Practical Work 1: Explore the Titanic, and ask it five questions

Objectives

By the end of this session you will be able to:

Create a project in rakoon-ds and read the workshop as a graph, not as a screen
Read a dataset card before the data, and know what a trap is
Tell apart the three tabs of the explorer: Overview, Statistics, Quality
Read a quality report: missing values and their patterns, outliers, duplicates, identifier columns, class balance
Turn a vague curiosity into five questions a chart can answer
Pin a figure and pour it into the written report

Prerequisites

A rakoon-ds account on https://rakoon-ds.apps.way-up.io (free, browser only, nothing to install), and the group join code your instructor gives you
A recent Chrome, Edge or Firefox. Computation runs either in your own tab (browser engine) or on the server; the header pill tells you which
The mission for this session, assigned to your group. Open a project, then the Mission button in the workshop header: the panel opens next to Report. Click Check after each step
No Python, no install, no spreadsheet. Everything happens in the studio
Transformation, chart and algorithm names come from the rakoon-ds registry and are served in French even when the interface is in English. Every step below gives you the French label you will click and, in code font, the registry key the mission checks against.

Data

This session uses a built-in dataset: nothing to download.

Dataset	Where	Size	Target
`titanic` (Titanic passengers)	Studio, Add a dataset → Built-in dataset	1 309 rows, 11 columns	`survivant` (yes / no)

Source: the titanic3 table (Harrell and Cason, Vanderbilt University), public domain, shipped with rakoon-ds. Column names are in French: classe, nom, sexe, age, freres_conjoints, parents_enfants, prix_billet, cabine, port_embarquement, canot, survivant.

Timing

The steps below add up to the announced duration. If you fall behind, Step 1 to Step 3 are the ones that must be finished.

#	What you do	Time
Step 1	Create the project and load the data	8 min
Step 2	Read the card before the data	7 min
Step 3	Overview and Statistics	12 min
Step 4	The Quality tab, point by point	15 min
Step 5	Five questions, five charts	20 min
Step 6	Your first report blocks	13 min
Total		75 min

Instructions

Step 1: Create the project and load the data8 min

Log in to the studio, then:

On the projects page, New project. Name it PW1 Titanic.
In the workshop, click Add a dataset (top left of the graph).
In the Built-in dataset list, pick Titanic (passagers).
Before clicking Add, click the Fiche (card) button next to it. Read it. It costs one minute and saves twenty.

A round node named titanic appears in the graph. Everything you do from now on hangs off that node: a transformation, a chart or a model is a new node, and the original is never modified.

Open the Mission panel now (button in the workshop header) and click Check. Step 1 should turn green. Get used to checking after every step: a validated step stays validated.

Step 2: Read the card before the data7 min

The dataset card is not decoration. It carries five traps that this dataset was chosen for. Write the answers to these three questions in your notes, you will need them in Step 6:

What does one row represent, and what is the target?
Which column does the card call a deliberate leak, and why is a leak worse than a missing value?
What accuracy does a model that always answers "no" already reach? (the card gives the number)

The card says it plainly: canot (lifeboat) tells you almost by itself who survived, because 477 of the 486 passengers whose lifeboat is known did survive. A column that is only filled in after the event you are predicting is not a feature. It is the answer.

Step 3: Overview and Statistics12 min

Click the titanic node, then Explore in the side panel. The explorer has three tabs.

Overview: the first rows. Scroll right to the end. Which columns are clearly text, which are numbers, which are numbers that are really categories?
Statistics: type, missing count, distinct count, min / mean / max per column, and the correlation matrix. Find the three columns with the most missing values.
Still in Statistics, click Pin: the profile becomes a node in the graph and can be poured into the report later.

Two readings to make now, out loud, with your neighbour:

classe holds 1, 2 and 3. It is a number in the file and a category in real life. What would a linear model do with it?
nom has 1 307 distinct values out of 1 309. What can a model possibly learn from a column that is nearly a row number?

