Machine Learning on Google Cloud Workbook
Machine Learning has been a big driver of business value for companies that have been successful in its adoption. Across industries, the key challenge is the process towards a successful adoption. Despite growing demand and democratization of the tools and techniques available, many companies still face significant challenges in getting models out of their local environments and deploying them into a robust production system.
The goal of this white paper is to cover all topics to consider when developing and deploying production machine learning systems on Google Cloud Platform (GCP). To remain succinct, it will not dive into the details of data warehousing on GCP as this is a big topic on its own. The paper will start with a brief introduction to Google Cloud as it is fundamental to understanding later sections. Afterwards, it will cover the basics of MLOps while being technology agnostic. The next section will cover the different ML services available on Google Cloud more in detail and the last section will then explain how those services can be used to implement the MLOps principles described.
Reading this white paper, written by our cloud engineers, will help orient you in getting started in the world of Machine Learning and MLOps on Google Cloud.
What’s inside?
The basics of Google Cloud
Introduction to Cloud resource hierarchy, IAM, services, locations, regions, and zones.
The basics of MLOps
Principles of MLOps, deployment pipelines, version control, CI/CD, feature store, and environment management.
Machine Learning services on Google Cloud
Overview of core services used in ML workflows:
- Cloud Storage
- BigQuery
- Vertex AI (including Training, Pipelines, Metadata Store, Model Registry, Feature Store, Workbench)
Applying MLOps on Google Cloud
End-to-end MLOps workflow:
- Data exploration and experimentation
- Using Feature Store
- Building pipelines (components and orchestration)
- Model training (AutoML, custom training, hyperparameter tuning)
- Model hosting (Vertex AI Endpoints, Cloud Run, etc.)
Conclusion & References
Summary of how MLOps and Google Cloud services combine to build scalable, reliable, and automated ML systems.
Download our ebook
