Principle:Interpretml Interpret EBM JSON Serialization
| Field | Value |
|---|---|
| Sources | Paper: InterpretML, Repo: InterpretML |
| Domains | Model_Persistence, Interoperability |
| Updated | 2026-02-07 |
Overview
A serialization procedure that converts a fitted EBM model's internal state into a portable JSON representation for storage and exchange.
Description
EBM JSON Serialization converts all model components (bins, scores, intercept, feature metadata, link function) into a JSON-compatible dictionary. It supports multiple detail levels: "minimal" (just for prediction), "interpretable" (for visualization), "mergeable" (for model merging), and "all" (complete state). Array data is converted from numpy to native Python types.
Usage
Use this when you need to save a trained EBM to a file, transmit it over an API, or share it with systems that can consume JSON.
Theoretical Basis
The serialization maps the mathematical model F(x) = B_0 + Sum f_t(x_St) to a JSON structure where each term t stores its bin definitions (cut points or category maps) and the lookup table f_t.