PDP

class ordinal_xai.interpretation.pdp.PDP(model: Any, X: DataFrame, y: Series | None = None)

Bases: BaseInterpretation

Partial Dependence Plot (PDP) interpretation method.

This class implements Partial Dependence Plots for interpreting ordinal regression models. PDPs show how a feature affects predictions while accounting for the average effect of other features. This helps understand the relationship between features and predictions in a model-agnostic way.

Parameters:

• model (ordinal_xai.models.base_model.BaseOrdinalModel) – The trained ordinal regression model to interpret

• X (pd.DataFrame) – Dataset used for interpretation, shape (n_samples, n_features)

• y (Optional[pd.Series], default=None) – Target labels, shape (n_samples,)

model

The trained model to interpret

Type:

ordinal_xai.models.base_model.BaseOrdinalModel

X

Dataset used for interpretation

Type:

pd.DataFrame

y

Target labels

Type:

Optional[pd.Series]

feature_names_

Names of features in the dataset

Type:

List[str]

Notes

• PDPs are global interpretation methods that show the average effect of a feature

• The method works by varying one feature while keeping others at their observed values

• For each feature value, predictions are made and averaged across all samples

• This helps understand the marginal effect of features on predictions

__init__(model: Any, X: DataFrame, y: Series | None = None)

Initialize the PDP interpretation method.

Parameters:

• model (ordinal_xai.models.base_model.BaseOrdinalModel) – The trained ordinal regression model to interpret

• X (pd.DataFrame) – Dataset used for interpretation, shape (n_samples, n_features)

• y (Optional[pd.Series], default=None) – Target labels, shape (n_samples,)

Raises:

ValueError – If X is not a DataFrame or y is not a Series

explain(observation_idx: int | None = None, feature_subset: List[str] | List[int] | None = None, plot: bool = False, max_features_per_figure: int = 12) → Dict[str, Dict[str, ndarray]]

Generate Partial Dependence Plots for the selected features.

This method computes and optionally visualizes the partial dependence of the model’s predictions on the selected features. For each feature, it shows how the average prediction changes as the feature value changes, while accounting for the effect of other features.

Parameters:

• observation_idx (Optional[int], default=None) – Ignored (PDP is a global method)

• feature_subset (Optional[Union[List[str], List[int]]], default=None) – List of feature names or indices to plot. If None, all features are used

• plot (bool, default=False) – Whether to create visualizations

• max_features_per_figure (int, default=12) – Maximum number of features to display per figure (for large datasets)

Returns:

Dictionary containing PDP results for each feature: - ‘grid_values’: Feature values used for plotting - ‘average’: Average predictions at each feature value

Return type:

Dict[str, Dict[str, np.ndarray]]

Raises:

• ValueError – If feature_subset contains invalid feature names or indices

• RuntimeError – If model is not fitted and cannot be fitted automatically

_plot_pdps(results: Dict[str, Dict[str, ndarray]], feature_subset: List[str], max_features_per_figure: int = 12) → None

Create visualization of Partial Dependence Plots with pagination for large feature sets.

Parameters:

• results (Dict[str, Dict[str, np.ndarray]]) – PDP results for each feature

• feature_subset (List[str]) – List of features to plot

• max_features_per_figure (int, default=12) – Maximum number of features to display per figure

_abc_impl = <_abc._abc_data object>