What is MLOps & why do we need it?

So, what is MLOps? The simplest explanation could be the DevOps principles and practices to the machine learning workflow. It’s a process to streamline the ML development and deployment. The goal is simple. <ul> <li>Faster experimentation and model development.</li> <li>Faster deployment of updated models and to production.</li> <li>Quality assurance.</li> </ul> Let’s describe the standard machine learning development and deployment flow. Typically, it starts with the data, which can be in a variety of formats: CSVs, JSON, Video, Image, etc. The first step is to clean up the data, because the quality of the data is highly correlated with the efficiency of the model. Better data, better model. But, to increase the quality of the data you need to have more data, and that’s where the first need of MLOps comes. The model is a final output, but it needs to be updated constantly. And that’s the second why of the MLOps. You need to optimize and streamline the process of building a model, which can have a bunch of subprocesses like: <ul> <li>Feature selection/generation.</li> <li>Algorithm selection.</li> <li>Hyperparameter tuning.</li> <li>Fitting the model.</li> <li>etc.</li> </ul> which we can call an experimentation to get the “best” model for the application and business. <a href="https://medium.com/@dmosyan/what-is-mlops-why-do-we-need-it-171e636a710e">Website</a>