Machine Learning Pipelines and MLOps with LightGBM
This chapter shifts the focus from data science and modeling problems to building production services for our ML solutions. We introduce the concept of machine learning pipelines, a systematic approach to processing data, and building models that ensure consistency and correctness.
We also introduce the concept of MLOps, a practice that blends DevOps and ML and addresses the need to deploy and maintain production-capable ML systems.
The chapter includes an example of building an ML pipeline using scikit-learn, encapsulating data processing, model building, and tuning. We show how to wrap the pipeline in a web API, exposing a secure endpoint for prediction. Finally, we also look at the containerization of the system and deployment to Google Cloud.
The main topics of this chapter are as follows:
- Machine learning pipelines
- An overview of MLOps
- Deploying an ML pipeline for customer churn
