Feature importance

fyt.core.feature_selection.feature_importance

Feature importance computation strategies using registry pattern.

This module provides a flexible system for computing feature importance using different methods (native, SHAP, permutation).

FeatureImportanceRegistry

Bases: ComponentRegistry[FeatureImportanceStrategy]

Registry for feature importance computation strategies.

compute(method, model, X, y=None, feature_names=None, **kwargs) classmethod

Compute feature importance using the specified method.

Parameters:

Name	Type	Description	Default
method	str	Feature importance method to use (string or enum).	required
model	BaseEstimator	Trained model instance.	required
X	DataFrame	Feature data.	required
y	Series | None	Target data (required for permutation method).	None
feature_names	list[str] | None	Optional list of feature names.	None
**kwargs		Method-specific parameters.	{}

Returns:

Type	Description
DataFrame | None	DataFrame with feature importance or None if computation failed.

Raises:

Type	Description
ValueError	If the method is not supported.

get_available_methods() classmethod

Get list of available feature importance methods.

Returns:

Type	Description
list[str]	List of available method keys.

FeatureImportanceStrategy

Bases: ABC

Abstract base class for feature importance computation strategies.

compute(model, X, y=None, feature_names=None, **kwargs) abstractmethod

Compute feature importance.

Parameters:

Name	Type	Description	Default
model	BaseEstimator	Trained model instance.	required
X	DataFrame	Feature data.	required
y	Series | None	Target data (required for some methods).	None
feature_names	list[str] | None	Optional list of feature names.	None
**kwargs		Additional method-specific parameters.	{}

Returns:

Type	Description
DataFrame | None	DataFrame with columns ['feature', 'importance'] or None if failed.

NativeFeatureImportance

Bases: FeatureImportanceStrategy

Compute feature importance from model's native attribute.

compute(model, X, y=None, feature_names=None, **kwargs)

Compute native feature importance.

Parameters:

Name	Type	Description	Default
model	BaseEstimator	Trained model with feature_importances_ attribute.	required
X	DataFrame	Feature data (used only for feature names).	required
y	Series | None	Not used.	None
feature_names	list[str] | None	List of feature names.	None
**kwargs		Not used.	{}

Returns:

Type	Description
DataFrame | None	DataFrame with feature importance or None if not supported.

PermutationFeatureImportance

Bases: FeatureImportanceStrategy

Compute feature importance using permutation importance.

compute(model, X, y=None, feature_names=None, **kwargs)

Compute permutation-based feature importance.

Parameters:

Name	Type	Description	Default
model	BaseEstimator	Trained model instance.	required
X	DataFrame	Feature data.	required
y	Series | None	Target data (required).	None
feature_names	list[str] | None	List of feature names.	None
**kwargs		Additional parameters for permutation_importance.	{}

Returns:

Type	Description
DataFrame | None	DataFrame with feature importance or None if failed.

ShapFeatureImportance

Bases: FeatureImportanceStrategy

Compute feature importance using SHAP values.