treemind.Explainer

class treemind.Explainer

The Explainer class provides methods to analyze and interpret a trained model by examining feature dependencies, split points, interaction effects, and predicted values. This class enables detailed inspection of how individual features and their interactions impact model predictions, allowing for a clearer understanding of the model’s decision-making process.

__call__(model)

The Explainer class provides methods to analyze and interpret trained models by examining feature dependencies, split points, interaction effects, and predicted values. This class enables detailed inspection of how individual features and their interactions impact model predictions, offering insights into the model’s decision-making process.

Parameters:

model (Any) – A trained model instance.

Return type:

None

analyze_feature(columns, back_data=None)

Analyzes feature interactions based on the model’s decision rules and computes metrics that quantify their combined influence on predictions.

Parameters:

• columns (int or list[int]) – The index or list of indices of the features to analyze for interactions. Each index corresponds to a feature in the input data.

• back_data (Optional[ArrayLike], default=None) – Optional data for updating the tree’s leaf counts dynamically. This data allows for re-calculating interaction metrics based on new or baseline data.

Returns:

A DataFrame containing the interaction analysis results with the following columns:

• feature_X_lower_bound, feature_X_upper_bound (float): Lower and upper bounds for each analyzed feature.

• interaction_value (float): The calculated metric representing the combined influence of the features.

• std_dev (float): The standard deviation of the interaction value.

• average_leaf_count (float): The average count of data points across relevant tree leaves.

Return type:

pd.DataFrame

count_node(order=2)

Counts the frequency of feature combinations used in decision splits across all trees in the model.

Parameters:

order (int, default=2) –

The number of features in each combination to count:

• order=1: Counts how often individual features are used in splits.

• order=2: Counts how often pairs of features appear together in splits.

• order=N: Counts combinations of N features.

Returns:

A DataFrame containing the following columns:

• feature_1_index, feature_2_index, …, feature_N_index (int): Indices of the features in the combination, where N equals the order parameter.

• count (int): The number of times this feature combination appears in the splits across all trees.

Return type:

pd.DataFrame