Google What-If Tool

The What-If Tool is an interactive visual interface designed to probe the behavior of trained machine learning models with minimal coding. It is intended for developers and researchers working on responsible ML systems who need to understand model performance across diverse inputs. Key capabilities include testing model performance in hypothetical situations, analyzing the importance of different data features, and visualizing model behavior across multiple models and subsets of input data. The tool also supports evaluation using different ML fairness metrics. It integrates with platforms such as Colaboratory notebooks, Jupyter notebooks, Cloud AI Notebooks, TensorBoard, TFMA, and Fairness Indicators. Compatible models include TF Estimators, models served by TF Serving, Cloud AI Platform Models, and any model that can be wrapped in a Python function. The tool supports binary classification, multi-class classification, regression, and tabular, image, and text data types. It is open source and available for contribution on GitHub.