Engineering MLOps // Emmanuel Raj // MLOps Meetup #69
MLOps community meetup #69! Last Wednesday we talked to Emmanuel Raj, Senior Machine Learning Engineer at TietoEvry.
//Abstract
The talk focuses on simplifying/demystifying MLOps and encourages others to take steps to learn this powerful SE method. We also talked about Emmanuel's journey in ML engineering, the evolution of MLOps, daily life, and SE problems, and what's next in MLOps (fusion of AIOps, EU AI regulations impact on MLOps workflow, etc).
//Bio
Emmanuel Raj is a Finland-based Senior Machine Learning Engineer. He is a passionate ML Researcher, Software engineer, speaker, and author. He is also a Machine Learning Engineer at TietoEvry and a Researcher at Arcada University of Applied Sciences in Finland. With over 6+ years of experience building ML solutions in the industry, he has worked on multiple domains such as Healthcare, Manufacturing, Finance, Retail, e-commerce, aviation, etc.
Emmanuel is passionate about democratizing AI and bringing state-of-the-art research to the industry. He has a keen interest in R&D in technologies such as Edge AI, Blockchain, NLP, MLOps, and Robotics. He believes the best way to learn is to teach and he is passionate about teaching about new technologies, that's one reason for writing a book and making an online course on MLOps.
Emmanuel is the author of the book "Engineering MLOps". The book covers industry best case practices and hands-on implementation to Rapidly build, test, and manage production-ready machine learning life cycles at scale. There is a big evolution happening in Data science for good, and we are moving away from notebooks and models sharing to a collaborative way of working via MLOps. We will discuss this big evolution of DevOps, MLOps, Data Engineering, Data Science, and Data-Driven business in the meetup.
// Other Links:
www.emmanuelraj.com
• Emmanuel Raj - MLOps: Automated Machine Le...
• Emmanuel Raj - Engineering MLOps - 5/26/2021
-------------- ✌️Connect With Us ✌️ ------------
Join our slack community: https://go.mlops.community/slack
Follow us on Twitter: @mlopscommunity
Sign up for the next meetup: https://go.mlops.community/register
Catch all episodes, Feature Store, Machine Learning Monitoring and Blogs: https://mlops.community/
Connect with Demetrios on LinkedIn: / dpbrinkm
Connect with Emmanuel on LinkedIn: / emmanuelraj7
// Link to live coding:
https://github.com/emmanuelraj7/mlops...
// Book
https://www.oreilly.com/library/view/...
https://www.amazon.com/Engineering-ML...
Timestamps:
[00:00] Introduction to Emmanuel Raj
[03:37] Emmanuel's background in tech
[04:18] "Software doesn't have to be deterministic. It can be without rules."
[05:48] Machine Learning failures
1. Lack of systematic way of developing AI software
2. Not focused on continuous learning and retraining models
3. Wrong strategy
4. Lack of strong governance
5. Lack of testing for the real world
[09:07] "One thing is to test with unit test and do integration test before going into production but also important to do testing for the real-world scenarios."
[10:10] Onyx solving the problem and Jupyter notebooks not ready for production
[12:13] "It's better to go as lean as simple as possible in the form of scripts."
[12:20] Building robust CI/CD pipelines
[14:58] Monitoring each phase of the Machine Learning life cycle
[15:48] Developing of AIOps
[16:20] AIOps in CI/CD pipeline from alert hells
[19:33] Where to start building MLOps capabilities
Phase 1: Build ML models using ML models
Phase 2: Deployment
Phase 3: Monitor
[22:00] Company legacy
[24:47] "Don't fall in love with tools, fall in love with the most optimal way to solve a problem."
[26:47] Open-source tools
[29:08] Security as a roadblock
[31:00] What comes next for MLOps?
[32:08] Most mature section of MLOps
3 Blocks:
1. Build Models
2. Deploy Models
3. Monitor Models
[34:40] Live Coding: Important aspects of MLFlow
[38:00] FastAPI microservice
[40:28] FastAPI terminal
[40:40] Local testing using Locust
[40:56] Run image and spin container
[41:10] Predict end-point and make a post request
[41:54] Load testing
[43:47] Spin up Locust in the IO server
[44:38] Mention your end-point
[45:20] Start Locust
[47:40] Push to production
[48:15] Automate
[50:00] Engineering MLOps is out!