Google What-If Tool
- Publisher: Google Research
- Status:
active - Version:
2023 - Release Date:
2019-01-01 - Date Added:
2025-12-11 - Source URL: What-If Tool Official Page
Summary
The What-If Tool (WIT) is a visual interface designed to explore and analyze trained machine learning models with minimal coding. It helps developers test hypothetical scenarios, understand model behavior across inputs, and evaluate fairness metrics, making it useful for responsible ML development.
Key Takeaways
- Probe ML model behavior and performance across a wide range of inputs.
- Analyze feature importance and visualize model outcomes across multiple models or subsets of data.
- Supports multiple task types (binary/multi-class classification, regression) on tabular, image, and text data, with tools for exploring AI fairness.
Additional Sources
- Developer Guide — instructions for contributing and using WIT.
- IEEE VAST ‘19 Systems Paper — research behind WIT.
Tags
ml, fairness, visualization, responsible-ai, tooling
License
Apache-2.0