I had some time over the end of year / new year to go over some machine learning concepts and cloud technologies. I have used cloud computing here and there - whether it was to spin up a Minecraft server to play with friends, develop a webpage to help with tutoring, or to develop AI applications for work - but I wanted to get an overview of what the three cloud providers had to offer. I also got to fill in some of my many gaps in my knowledge in this vast field of data science and machine learning. Overall, it was a good learning experience. As a side effect, I now have some new badges.
While there were some differences in how the three major cloud provides approach machine learning, their certifications focused on similar ideas: (a) taking models into production with scalability and reliability, (b) how to tackle issues such as privacy, bias, and explainability, and (c) how and when to leverage existing AI/ML solutions (cloud specific or through AutoML). In short, MLOps seemed to be a common theme amongst the three.
MLOps is said to be a combination of machine learning, devops, and data engineering. So it can be very challenging to learn. So here are some recommendations.
These certifications are aimed at professionals who already have some knowledge and experience in machine learning. Maybe you are a software engineer who is increasingly having to incorporate ML solutions. Perhaps you are a statistician dipping your toes in the cloud. Or you are a manager / consultant working in the AI field.
In any case, if you are new to programming and/or machine learning - you will need to start there.
📖 An Introduction to Statistical Learning by James, Witten, Hastie, and Tibshirani. It is never a bad idea to re-read this book for the "classical" machine learning models.
📖 Deep Learning with Python by Francois Chollet is another great introductory book, which covers deep learning models as well as working with images and text.
📖 Machine Learning Engineering by Andriy Burkov is a good book that touches on many of the MLOps topics that all machine learning engineers should know.
📖 Fundamentals of Data Engineering by Reis and Housley provides a very good overview of all things data engineering in a vendor/product-independent way.
🎓 Machine Learning DevOps Engineer Nanodegree by Udacity covered a lot of ground (including tools such as MLflow and FastAPI, and concepts such as writing clean code and automated testing). Despite its cost, Udacity is also a great platform. Their nanodegrees take a project-based learning approach. Very hands-on and a very industry-informed curation of projects. You won't pass unless your project meets all criteria - so the nanodegrees can be quite time consuming, but their feedback system is amazing, and you can resubmit your projects as many times as you need to keep improving.
🎓 Cloud Academy was a great learning platform - especially with the hands-on labs where they provide you temporary access to AWS, Azure, and Google Cloud, so that you can play around with the services you are learning about, rather than just reading about them. They have learning paths (which include video lessons, hands-on labs, quizzes, and practice exams) for the data engineering and machine learning certifications (and plenty more) on all three major cloud platforms.
Avoid using "brain dumps" (where people try to remember exam questions and share them). Not only is it cheating, but there is no quality control - the questions would be worded incorrectly, or the "answers" will be wrong almost all of the time. The "best" or "official" practice exams are:
- Microsoft Azure: Measure Up. Amazing quality, basically the same coverage, style, and difficulty as the real thing.
- AWS: Tutorials Dojo. Good explanations, and similar level of difficulty as the real thing.
- Google Cloud: This one doesn't have an "official" partner. But the closest thing I could find was Whizlabs. Varying quality from question to question, but reasonable quality. I can confirm that these were not brain dumps. And I can confirm that the questions were mostly useful to study.