Alibi (by Seldon)

Alibi Explain is a library of algorithms, referred to as explainers, that provide insights into the predictions of trained machine learning models. It is designed for users who need to understand model behavior, such as data scientists and machine learning engineers. The library helps answer questions about how predictions change based on feature inputs, which features are important for a given prediction, what minimal set of features must be changed to obtain a different prediction, and how each feature contributes to a model's output. The available explainers are constrained by three factors: the type of data the model handles (image, tabular, or textual), the task the model performs (regression or classification), and the type of model used (e.g., neural networks or random forests). Alibi Explain supports both white-box and black-box models, includes functionality for saving and loading explainers, and provides methods for model confidence and prototypes. Its key capability is offering a set of tools to generate explanations for model predictions, with the specific insights available depending on the characteristics of the model and data in question.