MLOps in Azure using Python SDK — Part 1

Introduction to MLOps using AzureML SDK

Source : <a href=’https://www.freepik.com/vectors/design'>Design vector created by macrovector — www.freepik.com</a>

Taking a Machine Learning project to production involves multiple components — Data Engineering, DevOps, and Machine Learning. The intersection of these components is MLOps. MLOps (Machine Learning + DevOps) is the process of taking a machine learning project to production — with the goal of automating and improving the quality of production models, while also focusing on business and regulatory requirements.

Key Phases in MLOps

1. Data Gathering
2. Data Analysis
3. Data Transformation/Cleaning
4. Model Training & Development
5. Model Validation
6. Deploying the Model
7. Serving the Model as a web app or an API
8. Model re-training

MLOps Pipeline (Image courtesy: Microsoft Azure Docs)

Another key feature in MLOps is that we will also want to track metrics across experiments — how did the model perform for example needs to be tracked across various hyperparameters so that we can choose the best model.