treemind.Explainer

class treemind.Explainer(model)

Extracts interpretable structure from tree models, showing how features and feature combinations influence predictions.

Parameters:

model (Any)

explain(degree, *, back_data=None)

Compute interaction metrics for feature groups of a specified degree.

Parameters:

• degree (int) – Interaction order: 1 for main effects, 2 for pairs, etc.

• back_data (array-like, optional) – Optional dataset used to re-weight statistics for baseline-specific explanations.

Returns:

A mapping from feature index tuples to per-class DataFrames.

Return type:

Result

count_node(degree=2)

Count how often feature groups appear in tree split rules.

Parameters:

degree (int, default=2) – Number of features in each group to count.

Returns:

A DataFrame listing feature groups of the given degree and how often they appear in tree split rules, sorted by descending count.

Return type:

pandas.DataFrame

class treemind.Result

Container holding feature–interaction statistics produced by Explainer.explain(). Acts like a mapping from feature-index tuples to per-class pandas.DataFrame objects with computed metrics.

The __getitem__ method simplifies this by returning a single DataFrame merged across classes, with a "class" column when applicable.

Notes

The content of Result is intended to be read-only.

property data: Dict[Tuple[int, ...], Dict[int, DataFrame]]

Internal dictionary mapping feature-index tuples to per-class DataFrames.

Returns:

The full internal representation of interaction data.

Return type:

dict

importance(combine_classes=False)

Calculate the mean absolute contribution metric (I_abs) for each feature or interaction.

The importance score for a group is:

\[I_{\text{abs}} = \frac{\sum_i \left| E[F\mid i]-\mu \right| \cdot \text{count}_i}{\sum_i \text{count}_i}\]

Parameters:

combine_classes (bool, default=False) – If True, aggregates per-class I_abs into a single weighted value per group. If False, returns a row per class.

Returns:

A DataFrame listing feature or interaction importances (I_abs), sorted by descending importance.

Return type:

pandas.DataFrame