Regularized-Regression Case (Housing/Price)

# Regularized-Regression Case (Housing/Price) Target: fit ridge/lasso on housing prices and explain the shrinkage effect in one paragraph a stakeholder can follow. ## 1. Baseline first Fit plain OLS on the train split and record holdout RMSE. This number is the anchor: regularization must beat it or you keep OLS. No anchor, no claim. ## 2. Ridge vs lasso, decided by the goal Ridge (L2) shrinks all coefficients smoothly — keeps every feature, tames collinearity. Lasso (L1) drives weak coefficients to exactly zero — automatic selection, at the cost of arbitrarily dropping one of two correlated features. Choose by reading the coefficient paths across alphas, not by fashion. ## 3. Trap: the leakage column The starter includes a column derived from the sale paperwork (a post-outcome proxy). Train RMSE looks miraculous; holdout collapses. Diagnose by feature-ablation: drop each column, re-score, and name the one whose removal destroys the miracle. ## Recap OLS anchor, ridge for shrinkage / lasso for selection via alpha paths, and a leakage ablation that names the proxy column before you trust any RMSE.