What-If Tool in one line: What-If Tool visually analyzes machine learning models and their behavior.
What-If Tool (WIT) is a plugin for Python notebooks and TensorBoard that provides insights into machine learning models. It helps machine learning systems developers understand their model’s performance via a visual interface, using just a few lines of code.
WIT can be used to interpret both classification and regression models, and trained on tabular, image, or text datasets.
When machine learning models are trained, it’s often unclear what biases the model might display or how it might perform on different data.
WIT allows developers to investigate machine learning models for performance and fairness. They can explore their model in different ways, including seeing the importance of specific features. Developers can also compare models against each other based on various metrics.
The insights WIT provides allow developers to make calculated tradeoffs when optimizing machine learning models.