Implementation:Interpretml Interpret Group Importance

Knowledge Sources	Interpretml Interpret
Domains	Machine_Learning, Interpretability
Last Updated	2026-02-07 12:00 GMT

Overview

The group importance module provides utility functions to compute and visualize the importances of groups of features or terms in a fitted Explainable Boosting Machine (EBM), enabling analysis of how sets of related features collectively contribute to model predictions.

Description

This research module extends EBM explanation capabilities by allowing users to group terms together and measure their joint importance:

• compute_group_importance: Computes the importance of a group of terms by summing their per-sample contributions and averaging the absolute values across all samples. For multiclass models, contributions are averaged across classes before aggregation.

• append_group_importance: Computes a group's importance and appends it to an existing (or new) global explanation object. The group importance appears in the Summary bar chart alongside individual term importances.

• get_group_and_individual_importances: Returns a sorted dictionary containing both individual term importances and group importances for specified term groups. Results are sorted in descending order.

• get_individual_importances: Returns a sorted dictionary of all individual EBM term importances.

• get_importance_per_top_groups: Creates incrementally larger groups starting from the most important individual term and expanding. Returns a pandas DataFrame showing how group importance grows as more terms are included.

• plot_importance_per_top_groups: Produces a Plotly line chart of the incremental group importance analysis.

Terms can be specified by name (string) or by index (integer). A "term" in this context covers both individual features and interaction pairs.

Usage

Use these functions when you want to understand how groups of related features (e.g., all demographic features, all lab test features) collectively contribute to an EBM's predictions. This is particularly useful for domain-specific feature grouping analysis or for understanding the marginal contribution of adding features to a group.

Code Reference

Source Location

• Repository: Interpretml_Interpret
• File: python/interpret-core/interpret/glassbox/_ebm/_research/_group_importance.py

Signature

def compute_group_importance(term_list, ebm, X, contributions=None):

def append_group_importance(
    term_list, ebm, X, group_name=None, global_exp=None,
    global_exp_name=None, contributions=None,
):

def get_group_and_individual_importances(
    term_groups_list, ebm, X, contributions=None,
):

def get_individual_importances(ebm, X, contributions=None):

def get_importance_per_top_groups(ebm, X):

def plot_importance_per_top_groups(ebm, X):

Import

from interpret.glassbox._ebm._research._group_importance import (
    compute_group_importance,
    append_group_importance,
    get_group_and_individual_importances,
    get_individual_importances,
    get_importance_per_top_groups,
    plot_importance_per_top_groups,
)

I/O Contract

compute_group_importance

Name	Type	Required	Description
term_list	list of str or int	Yes	List of term names or indices to group together
ebm	fitted EBM	Yes	A fitted Explainable Boosting Machine
X	numpy array	Yes	Samples used to compute group importance
contributions	numpy array	No	Precomputed contributions from ebm.eval_terms(X)

Name	Type	Description
importance	float	The group's importance (average of absolute summed contributions per sample)

append_group_importance

Name	Type	Required	Description
term_list	list of str or int	Yes	List of term names or indices
ebm	fitted EBM	Yes	A fitted EBM
X	numpy array	Yes	Samples for computing group importance
group_name	str	No	Custom name for the group (auto-generated from term names if omitted)
global_exp	EBMExplanation	No	Existing global explanation to append to (creates new one if omitted)
global_exp_name	str	No	Name for new global explanation if global_exp is None
contributions	numpy array	No	Precomputed contributions

Name	Type	Description
global_explanation	EBMExplanation	Global explanation with the group importance appended

get_importance_per_top_groups

Name	Type	Required	Description
ebm	fitted EBM	Yes	A fitted EBM
X	numpy array	Yes	Samples for computing group importance

Name	Type	Description
df	pandas DataFrame	DataFrame with columns: groups, terms_per_group, importances

Usage Examples

Compute Group Importance

from interpret.glassbox import ExplainableBoostingClassifier
from interpret.glassbox._ebm._research._group_importance import (
    compute_group_importance,
    append_group_importance,
)

ebm = ExplainableBoostingClassifier()
ebm.fit(X_train, y_train)

# Compute importance of a group of features
group_imp = compute_group_importance(
    ["Age", "MaritalStatus"], ebm, X_train
)
print(f"Group importance: {group_imp}")

# Append group importance to global explanation
global_exp = append_group_importance(
    ["Age", "MaritalStatus"], ebm, X_train,
    group_name="Demographics"
)
global_exp.visualize(key=None)

Incremental Top-K Group Analysis

from interpret.glassbox._ebm._research._group_importance import (
    get_importance_per_top_groups,
    plot_importance_per_top_groups,
)

df = get_importance_per_top_groups(ebm, X_train)
print(df)

plot_importance_per_top_groups(ebm, X_train)

Related Pages

• Principle:Interpretml_Interpret_Feature_Group_Importance