Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Take a real dataset (e.g. a Kaggle CSV) and go from raw data to a first honest model in a notebook. Clean the data (justify drop-vs-impute for missing values), do EDA (distributions, correlations, a couple of charts), engineer a few features (encode categoricals, scale where the model needs it), then train a simple supervised model (logistic/linear regression or a tree). CRUCIALLY: split into train/test, fit only on train, and evaluate on the held-out test set with the RIGHT metric (do not report accuracy on imbalanced data — use precision/recall/F1 or RMSE/R2). Note any data-leakage risks you avoided.
Sensible cleaning with justified missing-data handling; informative EDA (distributions/correlations) that guides the modeling.
Train/test split with fit on train only; the RIGHT metric chosen (not accuracy on imbalanced data); no evaluation on training data.
Categoricals encoded; scaling applied where the model needs it (not for trees); a couple of useful derived features.
Identifies and avoids obvious leakage (e.g. fitting a scaler before the split, or a target-derived feature).
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# An EDA + first model on a real dataset ## Goal Take a real dataset (e.g. a Kaggle CSV) and go from raw data to a first honest model in a notebook. Clean the data (justify drop-vs-impute for missing values), do EDA (distributions, correlations, a couple of charts), engineer a few features (encode categoricals, scale where the model needs it), then train a simple supervised model (logistic/linear regression or a tree). CRUCIALLY: split into train/test, fit only on train, and evaluate on the held-out test set with the RIGHT metric (do not report accuracy on imbalanced data — use precision/recall/F1 or RMSE/R2). Note any data-leakage risks you avoided. ## Acceptance criteria - [ ] EDA & cleaning: Sensible cleaning with justified missing-data handling; informative EDA (distributions/correlations) that guides the modeling. - [ ] Honest evaluation: Train/test split with fit on train only; the RIGHT metric chosen (not accuracy on imbalanced data); no evaluation on training data. - [ ] Feature work: Categoricals encoded; scaling applied where the model needs it (not for trees); a couple of useful derived features. - [ ] Leakage awareness: Identifies and avoids obvious leakage (e.g. fitting a scaler before the split, or a target-derived feature). ## Verification - [ ] Document installation and run commands. - [ ] Record automated checks and their exact commands. - [ ] Add representative output, screenshots, or a short demo where useful. - [ ] Confirm failure paths and known constraints. ## Decision log ### Decision title - Context: - Choice: - Alternatives considered: - Tradeoffs: ## Evidence - Link each rubric criterion to the file, test, or artifact that demonstrates it. ## Known limitations - List what is intentionally out of scope and what should be improved next.