Project Workshop

<p class="kicker">MST0052 · Lecture 15 · Fall 2026</p>

## Project Workshop<br>The Final Rehearsal

### Vegard H. Larsen — BI Norwegian Business School

---

## Where we are

L14 told you: **run the test once** and bring the project. Today is the **rehearsal** — submission is **Friday**.

![Semester timeline with four phases](../figures/semester-timeline.svg)

---

The toolbox is closed. Today you find out whether you can **defend** what you built.

A working session — present, listen, fix. No new methods; the method is each other.

---

## Today's plan

<p class="kicker">90 minutes · working session · bring your project</p>

- **How today runs** — groups, timing, roles
- **The pitch and the audit** — five parts, examiner-grade listening
- **~60 minutes of workshop**
- **Before Friday** — checklist, the last three days, a first look at December

---

## The clock

<div class="cols-3">
<div class="stat stat--blue">
<span class="stat-value">90</span>
<span class="stat-label">minutes of structured feedback —<br>the last before submission</span>
</div>
<div class="stat stat--red">
<span class="stat-value">3</span>
<span class="stat-label">days to the deadline —<br>Fri 20 Nov</span>
</div>
<div class="stat stat--green">
<span class="stat-value">Dec</span>
<span class="stat-label">the oral exam —<br>the real version of today</span>
</div>
</div>

---

## How today runs

---

## The format

<p class="kicker">Default · adjust to the room</p>

<div class="cols-3">
<div class="stat">
<span class="stat-value">4</span>
<span class="stat-label">people per group</span>
</div>
<div class="stat stat--blue">
<span class="stat-value">15 min</span>
<span class="stat-label">per person —<br>strictly timed</span>
</div>
<div class="stat stat--word stat--green">
<span class="stat-value">5+8+2</span>
<span class="stat-label">pitch + feedback<br>+ fix list</span>
</div>
</div>

---

## Roles

<div class="cols-3">
<div class="callout">
<h3>Presenter</h3>
<p>Drive from your notebook and results table. No slides needed.</p>
</div>
<div class="callout">
<h3>Your group</h3>
<p>Examiners for a day — at least one specific question each.</p>
</div>
<div class="callout">
<h3>Me</h3>
<p>Rotating between groups, ~10 minutes each. I flag methodology, not style.</p>
</div>
</div>

---

## The rehearsal

---

## The five-minute pitch

<p class="kicker">5 minutes · no slides needed</p>

1. **Problem** — target, metric, why it matters
2. **Data** — source, rows, features
3. **Pipeline** — split → pipeline → CV — the **L7 protocol**
4. **Results** — CV **mean ± std** and the test number, **touched once**
5. **Limitations** — what didn't work, what you'd try next

<p class="muted">End with the <strong>one question</strong> you want the group's opinion on.</p>

---

If a claim doesn't survive your **group** today, it won't survive the **examiner** in December.

Rehearsal, not defence — today the stakes are zero and the feedback is free.

---

## Listening like an examiner

<ul class="checklist">
<li>Is the <strong>target and metric</strong> clearly defined?</li>
<li>Is preprocessing <strong>inside the pipeline</strong>?</li>
<li>Is there a <strong>real baseline</strong> — and is it beaten?</li>
<li>Are the numbers read <strong>honestly</strong> — variability, not just the headline?</li>
<li class="no">"Looks good" — helps nobody</li>
</ul>

---

## The audit

<p class="kicker">Reference — screenshot this</p>

| Pitfall | The check | Lecture |
|---------|-----------|---------|
| **Leakage** | preprocessing fitted before the split — is it in the pipeline? | L3 |
| **Wrong metric** | accuracy on imbalanced classes — F1 / AUC instead | L5 |
| **No baseline** | majority class or mean predictor first — beat the bar | L5 |
| **Test set reused** | selection by CV + a written rule; test touched **once** | L7 |
| **Importance over-read** | impurity bias — permutation importance for the report | L10 |
| **No sensitivity check** | reseed, resplit — does the finding survive? | L13 |

