Data inspection is critical for all steps in the ML development workflow. But it is also time consuming: First of all it takes effort to build and maintain custom visualization pipelines. Second of all, it takes time to sift through data manually.
Spotlight enables you to build scalable data inspection workflows. With Spotlight you can
- Create visualizations for unstructured data with a single line of code
- Leverage model results to find relevant data segments quickly
- Create custom interaction templates for your team
In order to achieve this, the Spotlight UI provides a range of components that offer different perspectives on your data. You can add, remove, and rearrange these components as you wish to create a customized inspection layouts.
Once you have identified a component layout that works well for you, you can save it as a layout for future use. This way, you can quickly load your preferred layout whenever you need it, without having to reconfigure the UI each time.
Spotlight includes the Inspector and the Data Table to help you in analyzing and understanding individual datapoints. Other visualization options like the Similarity Map, the Scatter Plot and the Histogram aide you to understand the overall distribution of your data.
The Filter Bar is a special component that allows you to filter your data based on the values of your features.
Filtering and selecting datapoints can affect how data is presented in the other components and therefore can greatly help you in analyzing and reasoning about your data.
An overview over all available components can be found in the UI components section.