tempor.data.dataset module¶

Module defining the TemporAI dataset concept in BaseDataset and its derived classes.

tempor.data.dataset.EXCEPTION_MESSAGES = _ExceptionMessages()¶: Reusable error messages for the module.

Bases: ABC

Abstract base class representing a dataset used by TemporAI.

Initialize one of its derived classes (e.g. OneOffPredictionDataset, TimeToEventAnalysisDataset etc.) depending on the type of task.

See also tutorial tutorials/tutorial01_data_format.ipynb for examples of use.

Parameters:¶

time_series : data_typing.DataContainer¶: Data representing time series covariates of the samples. Will be initialized as TimeSeriesSamples.
static : Optional[data_typing.DataContainer], optional¶: Data representing static covariates of the samples. Will be initialized as StaticSamples. Defaults to None.
targets : Optional[data_typing.DataContainer], optional¶: Data representing target (outcome) feature(s) of the samples. Will be initialized as {TimeSeries,Static,Event}Samples depending on problem setting in the derived class. Defaults to None.
treatments : Optional[data_typing.DataContainer], optional¶: Data representing treatment (intervention) feature(s) of the samples. Will be initialized as {TimeSeries,Static,Event}Samples depending on problem setting in the derived class. Defaults to None.
**kwargs : Any: Additional keyword arguments to be passed to the derived class’s _init_predictive method.

predictive : PredictiveTaskData | None¶

property has_static : bool¶

A property returning whether the dataset has static data.

Returns:¶: Whether the dataset has static data.
Return type:¶: bool

property has_predictive_data : bool¶

A property returning whether the dataset has predictive data (targets or treatments).

Returns:¶: Whether the dataset has predictive data.
Return type:¶: bool

property predictive_task : PredictiveTask | None¶

A property returning the predictive task of the dataset (or None).

Returns:¶: The predictive task of the dataset.
Return type:¶: Union[data_typing.PredictiveTask, None]

validate() → None[source]¶: Validate integrity of the dataset.

property time_series : TimeSeriesSamplesBase¶

The property containing the time series covariates of the dataset.

Returns:¶: The time series covariates of the dataset.
Return type:¶: samples.TimeSeriesSamplesBase

property static : StaticSamplesBase | None¶

The property containing the static covariates of the dataset.

Returns:¶: The static covariates of the dataset.
Return type:¶: Optional[samples.StaticSamplesBase]

abstract property fit_ready : bool¶: Returns whether the BaseDataset is in a state ready to be fit on.

Split Dataset into train and test sets.

The arguments test_size … stratify are passed to sklearn.model_selection.train_test_split to generate the split.

Parameters:¶

test_size : Optional[float], optional¶: Passed to sklearn.model_selection.train_test_split. Defaults to None.
train_size : Optional[float], optional¶: Passed to sklearn.model_selection.train_test_split. Defaults to None.
random_state : Union[int, np.random.RandomState, None], optional¶: Passed to sklearn.model_selection.train_test_split. Defaults to None.
shuffle : bool, optional¶: Passed to sklearn.model_selection.train_test_split. Defaults to True.
stratify : Any, optional¶: Passed to sklearn.model_selection.train_test_split. Defaults to None.

Returns:¶

The split tuple (dataset_train, dataset_test).

Return type:¶

Tuple[Self, Self]

split(splitter: KFold | StratifiedKFold, **kwargs: Any) → Generator[tuple[Self, Self], None, None][source]¶

Generate dataset splits according to the scikit-learn splitter (Splitter). The kwargs are passed to the underlying splitter’s split method.

Example

>>> from sklearn.model_selection import KFold
>>> from tempor import plugin_loader
>>> data = plugin_loader.get("prediction.one_off.sine", plugin_type="datasource").load()
>>> kfold = KFold(n_splits=5)
>>> len([(data_train, data_test) for (data_train, data_test) in data.split(splitter=kfold)])
5

Parameters:¶

splitter : Splitter¶: A sklearn splitter.
**kwargs : Any: Additional keyword arguments to be passed to the splitter’s split method.

