chronax.models.tbats.tbats

`find_harmonics`

tbats_core.find_harmonics

Find optimal number of harmonics for period m using AIC.

Results are cached by (n, m, sum, std) so warm runs are free.

`find_harmonics(y, m)`

Parameter	Type	Default	Description
`y`	`jnp.ndarray`	-	(undocumented)
`m`	`int`	-	(undocumented)

Returns: Tuple[int, jnp.ndarray] (k, z_deseasonalised)

`tbats_model_generator`

tbats_core.tbats_model_generator

Fit a single TBATS specification (Box-Cox + optimisation + filter).

`tbats_model_generator(y, seasonal_periods, k_vector, use_boxcox, bc_lower, bc_upper, use_trend, use_damped_trend, use_arma_errors, ar_coeffs, ma_coeffs, seasonal_blocks=None)`

Parameter	Type	Default	Description
`y`	`jnp.ndarray`	-	(undocumented)
`seasonal_periods`	`Sequence[int]`	-	(undocumented)
`k_vector`	`jnp.ndarray`	-	(undocumented)
`use_boxcox`	`bool`	-	(undocumented)
`bc_lower`	`float`	-	(undocumented)
`bc_upper`	`float`	-	(undocumented)
`use_trend`	`bool`	-	(undocumented)
`use_damped_trend`	`bool`	-	(undocumented)
`use_arma_errors`	`bool`	-	(undocumented)
`ar_coeffs`	`Optional[jnp.ndarray]`	-	(undocumented)
`ma_coeffs`	`Optional[jnp.ndarray]`	-	(undocumented)
`seasonal_blocks`	`Optional[jnp.ndarray]`	`None`	(undocumented)

Returns: Dict

The returned dictionary contains the following keys:

Key	Type	Description
`fitted`	`jnp.ndarray`	Fitted values (one-step ahead predictions).
`errors`	`jnp.ndarray`	Filter innovations (errors).
`sigma2`	`jnp.ndarray`	Estimated variance of the innovations.
`aic`	`float`	Akaike Information Criterion.
`optim_params`	`jnp.ndarray`	Optimised parameter vector.
`F`	`jnp.ndarray`	Final transition matrix.
`w_transpose`	`jnp.ndarray`	Final observation vector (transposed).
`g`	`jnp.ndarray`	Final gain vector.
`x`	`jnp.ndarray`	Final state vector (shape `(1, d)`).
`k_vector`	`jnp.ndarray`	Vector of harmonic counts used for each seasonal period.
`BoxCox_lambda`	`float`	Optimised Box-Cox lambda, or `None`.
`p`	`int`	AR order.
`q`	`int`	MA order.
`ar_coeffs`	`Optional[jnp.ndarray]`	Optimised AR coefficients.
`ma_coeffs`	`Optional[jnp.ndarray]`	Optimised MA coefficients.
`seed_states`	`jnp.ndarray`	Initial state vector used for filtering.
`y_mu`	`jnp.ndarray`	(undocumented)
`y_sigma`	`jnp.ndarray`	(undocumented)
`description`	`Dict`	Empty dictionary (placeholder for model description).

`tbats_model`

tbats_core.tbats_model

Convenience wrapper: fit a single TBATS spec with no ARMA.

`tbats_model(y, seasonal_periods, k_vector, use_boxcox, bc_lower, bc_upper, use_trend, use_damped_trend, use_arma_errors)`

Parameter	Type	Default	Description
`y`	`jnp.ndarray`	-	(undocumented)
`seasonal_periods`	`Sequence[int]`	-	(undocumented)
`k_vector`	`jnp.ndarray`	-	(undocumented)
`use_boxcox`	`bool`	-	(undocumented)
`bc_lower`	`float`	-	(undocumented)
`bc_upper`	`float`	-	(undocumented)
`use_trend`	`bool`	-	(undocumented)
`use_damped_trend`	`bool`	-	(undocumented)
`use_arma_errors`	`bool`	-	(undocumented)

Returns: Dict

Returns the same dictionary structure as tbats_model_generator, but the description key is populated with the input configuration flags.

`tbats_selection`

tbats_core.tbats_selection

Auto-select the best TBATS configuration via AIC comparison.

`tbats_selection(y, seasonal_periods, use_boxcox, bc_lower, bc_upper, use_trend, use_damped_trend, use_arma_errors, early_stop_patience=None, early_stop_tol=0.5)`

Parameter	Type	Default	Description
`y`	`jnp.ndarray`	-	(undocumented)
`seasonal_periods`	`Sequence[int]`	-	(undocumented)
`use_boxcox`	`Optional[bool]`	-	(undocumented)
`bc_lower`	`float`	-	(undocumented)
`bc_upper`	`float`	-	(undocumented)
`use_trend`	`Optional[bool]`	-	(undocumented)
`use_damped_trend`	`Optional[bool]`	-	(undocumented)
`use_arma_errors`	`bool`	-	(undocumented)
`early_stop_patience`	`Optional[int]`	`None`	(undocumented)
`early_stop_tol`	`float`	`0.5`	(undocumented)

Returns: Dict (The fitted model dictionary selected by AIC.)

`tbats_forecast`

tbats_core.tbats_forecast

Multi-step mean forecast from a fitted model dictionary.

`tbats_forecast(mod, h)`

Parameter	Type	Default	Description
`mod`	`Dict`	-	(undocumented)
`h`	`int`	-	(undocumented)

Returns: Dict[str, jnp.ndarray]

The returned dictionary contains the following keys:

Key	Type	Description
`mean`	`jnp.ndarray`	The point forecast (inverse Box-Cox transformed if applicable).
`mean_bc`	`jnp.ndarray` or `None`	The point forecast in the Box-Cox transformed space, or `None`.

`compute_sigmah`

tbats_core.compute_sigmah

Parametric forecast standard deviations from a fitted model dictionary.

`compute_sigmah(mod, h)`

Parameter	Type	Default	Description
`mod`	`Dict`	-	(undocumented)
`h`	`int`	-	(undocumented)

Returns: jnp.ndarray

`tbats_forecast_batch`

tbats_core.tbats_forecast_batch

Vectorised forecasts: x_last has shape (B, d).

`tbats_forecast_batch(F, w, x_last, h)`

Parameter	Type	Default	Description
`F`	`jnp.ndarray`	-	(undocumented)
`w`	`jnp.ndarray`	-	(undocumented)
`x_last`	`jnp.ndarray`	-	(undocumented)
`h`	`int`	-	(undocumented)

Returns: jnp.ndarray

`compute_sigmah_batch`

tbats_core.compute_sigmah_batch

Vectorised sigmah: sigma2 and y_sigma have shape (B,).

`compute_sigmah_batch(F, w, g, sigma2, y_sigma, h, use_boxcox)`

Parameter	Type	Default	Description
`F`	`jnp.ndarray`	-	(undocumented)
`w`	`jnp.ndarray`	-	(undocumented)
`g`	`jnp.ndarray`	-	(undocumented)
`sigma2`	`jnp.ndarray`	-	(undocumented)
`y_sigma`	`jnp.ndarray`	-	(undocumented)
`h`	`int`	-	(undocumented)
`use_boxcox`	`bool`	-	(undocumented)

Returns: jnp.ndarray