Skip to content

Grading

One hundred points. Sixty from three team projects, forty from an individual exam.

20 Project 1
20 Project 2
20 Capstone
40 Final exam
60 points · built in teams 40 points · on your own

The split is the point. Sixty points reward what you can build with other people, which is how the work actually happens once you leave. Forty reward what you understand alone, which no team can carry for you.

Each block ends with one. Each is worth twenty points, graded against a rubric published before the project opens.

Project 1 · Validated baseline. A messy dataset, simple models only, and a pipeline you can defend.
Project 2 · Model comparison. Several model families, an equal tuning budget, and a claim about which differences are real.
Project 3 · Capstone. Your own problem, end to end, presented live.

Full specifications are on the projects pages.

Criterion Points What earns them
Method 8 Correct splitting, no leakage, evaluation that fits the problem
Execution 5 The code runs, the pipeline is reproducible, results are what you claim
Communication 4 A README and report a stranger can follow, limitations named
Craft 3 Repo conventions, commit history, code someone else can read

Where points are actually lost

Almost never on modelling. They go on a leaked feature, a repo that will not clone and run, a conclusion the measured variance does not support, and a README written at midnight. Three of those four are settled before you write a line of model code.

Each criterion is split into bands on the rubrics page, so you can see what separates seven points from five before you start. How a team grade becomes your grade, and how work is handed in, are on the rubrics and submission pages.

Forty points, individual, closed book except for one page of notes you write yourself.

Part Points What it asks
A · Concepts 15 Short answers about why, not what. "Your CV score is 0.91 and your test score is 0.62. Give three plausible causes."
B · Code reading 15 You are given a pipeline. Find what is wrong with it and explain the consequence.
C · Judgement 10 A scenario. Choose a metric, a split, and a baseline, and defend all three.

Nothing on the exam asks you to recall a formula. It tests whether you can diagnose, choose, and justify, which is what the projects spend fourteen weeks training.

How to prepare

Reread your own project logs. Most exam questions resemble something that went wrong in one of your three projects.

Work that clearly goes beyond a brief can earn a small bonus: an extension done properly, something built that helps the whole class, a real error found in these materials, or help given to other teams that they credit. Bonuses are modest and at my discretion, they are never needed for full marks, and the total is still capped at 100.

Points Grade
95 to 100 10
85 to 94 9
75 to 84 8
65 to 74 7
55 to 64 6
45 to 54 5
Below 45 4 (fail)

Passing needs both halves

You need 45 points overall and at least 15 of the 40 exam points. A team cannot carry you to a pass. This is the one place where the individual component is a gate rather than a weight.


Team formation, peer review, and deadlines are set out in each project brief when the project opens. See also the rubrics and how to submit.

Next: Week 1