Summary - Clustering

In this group you learned clustering, KMeans, and the elbow method.

Summary: clustering

In this group:

Clustering groups without labels; the numbers are arbitrary.
KMeans: assign → move to the mean; assumes round, similar clusters; scaling is a must.
Centroids are the clusters' typical examples; transform gives distances to the centres.
Choosing k: elbow (inertia_) and the silhouette score.
ARI for comparing with a known structure; plain accuracy misleads.

In the next groups you'll see how KMeans is initialised and hierarchical clustering.

Exercise KMeans and DBSCAN

Run KMeans(n_clusters=2, n_init=10, random_state=0) and DBSCAN(eps=0.3, min_samples=5) on the moon data. Print both ARI values against the true labels with round(..., 3) on separate lines.

Show example solution

from sklearn.datasets import make_moons
from sklearn.cluster import KMeans, DBSCAN
from sklearn.metrics import adjusted_rand_score

X, y = make_moons(n_samples=300, noise=0.05, random_state=0)

km = KMeans(n_clusters=2, n_init=10, random_state=0).fit_predict(X)
db = DBSCAN(eps=0.3, min_samples=5).fit_predict(X)
print(round(adjusted_rand_score(y, km), 3))
print(round(adjusted_rand_score(y, db), 3))

Test yourself

Which is more suitable than KMeans for moon-shaped clusters?
- DBSCAN
- KMeans with a larger k
- PCA
- Logistic regression
Show answer

DBSCAN — Density-based DBSCAN finds clusters regardless of shape.

Set up density-based clustering.

db = ____(eps=0.3, min_samples=5)

Show answer

db = DBSCAN(eps=0.3, min_samples=5)

Match the tools.

KMeans

Centroid-based
DBSCAN

Density-based
silhouette_score

Choosing k
adjusted_rand_score

Comparison with labels

Next lesson: KMeans algorithm

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies