DatasetSplit

chronax.data_types.DatasetSplit

Represents a dataset split (context/train/validate) for time series. This class encapsulates the start and end indices of a dataset split as produced by DataLoader, along with optional metadata.

__init__(self, start: int, end: int, metadata: Optional[Dict[str, Any]] = None)

Parameter	Type	Default	Description
start	int	-	Starting index (inclusive) of the split.
end	int	-	Ending index (exclusive) of the split.
metadata	Optional[Dict[str, Any]]	None	Arbitrary additional per-split metadata.

Dataset

chronax.data_types.Dataset

Container for time series dataset with all splits and metadata. This class holds the complete time series data including timestamps, target values, optional covariates, and metadata.

__init__(self, timestamps: List[float], target: List[List[float]], covariate: Optional[List[List[float]]] = None, metadata: Optional[Dict[str, Any]] = None)

Parameter	Type	Default	Description
timestamps	List[float]	-	Array of timestamps (same length as num_steps) or None
target	List[List[float]]	-	2D np.ndarray of Target values
covariate	Optional[List[List[float]]]	None	2D array of Covariate values
metadata	Optional[Dict[str, Any]]	None	Dictionary of metadata including frequency, start time, and other dataset properties.

generate_dataset_split(self, steps: list[tuple[str, int]], stride: int, max_windows: int)

Generate rolling windows over a time series with configurable segments.

Creates sliding windows over the time series data, where each window is split into multiple segments (e.g., context, train, validation) as specified.

Parameters:

Parameter	Type	Default	Description
steps	list[tuple[str, int]]	-	List of (segment_name, num_steps) tuples defining how to split each window. Example: [('context', 24), ('train', 12), ('validate', 6)]
stride	int	-	Number of time steps to advance between consecutive windows. stride=1 creates overlapping windows, stride=window_size creates non-overlapping.
max_windows	int	-	(undocumented)

Yields: tuple[int, Dataset] (Generator yielding (window_index, dataset) pairs where: window_index (int): Zero-based index of the current window; dataset (Dataset): Dataset object containing the window data with segment splits (context, train, validation, etc.) and metadata)

TaskResult

chronax.data_types.TaskResult

Contains results of a model task execution. This class stores the complete results of running a model on a task, including optimal hyperparameters, evaluation metrics, and dataset/task metadata.

__init__(self, optimal_hyperparameters: Dict[int, Dict[str, float]], final_evaluations: Dict[str, float], dataset_path: str, context_window: int, forecast_horizon: int, model_type: Literal["deterministic", "stochastic", "hybrid"], tuning_loss: str, dataset_normalize: bool, dataset_handle_missing: Literal["interpolate", "mean", "median", "drop", "forward_fill", "backward_fill"])

Parameter	Type	Default	Description
optimal_hyperparameters	Dict[int, Dict[str, float]]	-	Dictionary mapping window index to optimal hyperparameter dictionary for that window.
final_evaluations	Dict[str, float]	-	Dictionary mapping metric names to averaged evaluation scores across all windows.
dataset_path	str	-	Path to the dataset file used for this task.
context_window	int	-	Number of context steps used for training.
forecast_horizon	int	-	Number of steps forecasted ahead.
model_type	Literal["deterministic", "stochastic", "hybrid"]	-	Type of model used (deterministic for point forecasts, stochastic for probabilistic, hybrid for both).
tuning_loss	str	-	Loss metric used for hyperparameter selection.
dataset_normalize	bool	-	Whether the dataset was normalized.
dataset_handle_missing	Literal["interpolate", "mean", "median", "drop", "forward_fill", "backward_fill"]	-	Strategy used for handling missing values.