To MLOps, or not to MLOps? That is the question — the platform is the answer

🥛Rehydrate, and proceed with the article 🤠 Photo by Bram Naus on Unsplash

Out of all AI-centric buzzwords (e.g. #BigData, #DataInBusiness…you get the drift 😏) #MLOps has been creating a lot of traction in the last three years. This is evident in the worldwide interest for MLOps as seen in Google Trends, as shown in the plot below, where numbers (0 to 100) represent search interest relative to the highest point on the chart for the given time.

Note: The dip and rise on 1 January 2022 are due to a change in the Google Trends collection. Plot created using https://trends.google.com

Now, MLOps can be defined as “the extension of the DevOps methodology to include Machine Learning and Data Science assets as first-class citizens within the DevOps ecology” (source: the MLOps Roadmap 2022). In other words, MLOps extends the DevOps practice of continuously building, testing, and deploying code (DevOps), to data (Data Engineering) and models (Machine Learning), as represented below.

MLOps is supported by the triumvirate of Machine Learning, DevOps and Data Engineering. If you are a fan of Venn diagrams, or want to visualise the difference between Data Science and Machine Learning, head to https://www.carleton.edu/its/blog/what-is-data-science/. Plot created using Canva.

MLOps stands for Machine Learning (ML) Operations, and it is the “operations” part that is the key, as it aims to promote automation, scalability and reliability in the ML journey from…