03·The ML project workflow — from problem to deployment

Introduction

A professional ML project is not "train a model and ship it" but a cycle: problem framing → data collection → EDA and cleaning → feature engineering → model selection → training and tuning → hold-out evaluation → deployment → monitoring and retraining. The lesson clarifies the stages, identifies common pitfalls (data leakage, wrong baseline, model-centric instead of data-centric), and introduces MLOps concepts: pipeline, feature store, model registry, drift, shadow deployment, A/B test.