You must have noticed the buzz about MLOps.
MLOps is a set of processes, tools, and best practices for managing machine learning lifecycle and deploying ML models in production.
There has been an explosion of MLOps vendors and tools. Many of those are named as xyzFlow or xyzML:
These were just a few examples. Here is the relevant part of Matt Turck’s 2021 ML/Data tool landscape.
As if all that was not crazy enough…
You may think that these are small vendors or startups, what about the Big Three?
There is a remarkable similarity in the traditional software DevOps offerings from AWS, Azure, and Google Cloud. We can use their MLOps offerings as a template and make some sense out of it all.
So I looked and Amazon SageMaker and Google VertexAI, and there are some similarities in the tooling. But the MLOps worldview of the Big Three has not yet converged. It is apparent from their MLOps Lifecycle and ML Maturity Levels descriptions.
Amazon
Source: AWS whitepaper – MLOps: Emerging Trends in Data, Code, and Infrastructure
Microsoft
Source: Azure whitepaper – MLOps with Azure Machine Learning
The dust has not settled yet…
This is how the desktop software development world was back in the 1980s and the cloud development world in the 2000s. There were several competing methodologies, processes, and tools, and slowly coherence emerged. The same will happen for ML in 2–3 years. And then, everything about MLOps will be so obvious. Quite like how a caterpillar turns into a butterfly.
