MLOps¶
Training a model is a small fraction of the work of running one. A notebook that reaches 94% accuracy on a held-out split is a research result; MLOps is the discipline that turns that result into a system which keeps being correct, for months, while the world underneath it keeps changing, while ten other engineers keep touching the code, and while a regulator or a customer might ask "why did the model say that, on this exact input, on this exact day?" This chapter builds the full lifecycle: reproducibility (tying together DVC, chapter 14, and MLflow, chapter 15), CI/CD adapted to data and models (extending chapter 8), a mathematically guarded promotion process so a new model can never silently make things worse, serving patterns (online and batch), the feature store pattern that prevents the single most common production bug in ML systems, and — with full derivations and worked numeric examples — the statistics of monitoring for drift: PSI, the Kolmogorov–Smirnov test, and the hypothesis test that should gate every model promotion. …
🔒 La suite est en accès freemium — lecture complète 100 % gratuite
Tu lis ici l'aperçu libre. Le reste du chapitre (code, schémas, maths, exercices) fait partie du livre complet : crée un compte gratuit (30 secondes, aucun paiement) pour tout lire.