Time Series & Many-Series Forecasting

Theory

Cover time-series fundamentals: trend, seasonality, residuals, stationarity. Discuss classical baseline models (na√Øve last-value, moving averages, simple exponential smoothing) and appropriate error metrics (RMSE, MAE, MAPE). Emphasize the “many-series” setting in retail (thousands of product √ó store series) and what breaks from single-series methods: memory/compute constraints, the need for shared models or hierarchical structures, and proper time-based evaluation (rolling or expanding windows). Highlight how naive random splits cause leakage in time-series and why backtesting protocols are needed at scale. Include quick exercises computing simple baselines and errors on short synthetic series.

Technical

Technical focus on exploratory data analysis (EDA) and feature engineering with Spark. Show patterns for EDA at scale: stratified sampling, approximate quantiles, grouped statistics, and visualization via sampled extracts. For each project type, demonstrate typical feature engineering: interaction counts and recency for recommenders, degree distributions and edge weights for graphs, lagged values and calendar features for forecasting. In-class: teams extend their ETL to produce initial feature-ready tables and perform basic EDA relevant to their project (e.g., item popularity histograms, graph degree histograms, seasonality plots from aggregated sales).