Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Build a more ambitious project in an applied domain. Choose EITHER (a) a deep-learning task — fine-tune a PRETRAINED model (transfer learning) for image classification or text/NLP with PyTorch/TensorFlow/Hugging Face; OR (b) a recommender or a time-series forecaster (split BY TIME so no future leaks into training). For either: use proper train/validation/test (time-aware if time series), establish a strong baseline to beat, use metrics appropriate to the task, and control overfitting (regularization / early stopping / augmentation). Add a RESPONSIBLE-ML pass: either evaluate performance across subgroups (fairness) OR use SHAP/LIME to explain predictions, and note any bias or spurious-correlation risk. Document the modeling choices and what productionizing it would require (MLOps: serving + drift monitoring).
A transfer-learning deep model OR a recommender/time-series forecaster, with correct (time-aware if needed) splits, a baseline, and overfitting control.
Task-appropriate metrics on held-out data; the baseline beaten; no leakage (especially chronological for time series).
A fairness (per-group) or explainability (SHAP/LIME) pass + a bias/spurious-correlation note; a productionization note (serving + drift monitoring).
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# An applied ML project with responsible-ML checks ## Goal Build a more ambitious project in an applied domain. Choose EITHER (a) a deep-learning task — fine-tune a PRETRAINED model (transfer learning) for image classification or text/NLP with PyTorch/TensorFlow/Hugging Face; OR (b) a recommender or a time-series forecaster (split BY TIME so no future leaks into training). For either: use proper train/validation/test (time-aware if time series), establish a strong baseline to beat, use metrics appropriate to the task, and control overfitting (regularization / early stopping / augmentation). Add a RESPONSIBLE-ML pass: either evaluate performance across subgroups (fairness) OR use SHAP/LIME to explain predictions, and note any bias or spurious-correlation risk. Document the modeling choices and what productionizing it would require (MLOps: serving + drift monitoring). ## Acceptance criteria - [ ] Applied modeling: A transfer-learning deep model OR a recommender/time-series forecaster, with correct (time-aware if needed) splits, a baseline, and overfitting control. - [ ] Evaluation rigor: Task-appropriate metrics on held-out data; the baseline beaten; no leakage (especially chronological for time series). - [ ] Responsible ML & ops: A fairness (per-group) or explainability (SHAP/LIME) pass + a bias/spurious-correlation note; a productionization note (serving + drift monitoring). ## 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.