Implementation:Interpretml Interpret Process Terms For Merge

Field	Value
Sources	Repo: InterpretML
Domains	Federated_Learning, Machine_Learning
Updated	2026-02-07

Overview

Concrete tool for finalizing merged EBM model outputs by aggregating scores and cleaning bins, reusing the process_terms utility in the merge context.

Description

This implementation reuses process_terms and remove_extra_bins from _utils.py, plus restore_missing_value_zeros from _tensor.py, in the specific context of finalizing a merged EBM model. After harmonized scores from multiple models have been combined, these functions compute the final averaged scores, standard deviations, and clean up unused bin levels.

Usage

This is called internally by merge_ebms as its final step.

Code Reference

Field	Value
Source	interpretml/interpret
Primary File	python/interpret-core/interpret/glassbox/_ebm/_utils.py
process_terms Lines	200-230
remove_extra_bins Lines	259-296
generate_term_names Lines	233-234
Secondary File	python/interpret-core/interpret/glassbox/_ebm/_tensor.py
restore_missing_value_zeros Lines	69-82

Signatures:

def process_terms(bagged_intercept, bagged_scores, bin_weights, bag_weights):
def remove_extra_bins(term_features, bins):
def generate_term_names(feature_names, term_features):
def restore_missing_value_zeros(tensor, weights):

Import:

from interpret.glassbox._ebm._utils import process_terms, remove_extra_bins, generate_term_names
from interpret.glassbox._ebm._tensor import restore_missing_value_zeros

I/O Contract

Same as Process_Terms (see that page) plus restore_missing_value_zeros:

Function	Inputs	Outputs
process_terms	bagged_intercept, bagged_scores, bin_weights, bag_weights	intercept, term_scores, standard deviations
remove_extra_bins	term_features, bins	None (modifies bins in-place)
generate_term_names	feature_names, term_features	List of human-readable term name strings
restore_missing_value_zeros	tensor (ndarray), weights (ndarray)	None (modifies tensor in-place)

Usage Examples

# Called internally by merge_ebms:
# intercept, term_scores, stds = process_terms(
#     bagged_intercept, bagged_scores, bin_weights, bag_weights
# )
# remove_extra_bins(term_features, bins)
# term_names = generate_term_names(feature_names, term_features)
# for scores, weights in zip(term_scores, bin_weights):
#     restore_missing_value_zeros(scores, weights)

Related Pages

• Principle:Interpretml_Interpret_Model_Finalization_After_Merge
• Environment:Interpretml_Interpret_Python_Core_Environment
• Environment:Interpretml_Interpret_Native_Libebm_Environment