`True`	Use all rolling windows (True, recommended) vs only the last window per series (False).
`val_fraction`	`float`	`0.1`	Fraction of windows held out for validation (0 = none).
`val_check_steps`	`int`	`100`	Evaluate validation loss every this many steps.
`early_stop_patience_steps`	`int`	`-1`	Stop when val loss does not improve for this many consecutive checks; -1 = disabled.

`fit(self, y: SeriesLike, *, verbose: bool = False) -> NBEATSForecaster`

Train the model in-place on y.

Parameter	Type	Default	Description
`y`	`SeriesLike`	-	A single 1-D series or a list of 1-D series (panel mode).
`verbose`	`bool`	`False`	Print training / validation loss every `max(1, max_steps // 10)` steps.

Returns: Self (the fitted forecaster; sets self.model_).

`forecast(self, y: Optional[SeriesLike] = None, h: Optional[int] = None) -> jnp.ndarray`

Produce point forecasts from the last input_size of each series.

Parameter	Type	Default	Description
`y`	`Optional[SeriesLike]`	`None`	Series to forecast for (required; training data is not retained).
`h`	`Optional[int]`	`None`	Horizon override — must match `config.h` if provided.

Returns: jnp.ndarray of shape [h] for univariate input or [n_series, h] for panel input.

`predict(self, h: Optional[int] = None, X: Optional[jnp.ndarray] = None, level: Optional[List[Union[int, float]]] = None) -> dict`

Forecast from the series passed to :meth:fit.

Satisfies the BaseForecaster contract: unlike :meth:forecast, this does not take y explicitly — it reuses the series cached at fit time.

Parameter	Type	Default	Description
`h`	`Optional[int]`	`None`	Forecast horizon. Defaults to `config.h` when None.
`X`	`Optional[jnp.ndarray]`	`None`	Reserved for future exogenous regressors; unused.
`level`	`Optional[List[Union[int, float]]]`	`None`	Not yet supported for this model.

Returns: dict: {"mean": jnp.ndarray}. Raises: NotImplementedError

`fit_predict(self, y: SeriesLike, **kwargs) -> jnp.ndarray`

Fit on y and immediately forecast for the same series.

Parameter	Type	Default	Description
`y`	`SeriesLike`	-	(undocumented)
`**kwargs`	-	-	(undocumented)

Returns: jnp.ndarray (undocumented).

`with_config(self, **overrides) -> NBEATSForecaster`

Return a fresh (unfitted) forecaster with overridden config fields.

`repr(self) -> str`

(No summary provided in docstring.)

Attributes

`alias`

'NBEATS'

`fitted`

bool

Returns True if the forecaster has been trained.

`state`

TrainState

The internal training state holding parameters and optimizer state.

Raises: RuntimeError if the forecaster has not been fit yet.

NBEATSForecaster

fit(self, y: SeriesLike, *, verbose: bool = False) -> NBEATSForecaster

forecast(self, y: Optional[SeriesLike] = None, h: Optional[int] = None) -> jnp.ndarray

predict(self, h: Optional[int] = None, X: Optional[jnp.ndarray] = None, level: Optional[List[Union[int, float]]] = None) -> dict

fit_predict(self, y: SeriesLike, **kwargs) -> jnp.ndarray

with_config(self, **overrides) -> NBEATSForecaster

__repr__(self) -> str