Yields:¶

Tuple[Self, Self] – (dataset_train, dataset_test) for each split.

Bases: BaseDataset

A BaseDataset subclass for a dataset that does not contain any predictive data (targets or treatments).

property fit_ready : bool¶

Check if the dataset is ready to be fit on.

Returns:¶: Whether the dataset is ready to be fit on.
Return type:¶: bool

predictive : PredictiveTaskData | None¶

Bases: BaseDataset

A BaseDataset subclass for a dataset that can contain predictive data (targets or treatments).

This is an abstract class, to be derived from for different predictive task -specific Dataset s.

predictive : PredictiveTaskData¶

abstract property predict_ready : bool¶: Returns whether the PredictiveDataset is in a state ready to be predict ed on.

Bases: PredictiveDataset

A PredictiveDataset subclass for the one-off prediction problem setting, see BaseDataset docs.

In this setting: targets are required for fitting, will be initialized as StaticSamples.

predictive : OneOffPredictionTaskData¶

property fit_ready : bool¶

Check if the dataset is ready to be fit on.

Returns:¶: Whether the dataset is ready to be fit on.
Return type:¶: bool

property predict_ready : bool¶

Check if the dataset is ready to be predicted on.

Returns:¶: Whether the dataset is ready to be predicted on.
Return type:¶: bool

Bases: PredictiveDataset

A PredictiveDataset subclass for the temporal prediction problem setting, see BaseDataset docs.

In this setting: targets are required for fitting, will be initialized as TimeSeriesSamples.

predictive : TemporalPredictionTaskData¶

property fit_ready : bool¶

Check if the dataset is ready to be fit on.

Returns:¶: Whether the dataset is ready to be fit on.
Return type:¶: bool

property predict_ready : bool¶

Check if the dataset is ready to be predicted on.

Returns:¶: Whether the dataset is ready to be predicted on.
Return type:¶: bool

Bases: PredictiveDataset

A PredictiveDataset subclass for the time-to-event analysis problem setting, see BaseDataset docs.

In this setting: targets are required for fitting, will be initialized as EventSamples.

predictive : TimeToEventAnalysisTaskData¶

property fit_ready : bool¶

Check if the dataset is ready to be fit on.

Returns:¶: Whether the dataset is ready to be fit on.
Return type:¶: bool

property predict_ready : bool¶

Check if the dataset is ready to be predicted on.

Returns:¶: Whether the dataset is ready to be predicted on.
Return type:¶: bool

Bases: PredictiveDataset

A PredictiveDataset subclass for the one-off treatment effects problem setting, see BaseDataset docs.

In this setting: targets are required for fitting, will be initialized as TimeSeriesSamples; treatments are required for both fitting and prediction, will be initialized as EventSamples.

predictive : OneOffTreatmentEffectsTaskData¶

property fit_ready : bool¶

Check if the dataset is ready to be fit on.

Returns:¶: Whether the dataset is ready to be fit on.
Return type:¶: bool

property predict_ready : bool¶

Check if the dataset is ready to be predicted on.

Returns:¶: Whether the dataset is ready to be predicted on.
Return type:¶: bool

Bases: PredictiveDataset

A PredictiveDataset subclass for the temporal treatment effects problem setting, see BaseDataset docs.

In this setting: targets are required for fitting, will be initialized as TimeSeriesSamples; treatments are required for both fitting and prediction, will be initialized as TimeSeriesSamples.

predictive : TemporalTreatmentEffectsTaskData¶

property fit_ready : bool¶

Check if the dataset is ready to be fit on.

Returns:¶: Whether the dataset is ready to be fit on.
Return type:¶: bool

property predict_ready : bool¶

Check if the dataset is ready to be predicted on.

Returns:¶: Whether the dataset is ready to be predicted on.
Return type:¶: bool