MLOps Definition

MLOps

Some notes on how people are defining MLOps and what general principles we can draw from that.

Definition

  • “MLOps is a set of tools and best practices for bringing machine learning into production” (Huyen, p. 2).
  • “MLOps good” (ml-ops.org).
  • “MLOps good” (lazzeri presentation).

First, What is DevOps?

You’ve probably heard of DevOps before, and understanding this will help us to better understand MLOps.

DevOps Cycle:

  • Software is the artifact
  • Testing
  • Deployment
  • Monitoring

Principles:

  • You should have Developers who know something about Ops.
  • You should have Ops people who know something about software development.
  • Software is best put into production when development and operations are not separated into silos.
  • They need to be put together. They need to be considered together. Holistically.

For MLOps, it’s similar, but more.

  • Machine Learning
  • Data
  • Software