MLOps Notes 3.1: An Overview of Modeling for machine learning projects.
Welcome back, everyone! This is Akhil Theerthala. In the last article we have explored the standard practices and challenges faced during the deployment phase of the Machine Learning lifecycle. Now we take one more step back and visit the Modelling stage.
This part focuses on the critical challenges faced in developing machine learning models, like dealing with skewed datasets or the interesting case where the model performs worse after deployment even though it has very good test scores and many other problems.
Note: Since this part is a bit long, I have decided to split it into 3 different articles. One describes the details of developing the model, the other describes the standard procedures used for error analysis, and in the last one, we will look into more details about the data-driven approach for modeling.
Model-centric Approach vs. Data-Centric Approach
Until recently, there has been a lot of emphasis on choosing a suitable model for the project on a fixed benchmarked dataset, as most of the machine learning community was focused solely on research. This perspective is called Model-centric AI development. However, in recent days, we see a shift to another approach called Data-centric AI development, where…