Implementation:Interpretml Interpret Harmonize Tensor

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

Overview

Concrete tool for remapping EBM score tensors between different bin definitions provided by the InterpretML library.

Description

The _harmonize_tensor function remaps a score tensor from old bin definitions to new unified bin definitions. It handles both continuous (interpolation based on bin boundary overlap) and categorical (redistribution based on evidence weights) features. For interaction terms, it processes all dimensions.

Usage

This is an internal function called by merge_ebms. Not typically called directly.

Code Reference

Field	Value
Source	interpretml/interpret
File	python/interpret-core/interpret/glassbox/_ebm/_merge_ebms.py
Lines	23-277

Signature:

def _harmonize_tensor(
    new_feature_idxs,
    new_bounds,
    new_bins,
    old_feature_idxs,
    old_bounds,
    old_bins,
    old_mapping,
    old_tensor,
    bin_evidence_weight,
):

Import:

from interpret.glassbox._ebm._merge_ebms import _harmonize_tensor

I/O Contract

Inputs:

Parameter	Type	Required	Description
new_feature_idxs	list	Yes	Feature indices in the new merged model
new_bounds	list	Yes	New bin boundaries for each feature
new_bins	list	Yes	Target bin definitions
old_feature_idxs	list	Yes	Feature indices in the source model
old_bounds	list	Yes	Old bin boundaries for each feature
old_bins	list	Yes	Source bin definitions
old_mapping	list	Yes	Mapping from old to new bin indices
old_tensor	ndarray	Yes	Score tensor to remap
bin_evidence_weight	ndarray or None	No	Evidence weights for redistribution

Outputs:

Type	Description
np.ndarray	Tensor remapped to new bin dimensions

Usage Examples

# _harmonize_tensor is called internally by merge_ebms:
# new_tensor = _harmonize_tensor(
#     new_feature_idxs, new_bounds, new_bins,
#     old_feature_idxs, old_bounds, old_bins,
#     old_mapping, old_tensor, bin_evidence_weight
# )
# See merge_ebms() for the standard entry point.

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment