Foundations and preprocessing pipelines
Data cleaning, feature engineering, and scikit-learn pipelines
Foundations and Preprocessing Pipelines
Preprocessing is part of the model, not a separate cleanup step. This lecture covers the key preprocessing tasks (imputation, encoding, scaling), why the split-first rule prevents data leakage, and how scikit-learn’s Pipeline and ColumnTransformer keep the workflow reproducible. We also discuss how different models require different preprocessing — tree-based methods vs distance-based methods vs linear models.
Extra: working with agents
A 20-minute introduction to handing whole pieces of work to an AI agent: what makes something an agent, the kinds of agentic setups beyond coding, how to steer one with an instructions file, plans and permissions, and where agents help — and quietly mislead — when you build predictive models. Followed by a live demo.