alibi

Why we included this project

Teams that need to explain individual predictions get a lot of use out of Alibi, a Python library that covers both local and global explanation methods. It ships implementations of anchors, counterfactual search, integrated gradients, partial dependence, and ALE plots for tabular, text, and image models. Most methods treat the model as a black box, so they work against production prediction endpoints rather than models only available in memory, while the white-box options can use gradients for faster, more faithful explanations. That matters for credit scoring, medical ML, or any governed model work where regulators and auditors ask for reasons behind a decision. Alibi handles classification and regression and integrates with TensorFlow and Keras, so it slots into existing pipelines without forcing a rewrite.