Machine Learning on Relational Databases with TabPFN-Rel

Most business data lives across many linked tables, but most ML tools expect one flat table. In this webinar, we’ll show how TabPFN-Rel makes predictions directly on relational databases, without hand-built joins or feature engineering. You’ll then build a churn model yourself in a hands-on session.

What we’ll cover:

• What relational machine learning is
• RelArena - the new benchmark for relational ML and why we need it
• How TabPFN-Rel works: following the links between tables, building features from each entity’s history and predicting without task-specific training
• How it prevents data leakage by only using data from before each prediction date
• Defining your database and prediction task in YAML and SQL, and using a coding agent to write them
• How TabPFN-Rel performs on RelArena, our open benchmark for relational learning
• Hands-on: predicting seller churn on the Olist e-commerce database with the cookbook in Colab
• Live Q&A