MLOps Where Do I Even Start?. As I write this, I find myself asking… | by Lehlohonolo Makoti | May, 2024

As I write this, I find myself asking the same question and many more…

1. Where the *bleep* do I start munching down on this crazy assorted plate?
2. How do I strike a balance between starting to learn what’s practical (healthy) and mostly theoretical (not so healthy)? — Carbs are good, but of course, in moderation, and veges are definitely good.
3. Should I even be trying to get my fingers (cutlery aside) on everything? — what happened to the good old days of “sharing is caring” 😅

Well, venting aside, now it’s time to break it down, I hope by the end you would’ve picked up some practical next steps for your own situation.

Most of what is MLOps is loaned from DevOps, which is another whole beast, even Hercules would have a tough day at work dealing with.

The grand idea put forth in tech-meetups lingo is “Let’s operationalise model development”. Simply put; with MLOps we want to setup a system which defines a repeatable process for building and deploying Machine Learning models. This includes automating and streamlining stages in model creation, from data preparation, model training & validation + testing all the way to model deployment + monitoring.

The part that always gets left out is the PLANNING which I would argue forms the core of the entire system, MLOps at the end of the day just like any tech tool, should create business value. We need to always `start with the end in mind` when we take-off on the MLOps journey, the resourcing and budgeting for this effort can get very much out of hand, if not kept on a tight leash, and with all projects we need to ensure that the cure is not worse than the ailment:

Ailment: Seasonal Cold > Cure: Miracle Formula > Side Effect: Death

The paper (Hidden Technical Debt in Machine Learning Systems) that prompted the conversation that gave birth to MLOps addresses this very issue, ensuring that solutions are fit for purpose and we lessen the technical debt that piles up when we fiddle with ML systems.

Key Concepts in MLOps

• Efficiency: Streamlining processes to reduce time and resource expenditure.
• Consistency: Ensuring uniformity in model development and deployment.
• Scalability: Building systems that can grow with increasing data and model complexity.
• Repeatability: Creating processes that can be reliably repeated to produce similar results.

Keywords: Efficient | Consistent | Scalable | Repeatable

Over the years of my life as a data professional, I still appreciate to this day the time I took to learn and drill into my brain basics concepts like data modelling so I’ve chosen to start ground up on the introduction and I hope you will stay with me as I go through the hills and valleys of MLOps.

First Read: Introducing MLOps by Mark Treveil & The Dataiku Team. I will cover overarching chapter summaries and add some of my personalised seasoning which might help flavor the lessons to your liking.

• Part I. MLOps: What and Why (14 June 2024)
• Part II. MLOps: How (28 June 2024)
• Part III. MLOps: Real-World Examples (12 July 2024)

This structured approach ensures a thorough understanding of MLOps, from basic principles to real-world applications.

GitHub: https://github.com/lmakoti
Roadmap: https://roadmap.sh/u/lmakoti
LinkedIn: https://www.linkedin.com/in/lehlohonolomakoti/

1. Sweenor David, Hillion Steven, Rope Dan, Kannabiran Dev, Hill Thomas,O’Connell Michael, and Safari an O’Reilly Media Company. 2020. ML Ops :Operationalizing Data Science. https://go.oreilly.com/queensland-university-of-technology/library/view/-/9781492074663/?ar
2. Sculley, D., Holt, G., Golovin, D., Davydov, E., Phillips, T., Ebner, D., Chaudhary, V., Young, M., Crespo, J.-F., & Dennison, D. (2015). Hidden Technical Debt in Machine Learning Systems. In Advances in Neural Information Processing Systems 28 (NIPS 2015). Retrieved from https://papers.nips.cc/paper/2015/file/86df7dcfd896fcaf2674f757a2463eba-Paper.pdf