MLOps Guide

This site is intended to be a MLOps Guide to help projects and companies to build more reliable MLOps environment. This guide should contemplate the theory behind MLOps and an implementation that should fit for most use cases. Made by Arthur Olga, Gabriel Monteiro, Guilherme Leite and Vinicius Lima

What is MLOps?

MLOps is a methodology of operation that aims to facilitate the process of bringing an experimental Machine Learning model into production and maintaining it efficiently. MLOps focus on bringing the methodology of DevOps used in the software industry to the Machine Learning model lifecycle. In that way we can define some of the main features of a MLOPs project:

Data and Model Versioning
Feature Management and Storing
Automation of Pipelines and Processes
CI/CD for Machine Learning
Continuous Monitoring of Models

What is Contemplated on This Guide?

Introduction to MLOps Concepts
Tutorial for Building a MLOps Environment

MLOps Environment

This video shows how an example of workflow with a complete MLOps project. This exact project can be found here and is an example end-to-end made for this guide.

Architecture

The following diagram shows the complete MLOps flow used on the tutorial. Since the guide is modular, a team can choose to swap tools at any point due to project preferences and use cases.

Diagram

Project Tools

The main tools discussed in the guide are shown in the following table.

Tools	Function	Developer	License
IBM Watson ML	Deploying model as API	IBM	Proprietary
IBM Watson OpenScale	Monitoring Model in production	IBM	Proprietary
DVC	Data and Model Versioning	Iterative	Apache License 2.0
CML	Pipeline Automation	Iterative	Apache License 2.0
Terraform	Setups IBM infrastructure with script	HashiCorp	Mozilla Public License v2.0
	Code versioning		Proprietary
Actions	CI/CD Automation		Proprietary
Pytest	Python script testing	Pytest-dev	MIT
Pre-commit	Running tests on local commit	Pre-commit	MIT
Cookiecutter	Creating folder structure and files	Cookiecutter	BSD 3-Clause