Everything we did previously in this tutorial was based on projects, i.e. this logical unit that bundles models and datasets together. However, there are also interesting things you can do with the models in isolation once they are uploaded to Unbox.
In this part of the tutorial, we will briefly show how to do what-if analysis.
Once you've uploaded a model, it becomes available on the platform within the model registry. You can start sending one-off requests through the model to get insights.
Click on the Model on the sidebar to open the model overview page. The first thing you’ll see is the Inference playground. By modifying the feature values in the inference playground and then clicking on Run, you can quickly see what the model prediction would be and the associated explainability scores.
This is a great way to test hypotheses and, thus, make the error analysis procedure more scientific.
- Quickly test hypothesis about model behaviors;
- Compare two model predictions side-by-side.
Updated 2 months ago