Classical Models That Still Deliver

Trends come and go, but a lot of production value still comes from “classic” ML: recommenders that retrieve the right candidates fast, time series that forecast inventories and traffic, tabular models that crush structured data, and anomaly detectors that raise the right alarms. This chapter focuses on practical, Python-first patterns: a two-tower recommender, forecasting with PyTorch Forecasting, a hybrid tabular pipeline that mixes Torch embeddings with LightGBM, and a simple, dependable anomaly detector.