v0.1.1 — A2 v7 parity lock

A2 v7 parity lock

• Locks the served classifier to the submitted MLN601 Assessment 2 v7 contract.
• Validates dataset hashes, row counts, feature order, target semantics, split, parameters, exact metrics, and confusion matrix before writing artifacts.
• Adds exact fresh-retrain parity tests for predictions, probabilities, and internal tree arrays.
• Exposes the model contract and source submission commit through /health and /model/info.
• Keeps all existing prediction request and response contracts unchanged.

Verification: 26 tests passed; make parity passed 18 checks; ruff passed.

v0.1.0 — The leakage audit

Honest-metrics release. Auditing the same pipeline for MLN601 Assessment 2 surfaced 1,177 exact duplicate rows in the raw UCI files crossing the train/test split and inflating every published v0.0.1 metric.

Changed

• Dedup before split in ml/train.py: 6,497 raw rows → 5,320 unique. Provenance (raw_rows, duplicates_removed) recorded in metrics.json and served at GET /model/info.
• Grade model is now the assessment-approved balanced tree: DecisionTreeClassifier(gini, max_depth=5, min_samples_leaf=20, class_weight="balanced"). It trades some false alarms for catching 73% of genuinely low wines (was 59% at default weighting) — in a screening problem, the miss is the expensive error.
• Honest re-trained metrics:

Metric	v0.0.1	v0.1.0
Regression R²	0.50	0.41
Regression RMSE	0.61	0.66
Classification ROC-AUC	0.81	0.79
Classification sensitivity (low)	—	0.73
Classification specificity (high)	—	0.73

The models did not get worse — the evaluation got corrected: v0.0.1 was graded partly on rows it had already seen.

• metrics.json and /model/info now include sensitivity, specificity, F1 and the full confusion matrix; the Streamlit model card and About page lead with them.
• Tests re-pinned to the deduplicated numbers, plus a gate test mirroring the assessment approval criteria (AUC ≥ 0.75, sensitivity ≥ 0.70, specificity ≥ 0.70). 24 tests, ruff clean.

Live

• 🍷 App: https://sommelier-api.streamlit.app/
• 📜 API (Swagger): https://sommelier-api-yd1m.onrender.com/docs

v0.0.1 — First public release

The first end-to-end cut of sommelier-api: two ML models trained on the UCI Wine Quality dataset, served by a FastAPI backend and a Streamlit UI over one shared core — deployed, tested, and documented.

Models

• Score (regression): RandomForestRegressor(n_estimators=400) → predicted quality, R² 0.50 / MAE 0.44 / RMSE 0.61.
• Grade (classification): tuned DecisionTreeClassifier(max_depth=6, min_samples_leaf=20) → high (≥6) / low (<6), accuracy 0.74 / ROC-AUC 0.81.
• Re-trained deterministically (random_state=42) from the public CSVs — bit-identical to source. Pinned scikit-learn 1.9.0.

What's in it

• ml/ core — features.py (single source of truth for the 12-feature contract), train.py (reproduces both models → joblib + schema + metrics), predict.py (shared inference + A2 label-inversion guard).
• FastAPI — GET /health /features /model/info; POST /predict/score /predict/grade /predict. Pydantic v2 validation, Swagger, real metrics from training. Deployed on Render.
• Streamlit — tasting-room with quality gauge + high/low grade badge; local-default inference with automatic API fallback. Deployed on Streamlit Community Cloud.
• Quality & ops — 23 tests, GitHub Actions CI (ruff + pytest), Makefile, Dockerfile, render.yaml, pinned requirements.

Live

• 🍷 App: https://sommelier-api.streamlit.app/
• 📜 API (Swagger): https://sommelier-api-yd1m.onrender.com/docs
• 📝 Write-up: https://dev.to/lfariaus/i-gave-the-same-6497-wines-to-two-models-and-asked-them-different-questions-4hdn