Getting started
- The shape
- One question, one data set, one or more methods from the course "Does X predict, explain or separate Y in data Z?" If you can fill in all three today, you have a project. Regression and classification are taught by mid-October and cover most questions; the simplest method that fits is the right first model.
- Data first
- Find it before you fall in love with the question Most proposals that fail, fail because the data was never in hand. Download it, keep a copy, check the licence. Messy data is good — the cleaning is graded work. Where to look: our advice, Part I.
- How big
- One method taken all the way, or several methods, each with a job Taking a method all the way means more than fitting it and reporting a score: its assumptions checked against your data, its tuning explained, its mistakes examined. Several methods means each one is there because the one before it showed you something. Either way, we look for decisions that are yours and that you can explain; the grade follows what the project shows, not the hours behind it. Everything must run in the Nuvolos workspace.
- Propose early
- 9 November is the last day, not the day to aim for Two hundred words: the question, the data (already downloaded), the method. Approval comes by return email as proposals arrive, so a proposal sent in mid-October buys three extra weeks. A template, a worked example and a week-by-week timeline are in our advice, Part I.
What it is
- The work
- One project, carried out individually You choose a question, find data, apply methods from the course, and write it up. It is the only assessment in this course and carries 100 % of the grade.
- What you hand in
- One GitHub repository It holds the paper, the data, the code, and a video of about 15 minutes in which you present the project — there is no oral presentation. The paper is about ten pages in the SIAM conference style, with eight required sections. Submission is the repository — not an email attachment and not Moodle.
- How you submit
- One email with the repository link and the commit identifier The commit you name is the one that is graded, so nothing you push afterwards changes it — and nothing is lost if you keep working. Our advice shows the one command that prints it.
- Your GitHub account
- Must show your full name Set your real name in your GitHub profile before you submit. A submission from an account that cannot be matched to a registered student is not graded, and we will not guess whose it is.
- Two dates
- A 200-word proposal, then the project Email the proposal to the teaching assistant; the reply approving it is the sign-off, and work starts after that. Proposal by Monday 9 November 2026, 23:59; complete project by Sunday 20 December 2026, 23:59 (Zurich time). Late work is not accepted.
- AI tools
- Allowed, and expected — but declared Use assistants for code and prose. List every tool you used, and what you used it for, in the appendix. Undeclared use is misconduct; declared use is normal practice.
- The paper template
- SIAM article style, on Overleaf Click "Open as Template" and it creates a project in your account — nothing to download, no LaTeX to install, compiles in the browser. Open it in the first week and write into it from the start.
- Compute
- Your Nuvolos workspace — and that is what is provided It is enough for the methods this course teaches. Scope the project to fit it, at proposal stage. You may use your own compute at your own cost, but it earns no marks: the work is assessed, not the hardware.
- How the repository looks
- Split by stage — not one enormous notebook One notebook per stage: get the data, clean it, model it, make the figures. Each runs top to bottom from a fresh kernel. A single file holding the whole analysis cannot be checked, and cannot show which parts are yours.
- Code you did not write
- Attribute it, and it costs you nothing Libraries need no attribution. Anything else you did not write — a function from a forum, a chunk of a repository, code from an assistant — is named where it appears and listed in the appendix. Declared reuse is assessed on what you built with it; undeclared reuse is misconduct, whatever its size.
- If the result is negative
- An honest null result is not held against you A hypothesis that data does not support is a real finding, and a well-told account of why it failed is worth more than a result that was helped along.
The two documents
- Requirements
- Requirements for the semester project (PDF) The binding text: deadlines, what is submitted, how it is assessed. Where this page and the requirements differ, the requirements are what counts.
- Our advice
- Starting and shipping your project: our advice (PDF) Practical advice, and not binding. Part I: choosing a question, finding data, sizing the project, the proposal, a week-by-week timeline. Part II: folder layout, the README, reproducibility, the video, and the one test to run before you submit.
Gallery
Selected projects are published here, with their authors' permission and credited to them. Consent is asked for at submission, is entirely optional, and has no bearing on the grade.
The gallery opens with the first consenting projects from this cohort. Until then, the requirements above describe the standard the work is held to.