Example notebooks

Three end-to-end notebooks, fully executed with real outputs saved in place. They are the recommended reading order after Getting started.

1 — From a messy CSV to an actionable report

notebooks/01-messy-csv-to-decision.ipynb · English · data: UCI Online Retail (CC BY 4.0)

541,909 real e-commerce rows. A naive load silently destroys 9,291 invoices; editing the disclosed load plan rescues every one. Cancellations, duplicates, guest checkouts, honest seasonality, and a one-line edareport finale.

2 — Build a leakage-safe ML pipeline

notebooks/02-leakage-safe-pipeline.ipynb · English · data: IBM Telco churn + KDD Cup 09 (OpenML, Public)

Leakage measured instead of moralized. Preprocessing-before-split costs nothing on tame data (0.8465 vs 0.8465 -- said honestly); selection-before-split injects a bias a single check cannot even see; 20 splits and repeated CV expose it as systematic (+0.026, ahead in 9 of 10 repeats). Stars: featpipe / selectpipe FIT->APPLY artifacts and dextra.compat inside CV folds.

3 — قصة الأسعار: أسعار الغذاء في مصر (بالعربية)

notebooks/03-egypt-food-prices-ar.ipynb · Arabic narrative, English code · data: WFP Food Prices for Egypt via HDX (CC BY-IGO)

تحليل استكشافي كامل بالعربية على بيانات حقيقية من السوق المصري: حُبيبية مختلطة في عمود واحد، فقدان بنيوي لا يُعوَّض، انقطاع سلسلة عام 2022 يحسمه اختبار ت المزدوج لا المزاج، سعر الصرف الضمني من 5.7 إلى فوق 52 جنيهاً للدولار، والتضخم بالجنيه مقابل الدولار — مع قاموس مصطلحات ثنائي اللغة.

---

Each notebook installs its own extras in its first cell and downloads its data on first run; see notebooks/README.md for run instructions and notebooks/data/README.md for data licenses and attributions.