Online model inference with Feast
In the last section, we discussed how to use Feast in batch model inference. Now, it's time to look at the online model use case. One of the requirements of online model inference is that it should return results in low latency and also be invoked from anywhere. One of the common paradigms is to expose the model as a REST API endpoint. In the Model packaging section, we logged the model using the joblib
library. That model needs to be wrapped with the RESTful framework to be deployable as a REST endpoint. Not only that but the features also need to be fetched in real time when the inference endpoint is invoked. Unlike in Chapter 1, An Overview of the Machine Learning Life Cycle, where we didn't have the infrastructure for serving features in real time, here, we already have that in place thanks to Feast. However, we need to run the command to sync offline features to the online store using the Feast library. Let's do that first. Later...