Abstract
A key challenge in developing and deploying Machine Learning (ML) systems is understanding their performance across a wide range of inputs. To address this challenge, we created the What-If Tool, an open-source application that allows practitioners to probe, visualize, and analyze ML systems, with minimal coding. The What-If Tool lets practitioners test performance in hypothetical situations, analyze the importance of different data features, and visualize model behavior across multiple models and subsets of input data. It also lets practitioners measure systems according to multiple ML fairness metrics. We describe the design of the tool, and report on real-life usage at different organizations.
Highlights
People working with machine learning (ML) often need to inspect and analyze the models they build or use
This paper describes the design of the tool, and walks through a set of scenarios showing how it has been applied in practice to analyze Machine Learning (ML) systems
We have presented the What-If Tool (WIT), which is designed to let ML practitioners explore and probe ML models
Summary
People working with machine learning (ML) often need to inspect and analyze the models they build or use. Users can perform counterfactual reasoning, investigate decision boundaries, and explore how general changes to data points affect predictions, simulating different realities to see how a model behaves. An important motivation for these visualizations is to enable easy slicing of the data by a combination of different features from the dataset We refer to this as intersectional analysis, a term we chose in reference to Crenshaw’s concept of “intersectionality” [10]. This type of analysis is critical for understanding issues surrounding model fairness investigations [9] The tool supports both local (analyzing the decision on a single data point) and global (understanding model behavior across an entire dataset) model understanding tasks [11]. Our results suggest that supporting exploration of hypotheticals is a powerful way to understand the behavior of ML systems
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have