01
SHAP
SHapley Additive exPlanations — a game-theoretic approach to explain the output of any machine learning model. Assigns each feature an importance value for each prediction using Shapley values from cooperative game theory.
10 Libraries — Python Tools for Explainable AI
02
LIME
Local Interpretable Model-Agnostic Explanations — explains the predictions of any classifier or regressor in a faithful way by approximating it locally with an interpretable model around the prediction of interest.
03
ELI5
Explain Like I'm 5 — a Python library for visualizing and debugging various Machine Learning models using a unified API. Supports scikit-learn, XGBoost, LightGBM, Keras and more with feature weight and prediction visualization.
04
Shapash
A Python library by MAIF dedicated to the interpretability of Data Science models — making SHAP and LIME explainability results accessible and understandable to both data scientists and business stakeholders via interactive web dashboards.
05
Anchors
A method for generating human-interpretable IF-THEN rules (anchors) that explain the predictions of a machine learning model with high precision — sufficient conditions that "anchor" the prediction locally regardless of other feature changes.
06
XAI
A Machine Learning library designed with AI explainability at its core — providing tools to identify data bias, explain model predictions across protected groups, and visualize feature importance for responsible AI development.
07
BreakDown
A tool for explaining predictions of linear and non-linear models by decomposing the model's output into the contribution of each input feature — providing waterfall-style breakdowns of individual predictions for transparency.
08
interpret-text
A Microsoft library incorporating state-of-the-art explainers for text-based machine learning models — visualizing token-level explanations via a built-in interactive dashboard. Supports classical NLP models and transformer-based architectures.
09
AI Explainability 360 (AIX360)
An extensible open-source toolkit from IBM Research (now Linux Foundation Trusted AI) with 10 diverse explainability methods and 2 evaluation metrics — helping comprehend how ML models predict labels across the full AI application lifecycle.
10
OmniXAI
Omni eXplainable AI — a Salesforce library addressing common XAI problems in practice: a unified interface supporting SHAP, LIME, gradient-based, counterfactual and integrated-gradient methods across tabular, text, image and time-series data.