Skip to learning content

DATA SCIENCE PYTHON PLAYGROUND

Machine Learning · Learn / Refresh

← Choose a deckQUESTIONS · MODELS · EVIDENCE
DECK 5

Clustering and Discovery

Ask exploratory questions with K-Means and hierarchies.

01 / A different question

U01Core · teaching
Discovery has no prediction target: No target y, Find groups, Describe structureNo target yFind groupsDescribe structure

Discovery has no prediction target

State the exploratory population and question.

12–18 min · 3 practices
U02Core · teaching
Distance depends on scale and context: Original units, Scaled space, Distance changesRaw unitsScaled spaceDistances change

Distance depends on scale and context

Fit a meaningful distance representation for the exploratory population.

12–18 min · 3 practices

02 / K-Means

U03Core · teaching
K-Means assigns points to centroids: Assign point, Move centroid, RepeatAssign pointMove centroidRepeat

K-Means assigns points to centroids

Understand centroids, distances and assignments.

12–18 min · 3 practices
U04Core · teaching
Choosing k requires evidence and judgement: Inertia, Silhouette, Inspect profiles tooInertiaSilhouetteInspect profiles too

Choosing k requires evidence and judgement

Combine compactness, separation and useful interpretation.

18–25 min · 4 practices
U05Core · teaching
Describe clusters in meaningful units: Cluster label, Original units, Group profileCluster labelOriginal unitsGroup profilegroup 0mean lengthn rowsgroup 1mean lengthn rows

Describe clusters in meaningful units

Use sizes and original-unit profiles with named features.

12–18 min · 3 practices
U-R1Core · review
K-Means retrieval: Scale distance, Choose k, Profile groupsScale distanceChoose kProfile groups

K-Means retrieval

Retrieve earlier concepts before combining them.

15–20 min · 3 practices
U06Core · teaching
When K-Means geometry misleads: Elongated shape, Centroid limit, Check geometryElongated shapeCentroid limitCheck geometry

When K-Means geometry misleads

Recognise initialisation, scale and shape limitations.

12–18 min · 3 practices

03 / Hierarchical discovery

U07Core · teaching
Hierarchical merging: Small merges, Larger merges, DendrogramSmall mergesLarger mergesDendrogram

Hierarchical merging

Interpret Ward linkage and merge height.

25–40 min · 4 practices
U08Core · teaching
Turn a hierarchy into groups: Cut height, Group count, Describe groupsCut heightGroup countDescribe groups

Turn a hierarchy into groups

Choose a cut using evidence and purpose.

12–18 min · 3 practices
U09Core · teaching
Sampled hierarchies describe sampled rows: Sample rows, Keep row IDs, Interpret sampleSample nSample rowsKeep row IDsInterpret sample

Sampled hierarchies describe sampled rows

Keep population, sample and labels aligned.

12–18 min · 3 practices
U10Core · teaching
Interpret discovery without inventing truth: Fit without label, Profile groups, Avoid truth claimFit without labelProfile groupsAvoid truth claimNOW

Interpret discovery without inventing truth

Keep external labels and claims separate from fitting.

12–18 min · 3 practices
U-R2Core · review
Hierarchy retrieval: Linkage, Cut height, Sample scopeLinkageCut heightSample scope

Hierarchy retrieval

Retrieve earlier concepts before combining them.

15–20 min · 3 practices
U-K1Core · checkpoint
Discovery checkpoint: Prepare X, Find groups, Profile in unitsPrepare XFind groupsProfile inunits

Discovery checkpoint

Scale the transfer population, compare k evidence, profile groups and qualify their interpretation.

25–35 min · 1 practices