model-evaluationlisted

## Purpose Ensure models are evaluated rigorously before deployment decisions are made. ## When to use it After training, before declaring a model ready for deployment review. ## Checklist - **Split strategy**: holdout for large datasets, k-fold or stratified for smaller ones. - **Metric selection**: classification (precision/recall/F1/AUC-ROC/AUC-PR), regression (MAE/RMSE/MAPE/R²), ranking (NDCG/MAP). Pick metrics that match the business problem. - **Baseline comparison**: compare against a simple heuristic, dummy classifier, or previous model version. - **Confidence intervals**: report uncertainty around metrics, not just point estimates. - **Calibration**: for probabilistic models, check calibration curves. - **Per-slice evaluation**: evaluate on subgroups (by category, value range, data source) to find failure pockets. - **Failure mode documentation**: list known failure cases, edge behaviors, and conditions where performance degrades. ## Expected output An evaluation report with metrics, baseline comparison, and documented failure modes, logged to the experiment tracker.