Principle:Interpretml Interpret LIME Tabular Explanation
Metadata
| Field | Value |
|---|---|
| Sources | Paper: LIME |
| Domains | Interpretability, Feature_Attribution |
| Updated | 2026-02-07 |
Overview
A model-agnostic explanation method that approximates a black-box model locally with a sparse linear model to explain individual predictions.
Description
LIME (Local Interpretable Model-agnostic Explanations) explains individual predictions by perturbing the input around the sample of interest, observing the black-box model's output on the perturbed samples, and fitting a simple interpretable model (e.g., sparse linear regression) to the local neighborhood. The coefficients of this local model serve as feature importance scores.
Usage
Use LIME when you need quick, local feature attributions for any model and want a simpler, faster alternative to SHAP.
Theoretical Basis
The LIME objective is:
Where f is the black-box model, g is the simple model, πx is a proximity kernel, and Ω penalizes complexity. LIME samples z' ~ N(0,1), maps to z via inverse, evaluates f(z), and fits g as weighted ridge regression.