Project A — The Canteen Data Mini-Study

Individual or pairs (second semester appropriate) · Data you generate.

The brief

The canteen committee wants to know how students actually use it, but has no data. You will simulate three weeks of canteen sales with a documented generator, then analyse your own data as if it were real — including all the honesty problems that creates.

Part 1 — The generator (program)

Write generate.py producing sales.csv with columns day, item, price, count:

• 15 weekdays (three weeks), 4 items with fixed prices you choose;
• each day's counts random within item-specific ranges (e.g. sandwich 20–60, soup 10–40 …), with one documented weekday effect of your choice (e.g. Fridays sell 30% fewer hot meals);
• a fixed random seed so your data is exactly reproducible;
• deliberate imperfections, each occurring at low, stated rates: occasional negative counts, one duplicated day per week, and rare off-menu items.

Part 2 — Clean and profile (program + report)

Write clean.py that reports (not silently fixes) every defect it finds: rows dropped, rows fixed, duplicates merged — each with a count and a rule. Produce a one-page profile: total sales, best day, best item, and the distribution of daily totals.

Part 3 — Answer three questions (report)

1. Which item earns the most across the three weeks, and does the weekday effect survive the missing/duplicate handling?
2. Is the "best day" stable if you re-run the generator with a different seed? (Run 5 seeds; report the spread.)
3. What question about the canteen can this data not answer, and why?

Analysis section

• Efficiency: what grows with the number of rows; show the arithmetic.
• The named design decision the course did not make for you (e.g. your weekday-effect representation, your duplicate-merge rule).
• Limitations: what the simulation assumes that real canteens violate.

Data rules

• Everything is generated by you; the generator is part of the submission.
• Seed reproducibility is mandatory: graders re-run your generator.
• Every cleaning rule is stated in the report — silent deletion fails the integrity bar for this project.