BaseModel

chronax.base_model.BaseModel · inherits ABC

Abstract base class for traditional time series forecasting models. This class provides a unified interface for training, prediction, and evaluation of traditional time series forecasting models. It handles configuration management, data preprocessing, and evaluation metrics computation.

__init__(self, params: Dict[str, Any], settings: Dict[str, Any] | None = None, ParamsClass: PydanticBaseModel | None = None)

Initialize the base model with validated hyperparameters and runtime settings.

Parameter	Type	Default	Description
params	Dict[str, Any]	-	Raw hyperparameters chosen for the current training run.
settings	Dict[str, Any] \| None	None	Model-level execution configuration (device, seed, etc.). Defaults to empty dict.
ParamsClass	PydanticBaseModel \| None	None	Pydantic schema used to validate and coerce params.

train(self, y_context: np.ndarray, y_target: np.ndarray, timestamps_context: np.ndarray, timestamps_target: np.ndarray, x_context: Optional[np.ndarray] = None, x_target: Optional[np.ndarray] = None, **kwargs: dict) -> "BaseModel"

Train the model on given data.

Parameter	Type	Default	Description
y_context	np.ndarray	-	Context window used to initialise the model before fitting.
y_target	np.ndarray	-	Segment used for supervised optimisation during tuning or evaluation.
timestamps_context	np.ndarray	-	Timestamp index aligned with y_context.
timestamps_target	np.ndarray	-	Timestamp index aligned with y_target.
x_context	Optional[np.ndarray]	None	Optional covariate data aligned with y_context, shape (num_steps_context, num_covariates).
x_target	Optional[np.ndarray]	None	Optional covariate data aligned with y_target, shape (num_steps_target, num_covariates).
**kwargs	dict	-	(undocumented)

Returns: BaseModel (The fitted model instance.)

predict(self, y_context: np.ndarray, timestamps_context: np.ndarray, timestamps_target: np.ndarray, x_context: Optional[np.ndarray] = None, x_target: Optional[np.ndarray] = None, **kwargs: dict) -> np.ndarray

Generate predictions for the target time steps.

Parameter	Type	Default	Description
y_context	np.ndarray	-	Context window used for prediction initialization, shape (num_steps_context, num_targets).
timestamps_context	np.ndarray	-	Timestamp index aligned with y_context, shape (num_steps_context,).
timestamps_target	np.ndarray	-	Timestamp index for prediction targets, shape (num_steps_target,).
x_context	Optional[np.ndarray]	None	Optional covariate data aligned with y_context, shape (num_steps_context, num_covariates).
x_target	Optional[np.ndarray]	None	Optional covariate data aligned with timestamps_target, shape (num_steps_target, num_covariates).
**kwargs	dict	-	Additional keyword arguments for model-specific prediction parameters (e.g., freq, num_samples for stochastic models).

Returns: np.ndarray (Predicted values. Shape depends on model type: Deterministic: (num_steps_target, num_targets); Stochastic: (num_samples, num_steps_target, num_targets); Hybrid: Tuple of (point_forecasts, samples))

compute_metrics(self, y_true: np.ndarray, y_pred: np.ndarray, **kwargs) -> Dict[str, float]

Compute all evaluation metrics between true and predicted values using the MetricRegistry class.

Parameter	Type	Default	Description
y_true	np.ndarray	-	True target values (ndarray, shape [num_steps, num_features])
y_pred	np.ndarray	-	Predicted values (ndarray, shape [num_steps, num_features])
**kwargs	-	-	(undocumented)

Returns: Dict[str, float] (Dictionary of computed evaluation metrics (from evaluation.metrics))

get_params(self)

Get the current model parameters.

Returns: Dict[str, Any] (Dictionary of model parameters)

set_params(self, **params: Any) -> "BaseModel"

Set model parameters.

Parameter	Type	Default	Description
**params	Any	-	Model parameters to set (merged into current validated params)

Returns: Self (The model instance with updated parameters)

resolve_weights_path(hf_id: str) -> str

Return a local FUSE path for hf_id if MODEL_WEIGHTS_PATH is set and the directory exists, otherwise return the original HuggingFace identifier.

Parameter	Type	Default	Description
hf_id	str	-	(undocumented)

Returns: str

set_attrs(self, **attrs: Dict[str, Any])

Map validated settings onto the instance for ergonomic access.

| Parameter | Type | Default | Description | |-----------|---------------|-------------| | **attrs | Dict[str, Any] | - | Arbitrary attributes sourced from the settings dictionary. |

get_model_summary(self) -> Dict[str, Any]

Get a summary of the model's properties and performance.

Returns: Dict[str, Any] (Dictionary containing model summary information)

validate_covariate_support

chronax.base_model.validate_covariate_support

Raise ValueError when covariates are provided in an unsupported configuration.

Parameter	Type	Default	Description
x_context	Optional[np.ndarray]	-	Past covariate data (None if not provided).
x_target	Optional[np.ndarray]	-	Future covariate data (None if not provided).
supports_past_only	bool	-	Model can use x_context alone.
supports_future_only	bool	-	Model can use x_target alone.
supports_both	bool	-	Model can use x_context and x_target together.
model_name	str	-	Model name for error messages.

Raises: ValueError

validate_inputs

chronax.base_model.validate_inputs

Decorator to validate input shapes for train/predict methods.