What Is MLOps According to the MLOps Guide?

Updated

MLOps is an operations methodology for moving an experimental machine learning model into production and maintaining it efficiently, according to the MLOps Guide. It applies DevOps practices from the software industry to the machine learning model lifecycle. This definition fits teams that already train models but struggle to keep them running reliably once deployed.

The core idea

The MLOps Guide frames MLOps as the bridge between experimentation and production. A model that performs well in a notebook is not yet a product. MLOps covers the practices needed to close that gap and keep the model healthy afterward.

The guide defines five main features of an MLOps project:

  • Data and model versioning — tracking which data and which model produced a given result
  • Feature management and storing — organizing the inputs models depend on
  • Automation of pipelines and processes — reducing manual steps between training and deployment
  • CI/CD for machine learning — applying continuous integration and delivery to ML artifacts
  • Continuous monitoring of models — watching deployed models for degradation or drift

How MLOps relates to DevOps

The guide's central claim is that MLOps brings DevOps methodology to the ML model lifecycle. Where DevOps treats code as the versioned, tested, continuously delivered artifact, MLOps extends that treatment to data, features, and models as well. That extension is what makes ML operations harder than ordinary software operations: the artifact being shipped changes as data changes.

What the guide covers

The MLOps Guide is intended to help projects and companies build a more reliable MLOps environment. It combines theory with an implementation meant to fit most use cases. Its structure moves from concepts to a working environment:

Section Content
Principles of MLOps Versioning, automation, feature storing, CI/CD for ML, continuous monitoring
Implementation Guide Introduction, tools, and project structure
Environment IAM access, setting up the IBM environment with Terraform, managing the deployment space
Versioning What DVC is, data versioning, working with pipelines
Deployment Deployment with Watson Machine Learning
CI/CD for Machine Learning Implementing tests, continuous integration with CML and GitHub Actions, continuous delivery with CML, GitHub Actions, and Watson ML
Monitoring Monitoring with IBM OpenScale

Tools discussed

The guide names the following tools and their functions:

Tool Function Developer
IBM Watson ML Deploying model as API IBM
DVC Data versioning and pipelines —
CML CI/CD for machine learning —
IBM Watson OpenScale Continuous monitoring IBM
Terraform Environment setup —
Cookiecutter Starting a new project —

The guide notes that it is modular, so a team can swap tools at any point based on project preferences and use cases. The tool table above reflects only what the guide lists; it is not a claim that these are the only viable options.

Who this is for

Use the MLOps Guide if you want both the conceptual framing of MLOps and a concrete end-to-end example to follow. The guide includes a video showing an example workflow with a complete MLOps project, and states that the exact project is available as an end-to-end example built for the guide. If you only need the definition, the five features above are the short answer. If you need to build the environment, start with the Implementation Guide and Environment sections.