Covers: theory of MLOps
Estimated time needed to finish: 30 minutes
Questions this item addresses:
  • Motivation for MLOps?
  • Designing ML-powered Software?
  • End-to-End ML Workflow Lifecycle?
  • Three Levels of ML-based Software?
  • MLOps Principles?
  • State of MLOps (Tools & Frameworks)?
How to use this item?

There are in total 6 parts, but the following two are more interested to me:

  • End-to-End ML Workflow Lifecycle. This article gives you the overall implementation components of a end-to-end machien learning workflow. It introduces:

    • What is data engineering and how this fit into the workflow
    • What is model engineering and how this fit into the workflow
    • what is model deployment and how this fit into the workflow.
  • MLOps Principles. Only knowing the components are not enough. In order to follow the best practice of MLOps, this article illustrates what are the principles we should follow so that we can build a robust, reliable ML product with less possibily of useless effort, including:

    • Automation
    • Continuous X (continuous integration, Continuous delivery, continuous training and continuousmonitoring)
    • Versioning
    • Experiments Tracking
    • Testing
    • Monitoring
    • "ML Test Score" System
    • Reproducibility
    • Loosely Coupled Architecture (Modularity)
    • ML-based Software Delivery Metrics (4 metrics from “Accelerate”)
Author(s) / creator(s) / reference(s)
INNOQ: Dr. Larysa Visengeriyeva, Anja Kammer, Isabel Bär, Alexander Kniesz, and Michael Plöd
0 comment
Recipe
publicShare
Star0

What Is MLOps?

Contributors
Total time needed: ~3 hours
Objectives
For me, a successful ML app is not only about the accuracy of the Model, but also the infrastructure around the model to make the model a scalable, stable application. This is where MLOps becomes important, to apply DevOps best practices into ML project.
Potential Use Cases
For the ones that are interested in MLOps or want to make your ML product more robust.
Who is This For ?
BEGINNERPeople interested in MLOps
Click on each of the following annotated items to see details.
ARTICLE 1. What Is MLOps?
  • What is MLOps?
20 minutes
ARTICLE 2. MLOps: Continuous delivery and automation pipelines in machine learning
  • How to implement and automate CI, CD, CT for ML systems?
  • What are the differences between DevOps and MLOps?
  • What are the data science steps for ML?
  • What are the 3 maturity level of MLOps?
40 minutes
REPO 3. MLOps on Azure
  • How does Azure ML help with MLOps?
  • What are MLOps Best Practices?
  • What are MLOps Solutions using Azure?
  • What are the key challenges Azure with to solve with MLOps?
60 minutes
ARTICLE 4. Machine Learning Operations
  • Motivation for MLOps?
  • Designing ML-powered Software?
  • End-to-End ML Workflow Lifecycle?
  • Three Levels of ML-based Software?
  • MLOps Principles?
  • State of MLOps (Tools & Frameworks)?
30 minutes

Concepts Covered

0 comment