subtopic

Build the preprocessing Pipeline (ColumnTransformer)

Package scaling and encoding into one ColumnTransformer + model Pipeline and verify it reproduces the manual result. Concepts: Pipeline; ColumnTransformer; reproducibility; serving consistency Project A: Your notebook's pipeline used only raw numerics + merged categoricals, while the manual path used engineered and ordinal features → results are not comparable. Put the engineered features into the pipeline too. Project B: Pipeline with calendar, lag and weather features + linear regression baseline.