Summary
In this chapter, our aim was to look at how model training and scoring change with the feature store. To go through the training and scoring stages of the ML life cycle, we used the resources that were created in the last chapter. In the model training phase, we looked at how data engineers and data scientists can collaborate and work towards building a better model. In model prediction, we discussed batch model scoring and how using an offline store is a cost-effective way of running a batch model. We also built a REST wrapper for the online model and added Feast code to fetch the features for prediction during runtime. At the end of the chapter, we looked at the required changes if there are updates to features during development.
In the next chapter, we will continue using the batch model and the online model that we built in this chapter, productionize them and look at what the challenges are once the models are in production.