Modeling in Databricks
After features have been created and stored as feature tables, we create training sets and focus on model training. We will cover modeling in terms of leveraging Databricks to facilitate the model lifecycle in Chapter 6. In Chapter 7, we’ll discuss the Unity Catalog Registry and how to use it to track an enormous amount of information from the associated experiments, in addition to details such as model lineage. You can register multiple versions of a model at every stage and can give these different versions aliases, such as champion and challenger, or a more specific alias referring to versions A and B in an A/B test. See aliasing in Figure 2.6.
Figure 2.6 – A user can alias a model with specific names for A/B testing or multi-armed bandits
In Chapter 7, we demonstrate how to trigger a testing script to test every model before having a human review it. Testing models is an efficient practice to reduce the time to...
