Welcome to this third article in my MLOps sequence. In the primary article, we explored Docker and the way it simplifies utility packaging. In the second article, we managed machine studying fashions utilizing MLflow, Azure, and Docker. Now, on this third half, we’ll deliver all the pieces collectively by constructing a FastAPI utility that serves our beforehand saved mannequin on Azure. This enables us to create a prediction service that may be accessed globally!
An API is sort of a bridge. Everytime you work together with a library in Python, you’re utilizing its API. It’s the general public a part of an utility you can work together with, whereas all the pieces behind it’s hidden.
APIs are generally used to speak with net functions, they usually present a set of URLs that return information (You ship a request with some parameters and get a response again). Most frequently, the info comes again in codecs like JSON or XML, that are straightforward to parse. That is completely different from web sites that return HTML, which incorporates data for rendering pages. With APIs, you get simply the uncooked information.
Some APIs are public, whereas others are non-public. When constructing an API, you resolve what information to share, the best way to…
