Learning MLOps (Gradually) for Free Through Blog Posts & Podcasts

Published in

Ml Ops by Mikiko Bazeley

6 min readSep 27, 2022

Introduction

It’s 2022 and you’re interested in learning about MLOps (and maybe even pivoting to an MLOps Engineer role).

There’s a ton of content out there on every topic you can imagine, from data science best practices, model algorithms, software engineering principles, system designs etc.

And maybe that’s one of the problems – there’s so much content out there, it feels overwhelming & panic inducing, especially if you don’t know what questions to ask.

My goal is to highlight some of the best free talks, papers, blog posts, and communities out there that are focused on doing MLOps and production ML.

In a separate post I’ll dive deeper into how to combine the links below with my favorite books, courses, and subscriptions to create a personalized roadmap to help develop and grow your MLOps career.

Criteria

How I chose the recommendations:

I have listened to them & enjoy them ❤️;
They resonate with my experiences (or the experiences of people I know) 👂;
They are less than an hour to listen to (on 2X mode) or are less than a couple hours of skimming & light reading ⏲;
They are free – not freemium, not hidden behind a paywall… Just straight up free. 💰

With that being said some of the truly BEST content out there is in the form of books, courses, or paid subscriptions (which I’ll tackle in a different blog post).

MLOPs 101

What is MLOps?

Machine learning operations has many definitions but for the most part, agreement is that it’s a practice that involves establishing processes & tooling (whether bought or built or Frankensteined) in order to help models go from exploratory MVPs to productionionized models that can be used in web and mobile applications, devices, or via API.

MLOps is the business of shipping many models and pipelines as effectively as possible and getting them through the ML lifecycle.

A fun video by my friend Ken Jee illustrates his take on MLOps using papayas. I also attempt to tackle different topics on MLOps in less than 60 seconds with my MLOps Tool Snacks & MLOps Terminology Snacks.

Are you the type that loves flashcard style studying?

Then look not further than MLOps Wiki | AI Observability | Censius which tries to define some of the most common terminology you’ll come across in MLOps.

A website I especially love recommending for people trying to wrap their heads around all the different component of an MLOps system Ml-ops.org that has useful templates and framing around principles & stacks.

If you’re interested in further understanding how different MLOps related teams and work streams interact in an enterprise setting & you feel like you have a handle on what MLOps is, Machine Learning Operations (MLOps): Overview, Definition, and Architecture covers those questions and more.

But as Patrick Wendell from Databricks notes, the fact that the diagram is representative of many enterprise (and non-enterprise) companies out there might be part of the problem.

Where did MLOps Come From?

Advances in technology don’t come out of thin air.

As a new engineer or entrant into the field of MLOps, it’s easy to point fingers and question the disarray that is ML systems or the MLOps ecosystem.

But history and context are important in every field of engineering and MLOps is no different.

For example, some fun facts I summarized in this thread here (note: Years are rough approx, depending on sources I used):

~ 2008: Github Site Launched
~ 2009: DevOps term coined (Allspaw & Hammond)
~ 2010’s – Big Data Era with 3 V’s Popularized
~ 2013 – HBR Data Scientist Article Published, Initial Release of Anaconda Distribution
~ 2013 – Docker Introduced
~ 2014 – Machine Learning: The High-Interest Credit Card of Technical Debt Paper
~ 2015 – Initial Release of Tensorflow, Kubernetes 1.0, Hidden Technical Debt Paper
~ 2016 – PyTorch Initial Release
~ 2017 – ML Test Paper published
~ 2018 – First Stable Release of Jupyter Notebooks

and so on.

You get the picture. There were waves, stages, and stories created throughout the years that set the stage for the tools and technologies developing in the here and now.

To understand where MLOps came from, I highly recommend checking out these talks on The Evolution of DevOps and the Birth of MLOps with Sam Ramji, The Godfather Of MLOps // D. Sculley // MLOps Coffee Sessions #32 with D. Sculley, the early days of Tensorflow development with Jeff Dean Interview - Systems and Software for Machine Learning at Scale and even the early days of Jupyter & The Evolution of ML Tooling with Brian Granger.

Does MLOps Have Value?

That’s a hot-button question if there ever was one.

On the one hand, Lakshmanan of Google fame would hook you into saying “no” with his piece “No, you don't need MLOps” (although some folks in my network have argued the title is really a bait & switch and should be “Yeah, it’s probably good have in small doses”).

Let’s assume that you’ve bought into the benefits of implementing MLOps in your team or organization. The folks in this talk Machine Learning Engineering for Production (MLOps) describe the value of MLOps for teams and orgs at any stage.

Industrial MLOps

What does an MLOps engineer do?

For some useful comparisons and contrasts:

MLOps vs Data Engineering – What Does it Really Mean To Do MLOps & What Is The Data Engineer's Role? with Demetrios Brinkmann & David Aponte.
ML Engineering & Productionizing ML: Reflecting on a Year of Making Machine Learning Actually Useful | Shreya Shankar & Thoughts on ML Engineering After a Year of my PhD | Shreya Shankar from Shreya Shankar.

I also talk about my experiences as an MLOps Engineer in this conversation with Ken Jee. A wealth of user stories and experiences however can be found at the MLOps Community Youtube Channel.

Who is doing MLOps & How?

Noah Gift (one of the co-authors of “Practical MLOps” along with Alfredo Deza) starts describing some of the principles and practices that MLOps involve in this conversation Practical MLOps // Noah Gift // MLOps Coffee Sessions #27.

These principles and practices are further fleshed out in Practitioners guide to MLOps: A framework for continuous delivery and automation of machine learning.

Some of the most popular conferences that cover MLOps use cases and system designs at various companies doing MLOps include the Toronto Machine Learning Summit, MLOps World, TWIML Con, and REWORK.

James Le captures detailed notes on each of these conferences, including the presentations by companies like Spotify, Zalando, & even Mailchimp!

Conclusion

When I began my journey of trying to make ML useful and productionize models years ago I remember feeling overwhelmed by the swamp of content out there.

Do I get started on reading papers? Sign up for a course? Which course to sign up for?

If you are serious about pivoting into MLOps and come from a non-engineering or non-adjacent engineering background, there will come a time when the next best learning step is picking a couple of the awesome courses and books that are now out on the market.

If you’re just starting your MLOps career or you’re still not sure whether to take the plunge and sign up for a course, I hope you know there are other high-quality free options, like the ones linked in the post above.

Let me know if you think there are any talks or blog posts that deserve to be on the list!

In a future blog post, I’ll start digging into how to select the right books and courses so you can create your own MLOps Engineer career roadmap.

Interesting in Staying Updated?

Thanks for the read! My name is Mikiko & I work as an MlOps Engineer.

If you’re interested in learning more about MLOps, production ML, and distributed systems + cloud development, I also publish:

Videos on Youtube at The MLOps Engineer
Technical walkthroughs on Substack
Snark and saltiness on Twitter
Professional sounding/career oriented stuff on LinkedIn
Code gists and TODOs on Github

And I take support and patronage in the form of coffees ☕!