Feast GCP Driver Ranking Tutorial is a tutorial project that demonstrates building and serving a machine learning model using Feast, a feature management platform, with Google Cloud Platform services.
The tutorial addresses the challenge of managing features consistently across model training and serving environments. It walks through building a linear regression model that predicts driver trip completion using Feast to orchestrate feature engineering. The approach progresses from local development to production-grade setup, showing how to ingest features into Feast, train a model with data from BigQuery, and serve predictions through different online stores.
The tutorial suits developers learning Feast who have access to GCP resources and want to understand the full lifecycle from local experimentation to cloud deployment. It demonstrates both a basic local mode for quick iteration and an advanced mode using Firestore for production scenarios. The progression from SQLite to Firestore online stores illustrates how to move from development to production infrastructure, and the optional use of Google Cloud Storage for feature registry sharing shows how multiple systems can coordinate on feature definitions.
The project is structured as a step-by-step guide with numbered sections covering installation, local feature store setup, model training, and progressive testing against different online stores. It includes configuration examples showing how to switch the provider field from local to GCP and demonstrates the code changes needed when moving from development to production infrastructure.
