Understanding MLOps
MLOps is an emerging domain that takes advantage of the maturity of existing software development processes—in other words, DevOps combined with data engineering and ML disciplines. MLOps can be simplified as an engineering practice of applying DevOps to ML projects. Let's take a closer look at how these disciplines form the foundation of MLOps.
ML
ML projects involve activities that are not present in traditional programming. You learned in Figure 2.3 that the bulk of the work in ML projects is not model development. Rather, it is more data gathering and processing, data analysis, feature engineering (FE), process management, data analysis, model serving, and more. In fact, according to the paper Hidden Technical Debt in Machine Learning Systems by D. Sculley et al., only 5% of the work is ML model development. Because of this, MLOps is not only focused on the ML model development task but mostly on the big picture—the entire ML project...
