Tags

Tags are class-level dictionaries (_tags) that describe component capabilities. For an explanation of tag resolution, MRO merging, and dynamic tags, see Extending Yohou.

Top-Level Tags

These apply to all estimator types.

Tag	Type	Default	Description
estimator_type	str or None	None	One of "transformer", "forecaster", "scorer", "similarity", "splitter"
requires_fit	bool	True	Whether the estimator needs to be fitted before use
non_deterministic	bool	False	Whether the estimator produces non-deterministic results

Input Tags (input_tags)

Tag	Type	Default	Description
requires_time_column	bool	True	Whether a "time" column is required in input DataFrames
pairwise	bool	False	Whether the estimator expects pairwise inputs (e.g., distance matrices)
allow_nan	bool	False	Whether NaN values are allowed in inputs
min_value	float or None	None	Minimum value constraint for inputs (exclusive)

Target Tags (target_tags)

Tag	Type	Default	Description
required	bool	False	Whether fit() requires the y parameter
min_value	float or None	None	Minimum value constraint for target values (exclusive)

Transformer Tags (transformer_tags)

Tag	Type	Default	Description
stateful	bool	False	Whether the transformer maintains an observation window
invertible	bool	False	Whether inverse_transform is available
preserves_dtype	bool	False	Whether the transformer preserves input data types

Forecaster Tags (forecaster_tags)

Tag	Type	Default	Description
forecaster_type	frozenset or None	None	Capabilities: {"point"}, {"interval"}, {"class_proba"}, or combinations
stateful	bool	False	Whether the forecaster uses an observation horizon mechanism
uses_reduction	bool	False	Whether the forecaster converts to tabular regression
uses_target_transformer	bool	False	Whether a target transformer is applied before fitting
uses_feature_transformer	bool	False	Whether a feature transformer is applied before fitting
supports_panel_data	bool	True	Whether the forecaster handles panel data
supports_time_weight	bool	False	Whether time weighting is supported during training
supports_vintage_weight	bool	False	Whether vintage weighting is supported
requires_exogenous	bool	True	Whether X_actual must be provided at fit() time
tracks_observations	bool	True	Whether observation tracking follows the standard pattern

Scorer Tags (scorer_tags)

Tag	Type	Default	Description
prediction_type	str or None	None	One of "point", "interval", "class_proba"
lower_is_better	bool	True	Whether lower scores indicate better performance
requires_calibration	bool	False	Whether calibration data is needed from fit()
symmetric	bool	False	Whether conformity scores are symmetric
multiplicative	bool	False	Whether residuals are normalized by prediction magnitude

Splitter Tags (splitter_tags)

Tag	Type	Default	Description
splitter_type	str or None	None	One of "expanding", "sliding", "gap"
supports_panel_data	bool	False	Whether the splitter handles panel data
produces_non_overlapping_tests	bool	True	Whether test sets across splits are guaranteed non-overlapping
stateful	bool	False	Whether the splitter maintains state across split() calls

Similarity Tags (similarity_tags)

Tag	Type	Default	Description
symmetric	bool	True	Whether similarity(A, B) == similarity(B, A)
requires_predictions	bool	True	Whether predictions (y_pred) are needed in addition to actuals
produces_weights	bool	True	Whether the measure produces weights for conformity scores

Dynamic Tags

Some tags are computed at runtime rather than declared in _tags. Override __sklearn_tags__() to set these programmatically.

Tag	Computed When	Description
forecaster_tags.stateful	target_transformer or feature_transformer has stateful=True	Base forecaster automatically marks itself stateful when any attached transformer is stateful
forecaster_tags.forecaster_type	Multiple base classes are combined	A forecaster that inherits both point and interval capabilities gets {"point", "interval"} automatically
forecaster_tags.requires_exogenous	Subclass overrides __sklearn_tags__()	Set False for forecasters that never use X_actual

Override example:

from yohou.utils.tags import Tags

class MyForecaster(BasePointForecaster):
    def __sklearn_tags__(self) -> Tags:
        tags = super().__sklearn_tags__()
        tags.forecaster_tags.requires_exogenous = False
        tags.forecaster_tags.stateful = True
        return tags

For the full explanation of tag resolution, MRO merging, and how dynamic tags interact with discovery and testing, see Extending Yohou.

See Also

• Extensions: base classes and extension packages for custom components
• Data Catalog: bundled datasets for testing and examples
• Extending Yohou: conceptual background on tag resolution, MRO merging, and dynamic tags