---

## While you work

<p class="kicker">15 min per person · 5 pitch + 8 feedback + 2 fix list</p>

**Present**   `Problem` `Data` `Pipeline` `Results` `Limitations`

**Listen for**   `leakage` `metric` `baseline` `test-once` `sensitivity`

</div>

<p class="muted">Write the fix list — verbal promises don't survive the weekend.</p>

---

## Before Friday

---

## The submission checklist

<ul class="checklist">
<li>One <strong>PDF report</strong> — problem → data → methodology → results → reflection</li>
<li><strong>AI-use statement</strong> — tools, what for, how you verified (L2)</li>
<li><strong>CV mean ± std and the test score</strong> — reported once</li>
<li class="no">Padding for length</li>
</ul>

<p class="muted">Code is <strong>not</strong> submitted — sharing it on GitHub is encouraged, but ungraded. Quality beats length; full detail in the assessment guidelines.</p>

---

## The last three days

<p class="kicker">Deadline · Friday 20 November 2026 · 12:00</p>

<div class="cols-3">
<div class="callout">
<h3>Today + Wednesday</h3>
<p>Fix the audit findings — biggest first.</p>
</div>
<div class="callout">
<h3>Thursday</h3>
<p>Fresh folder, run from scratch — numbers match the report.</p>
</div>
<div class="callout">
<h3>Friday</h3>
<p>Proofread, AI statement, submit early — not at 23:59.</p>
</div>
</div>

---

## Wrap-up

---

## December — the real version of today

- "Why **this model** over that one?"
- "What does the number in **your table** mean — how was it computed?"
- "What's the **biggest weakness**?"

The questions that made you **pause** today are your prep list.

<p class="muted">L16 — tomorrow — covers the oral exam in depth.</p>

---

You have heard the questions. Answer them **on paper by Friday** — and **out loud in December**.

The rehearsal ends today. Ship it.

---

## Three words for this lecture

<div class="cols-3">
<div class="stat stat--word stat--blue">
<span class="stat-value">Rehearse</span>
<span class="stat-label">say it aloud today — December<br>is too late to find the gaps</span>
</div>
<div class="stat stat--word stat--green">
<span class="stat-value">Listen</span>
<span class="stat-label">specific beats kind —<br>"looks good" helps nobody</span>
</div>
<div class="stat stat--word">
<span class="stat-value">Fix</span>
<span class="stat-label">three days — methodology<br>first, prose last</span>
</div>
</div>

---

## Questions

Backup slides below — press down ↓

## Grade descriptors in full

<p class="kicker">Backup · grading</p>

| Grade | What it looks like |
|-------|--------------------|
| **A** | Ambitious, rigorous, well defended. Independent thinking. |
| **B** | Solid work with at least one aspect beyond the basics. |
| **C** | Standard methods, correctly applied. Competent but not deep. |
| **D** | Works end-to-end but noticeable gaps in reasoning or validation. |
| **E** | Minimal viable submission. Significant methodology flaws. |
| **F** | Cannot defend the work or missing major components. |

## Reproducibility, concretely

<p class="kicker">Backup · Thursday's check</p>

- **Pinned dependencies** — `requirements.txt` or `environment.yml`
- **Random seeds** — set them, document them
- **A README** — how to run the pipeline from scratch
- **A pipeline** — not a notebook of scattered cells

</div>

## If your group finishes early

<p class="kicker">Backup · spare time</p>

Run a **mock-examiner round** — each member asks one question:

- Why **this dataset**?
- Why **X over Y**?
- How did you **validate**?
- **Biggest weakness**?
- What would you do **next**?

---

## What's next

**Lecture 16:** Summing up and Q&A — **tomorrow, Wed 18 Nov**

- The course as one arc
- The oral exam in depth
- Final submission reminders