Step 4: The Quality tab, point by point15 min

Third tab: Quality. It is the tab that reads the dataset against you. At the top, Points to watch, worst first. Below, four cards. Fill this table in your notes from what you see:

What the tab says	Column(s)	Number	What you decide
Identifier-like column
Missing values, worst column
Second and third worst
Most frequent missing pattern
Duplicate rows
Class balance (majority / minority)

For reference, here is what the studio reports on this dataset (measured on the instance, 2026-09-09), so you can check your reading:

nom flagged as an identifier: 1 307 distinct values, ratio 0.998
age missing 263 times (20.1 %), cabine 1 014 times (77.5 %), canot 823 times (62.9 %)
Most frequent pattern: cabine and canot empty together, 538 rows
Duplicate rows: 0
Balance: 809 "non" (61.8 %) against 500 "oui" (38.2 %), ratio 1.62

The patterns block is the one nobody reads and the one that talks. Columns that go empty together share a cause. Here, no known cabin and no known lifeboat is, most of the time, the same passenger: third class.

Step 5: Five questions, five charts20 min

Back to Explore, tab Overview, chart selector on the right. The line Most informative: suggests charts ranked by how much they tell you; you are free to ignore it.

Produce five charts, one per question. Suggested set (the French label is what you click, the key in brackets is what the mission checks):

Question	Chart	How to set it up
How is age distributed, and where are the holes?	Histogramme (`histogram`)	Variable: `age`
Does survival depend on sex?	Comptage / moyenne (barres) (`bar`)	Colonne (axe X): `sexe`, Séparer par: `survivant`
Does it depend on class?	Comptage / moyenne (barres) (`bar`)	Colonne (axe X): `classe`, Séparer par: `survivant`
Which variables move together?	Corrélations (heatmap) (`correlation`)	All numeric columns
Did survivors pay more?	Distributions (violin) (`violin`)	Valeur: `prix_billet`, Grouper par: `survivant`

For each chart: click Pin, then write one sentence of what it shows. A chart without a sentence is not a result, it is wallpaper.

The correlation heatmap mixes three measures: Pearson between two numeric columns, Cramer's V between two categorical ones, and the correlation ratio between a numeric and a categorical one. The legend under the chart says which is which. Do not read a Cramer's V as if it were a Pearson.

Step 6: Your first report blocks13 min

Open the Report panel (workshop header). Seven sections, each with its guiding questions in grey.

Section Question et contexte: write who would use an answer to "who survives" and how you would know the answer is any good. Aim for 5 to 8 lines.
Section Données: what a row is, where the data comes from, what the target is, and the three numbers you wrote down in Step 4 (missing values, balance, the leak). Aim for 8 to 12 lines.
Go back to two of your pinned charts and click Add to report. The chevron next to the button lets you choose the section: put them under Exploration.
Under each block, write the caption. The field saves itself.

Click Check in the mission panel. Five steps out of five should be green. Then click Lecture (Read) at the top of the report panel: your report opens as a document in a new tab. That document is the deliverable of every session of this course.

What you should have

A project PW1 Titanic whose graph holds the dataset node, a pinned profile and five pinned charts
A filled Quality reading: one identifier column, three columns with holes, one missing pattern, zero duplicates, a 62 / 38 class balance
Five charts, each with one written sentence
Two report sections written, two figure blocks poured in
The dep-1 mission at 5 / 5

Deliverables

Mission: dep-1 validated (the instructor sees it in the mission progress table, no file to send)
Report: sections Question et contexte and Données written, two figures with captions. Export it with Exporter le rapport → Markdown (.md) and keep the file
One sentence answering: what would you refuse to put into a model that predicts survival, and why?

Bonus

Train a Foret aleatoire (rf_clf) on survivant with all columns, then a second one without canot. On the reference instance the accuracy goes from 0.966 to 0.820. Explain the 15 points to somebody who has never heard of machine learning.
In the Quality tab, look at the outliers card for prix_billet. 171 fares are outside the Tukey bounds. Are they errors?
Rename the pinned profile node and reorganise the graph (Reorganize button): with ten nodes it is already worth doing.

Resources

Session 1 slides (the lecture this practical work follows)
Course page: both programmes, all fifteen sessions
rakoon-ds studio
Dataset card: Add a dataset → Titanic (passagers) → Fiche
The mission JSON, if you are curious about how it is checked: missions/dep-1.json